--- title: "Hermes Plugin Hackathon" url: "https://aaddrick.com/hermespluginhackathon" --- # Hermes Plugin Hackathon ## Hermes Plugin Hackathon *Every plugin entered in the Hermes Agent Plugin Hackathon, in one place. Search them, filter by what they do, and see the screenshots and demos.* - **1508** plugins - **1187** makers - **743** merged Pull requests to NousResearch/hermes-agent opened during the hackathon that add a catalog plugin or call themselves entries, from a snapshot taken Nov 11, 2026. The web page has screenshots, filters and search; the raw data is at https://aaddrick.com/hermespluginhackathon.json. ### Most plugins entered | Maker | Plugins | | --- | ---: | | salch-cred | 13 | | mojomast | 11 | | arjunbagga-xyz | 8 | | josselinlbe | 8 | | tera-sarah | 8 | | flabanabba8 | 7 | | percyhendrux123-prog | 7 | | tuancookiez-hub | 7 | | Whydiriki700 | 7 | | 100yenadmin | 6 | ### Hooks they use most | Hook | Plugins | | --- | ---: | | pre\_tool\_call | 388 | | pre\_llm\_call | 341 | | post\_tool\_call | 248 | | on\_session\_end | 114 | | transform\_llm\_output | 96 | | post\_llm\_call | 92 | | on\_session\_start | 90 | | transform\_tool\_result | 75 | 99 entries came in during the last hour before the deadline. 167 people entered more than one plugin. ### Every entry, newest first ### [offense](https://github.com/NousResearch/hermes-agent/pull/154115) Tools · by offense-research · open · 11-11 Spin up isolated Offense security sandboxes and summon swarms of independent investigators against a verified staging target: register and verify control of the target, launch a focused sandbox or a broad swarm with per-agent budgets, read deduplicated findings and checksummed evidence, verify a candidate in a fresh sandbox, and retest after a fix. ### [Deadline Guard](https://github.com/NousResearch/hermes-agent/pull/154112) Automation · by Snedz · open · 11-11 Deadline governance for Hermes. Reads a local Markdown deadlines table and exposes check\_deadlines, next\_founder\_steps, propose\_status\_update, and deadline\_guard\_cron: commitments are bucketed overdue / 24h / 72h / 7d, founder-gated next steps are surfaced before they go critical, and a status can never be recorded as 'submitted' without an http(s) receipt URL - the proposal comes back unapplied for a human to confirm. Ships a daily-check automation blueprint (08:00 cron) and a bundled deadline-guard skill. Validated on Hermes 0.23.0: hermes plugins validate --install-deps exit 0, all 15 checks pass; 47 tests pass. ### [portal-engine](https://github.com/NousResearch/hermes-agent/pull/154111) Tools · by bazziba-it · open · 11-11 Interactive "portal" engine for Hermes: OpenCV + MediaPipe real-time hand tracking (up to 4 hands) with glowing-hand, portal-ring, pass-through and particle effects, camera-to-screen calibration (auto / manual / checkerboard), and a video output bridge (Syphon on macOS, Spout on Windows, NDI cross-platform, virtual camera on Linux, shared-memory fallback) so MadMapper, Resolume, TouchDesigner, vMix or OBS can consume the "PortalEngine" stream. 8 agent tools + on\_session\_end cleanup hook + bundled skill. ### [injection-range — prompt-injection fire drill](https://github.com/NousResearch/hermes-agent/pull/154110) Tools · by browncyber · open · 11-11 A prompt-injection fire drill for your own agent. /drill run plants eight documents, one at a time, each carrying a test injection (plain override, hidden HTML comment, forged system message, memory poisoning, base64, reply hijack, and two delegated-task cases where you tell the agent to follow a document), gives the agent an ordinary task with each, blocks any tool call that carries a drill canary before it runs (pre\_tool\_call), and scores each scenario as held, neutralized by a guard, or compromised from the tool calls and model requests it observed, never from the model's own account. Use it to test a model, or a guard plugin, before you trust the agent with real documents. ### [Barrier Shield](https://github.com/NousResearch/hermes-agent/pull/154107) Tools · by arjunbagga-xyz · open · 11-11 Control-Barrier Function (h(x) \>= 0) and Lyapunov continuous safety shield that solves a real-time Quadratic Program (QP) to smoothly dampen aggressive tool actions. ### [timecapsule](https://github.com/NousResearch/hermes-agent/pull/154106) Tools · by brayedan · open · 11-11 Sealed messages for later — seal a note so it cannot be read until its release condition is met (a date, N days of silence, or the sealer's explicit release), with a grace ladder and slow, announced deletion. Local-only, stdlib-only, no network. ### [Paranoy](https://github.com/NousResearch/hermes-agent/pull/154105) Tools · by DeleganceAI · open · 11-11 Read and write a Paranoy evidence ledger from Hermes: search, file and link claims, evidence and contradictions in shared collections, with roles (Scout, Claim Miner, Linker, Gap Finder, Fool, Ponder, Ledger) and optional capture of chat messages as evidence. The agent's Nostr key stays in Hermes as a declared secret; writes are signed in-process and the model never holds the key. ### [Hermes WebMCP Bridge](https://github.com/NousResearch/hermes-agent/pull/154104) Web · by NifasathA · open · 11-11 Permission-aware, ephemeral WebMCP capabilities with origin allowlists, schema validation, conservative risk classification, exact one-call approvals, and safe revocation. ### [Plastis3D](https://github.com/NousResearch/hermes-agent/pull/154103) Desktop · by midnightbobarun · open · 11-11 A local 3D scene editor for Hermes Desktop, with agent tools to build and edit scenes. ### [specsentry](https://github.com/NousResearch/hermes-agent/pull/154101) Tools · by birdcoin0 · open · 11-11 Evidence-backed API contract drift analysis using OpenAPI and HAR traffic, with deterministic verdicts, coverage intelligence, and reproducible regression tests. ### [hermes-skill-router](https://github.com/NousResearch/hermes-agent/pull/154098) Tools · by mortang2410 · open · 11-11 Removes the always-on skill catalog from every request and injects a short fused shortlist instead. Learns which skills you loaded for similar prompts. ### [Lyrics](https://github.com/NousResearch/hermes-agent/pull/154097) Desktop · by amitamit10 · open · 11-11 Time-synced lyrics for the song playing on Spotify: a Hermes Desktop page that highlights and centres the current line as the song plays, click any line to seek, with a per-user sync offset. Lyrics from the public LRCLIB API; playback through the official spotify plugin. ### [Sidecar for PowerPoint](https://github.com/NousResearch/hermes-agent/pull/154095) Tools · by AtlasOmnia · open · 11-11 Sidecar by Atlas Omnia — Your Hermes agent, in the sidebar of Microsoft Excel, Word, and PowerPoint. Sidecar puts your Hermes agent in a side panel next to the file you're working on. Ask it to draft, edit, analyze, or build without leaving your document, spreadsheet, or deck. In PowerPoint it reads slides, adds slides, text boxes and shapes, replaces and formats text, and reorders, deletes or re-lays out slides. After one manual local-CA trust step, connect with one click; approval authorizes subsequent chat-driven edits, not per-edit confirmation; runs on your machine, no API key files (protected TLS keys in plugin data). Your agent: it runs on your own Hermes setup, with the models and tools you already use. Setup: install the plugin, enable the Hermes API server, manually trust the per-install local CA using the printed command (never run automatically), load the sidebar manifest into PowerPoint, click Connect to Hermes and allow it in Hermes with /sidecar-powerpoint. ### [Sidecar for Word](https://github.com/NousResearch/hermes-agent/pull/154094) Tools · by AtlasOmnia · open · 11-11 Sidecar by Atlas Omnia — Your Hermes agent, in the sidebar of Microsoft Excel, Word, and PowerPoint. Sidecar puts your Hermes agent in a side panel next to the file you're working on. Ask it to draft, edit, analyze, or build without leaving your document, spreadsheet, or deck. In desktop Word for Mac it reads the document and selection, inserts and replaces text, applies exact-match edit plans, formatting, styles and comments. After one manual local-CA trust step and pane approval, the agent may read and write the bound document; this is not per-edit approval. No API keys stored by the plugin; TLS private keys are protected in plugin data. Your agent: it runs on your own Hermes setup, with the models and tools you already use. Setup: install the plugin, enable the Hermes API server, manually trust the per-install local CA using the printed user-scope command, load the sidebar manifest into Word, click Connect to Hermes and allow it in Hermes with /sidecar-word. ### [Sidecar for Excel](https://github.com/NousResearch/hermes-agent/pull/154093) Tools · by AtlasOmnia · open · 11-11 Sidecar by Atlas Omnia — your Hermes agent in desktop Excel for Mac. The plugin runs an HTTPS localhost bridge on ports 47640-47644, exposing 28 sidecar\_excel\_\* tools and opt-in workbook plans. Manual CA trust and one sidebar Allow are required. Allowing lets the configured site use your Hermes agent and all its tools until revoke; use a site you control or trust and open only documents you trust. API\_SERVER\_KEY is read through Hermes, never sent to the sidebar or stored by the plugin; TLS private keys are owner-only. For TLS it can only vouch for localhost. The trust command does not limit the certificate to TLS, so protect the key file and remove the trust when you stop using Sidecar. Latest live run: 2026-11-11 about 00:26 in desktop Excel for Mac, plugin and sidebar built from a7b3fd7. Connect, Allow, a 3-row table with header in A1:B4 and =SUM(B2:B4) in B5 read back from the workbook; revoke dropped the sidebar to needs your OK. The bridge answered 200 to the configured origin and 403 to another origin. The sidebar used a private test site via addin\_base, with CA trust in the login keychain; this Mac also has it in the System keychain. Earlier the same night at 35ef09e (behaviour unchanged since): reversal on request by clearing the range (no real undo), plugin reload with the sidebar open (it reconnected by itself and edited again), approval remembered after restarting Excel, next Connect after revoke asked for approval again. Not tested: Windows desktop Excel, dark theme, keyboard-only use, public Pages origin, TLS-only trust or all workbook tools/plans. Reopen the sidebar if it does not reconnect. The Excel test Mac also has System-keychain trust; that was not required in our Word and PowerPoint runs, which used login-keychain trust only. The default sidebar site is https://atlasomnia.github.io/sidecar-excel/ (publication pending); it is outside the catalog pin, and its origin is shared with other account Pages sites. Office.js loads from Microsoft. Hermes may send chat/workbook data to the configured provider. CLI approval/status never starts a listener. Remove matching CA trust manually before deleting the plugin tls directory; uninstall/revoke does not remove OS trust. No telemetry or self-updater. Apache-2.0; LICENSE and NOTICE included. No Microsoft or Nous Research endorsement. ### [Workbench by Atlas Omnia](https://github.com/NousResearch/hermes-agent/pull/154092) Tools · by AtlasOmnia · open · 11-11 Workbench by Atlas Omnia — Open and edit Word, Excel, PowerPoint, OneNote and Outlook for the web in Hermes Desktop. Live-tested in the automated harness and real Hermes Desktop sessions on macOS and Windows; see docs/live-evidence.md. Every edit requests Hermes approval and returns a readback/undo receipt; undo has exceptions (large Word snapshots, Excel over 10,000 cells, some PowerPoint edits and OneNote page creation), and history is lost on pane reload. powerpoint\_add\_speaker\_notes returns Unsupported. Requires Microsoft sign-in and Upload My Add-in. A loopback HTTP bridge on 127.0.0.1:47615-47619 accepts the configured add-in origin and a user-typed single-use code, then a per-pane token; idle listeners release after ten minutes. Default add-in site: https://atlasomnia.github.io/workbench/office/live/ (GitHub Pages, outside this catalog pin; rebuild with scripts/build\_addin.py and compare SHA256SUMS). Live runs used a private site; the public origin is first exercised at publication. The pane loads Microsoft Office.js. Document data enters the local Hermes conversation; OneNote pages and email indexed on request persist unencrypted in brain.db. Reads are capped at 20,000 characters for Word and Outlook bodies. No Graph, credentials, telemetry, shell commands or unattended runs. Outlook reads only the open item; mail and calendar drafts are never sent or saved and recipients/attendees never changed. Image URL fetches alone use outbound HTTPS (public addresses, checked redirects, PNG/JPEG, 4 MiB); local images are restricted to allowed folders. Full disclosures and known limits: README and docs/reference.md. Not affiliated with Microsoft or Nous Research. ### [Hermes AI Lab](https://github.com/NousResearch/hermes-agent/pull/154091) Models · by BZK-kais · open · 11-11 Prepare experiments, compare local models and activate qualified configurations in Hermes. ### [Butler](https://github.com/NousResearch/hermes-agent/pull/154090) Automation · by haykguy · open · 11-11 Turns Hermes into a proactive personal assistant. You set your daily non-negotiables and Butler checks in until they are done or deferred; you tell it about a commitment and it follows up before it slips. You choose its tone (Gentle to Strict), how persistent it is, quiet hours, time zone and a daily cap on check-ins, and can pause it in one click. A concierge-style Hermes Desktop page shows today's non-negotiables, upcoming commitments, recent activity and preferences, with a 4-step Quick Setup; the same controls are available to the agent as tools in chat. ### [Crucible](https://github.com/NousResearch/hermes-agent/pull/154089) Tools · by samshapley · open · 11-11 Adversarial research and code audits. Proposer agents commit quoted claims, attackers try to break them, a blind referee checks every quote against its source and re-runs every proof of concept in a sandbox, and you get a map of what survived and what was held back. ### [whoami](https://github.com/NousResearch/hermes-agent/pull/154088) Desktop · by atostivint · open · 11-11 Coloured pill in the composer showing which Hermes profile a new session will use. ### [SMB Growth Agent](https://github.com/NousResearch/hermes-agent/pull/154087) Tools · by jcolano · closed · 11-11 Local-first growth diagnosis, explainable opportunity ranking, 30-day plans and approval-gated simulated actions for small businesses. Includes two fictional demo businesses and nine tools. No outbound network calls, external execution, credentials, telemetry, or payments. Uses locally stored profile and action state only. ### [Hermes Presenter](https://github.com/NousResearch/hermes-agent/pull/154086) Voice · by violetSoftware · open · 11-11 Present a selected JSON slide deck with the calling local Hermes conversation as co-presenter. ### [Finix](https://github.com/NousResearch/hermes-agent/pull/154085) Automation · by web3blind · open · 11-11 Encrypted profile backups with cloud readback verification and experimental default-profile recovery. ### [hermes-creator-forge](https://github.com/NousResearch/hermes-agent/pull/154084) General · by zemhoutepro-beep · open · 11-11 Four Hermes skills for authoring and reviewing skills, workspace operations, and host-provided Python execution. ### [ELM for Hermes](https://github.com/NousResearch/hermes-agent/pull/154082) Memory · by Miawd3 · open · 11-11 Local source-backed memory with bounded retrieval and explicit human control. ### [CoinFello](https://github.com/NousResearch/hermes-agent/pull/154081) Tools · by CoinFello · open · 11-11 CoinFello for Hermes: the agent reads your CoinFello automation wallet (balances, positions, LP ranges, open orders, activity, prices, yields, swap and bridge quotes), prepares sends and swaps you approve in Hermes, and talks to Fello, CoinFello's DeFi agent. A desktop pane shows the wallet and portfolio with send, swap, earn and a daily trading allowance, and draws CoinFello's MCP App cards. Sign-in is the plugin's oauth: entry (hermes auth add coinfello:coinfello, OAuth 2.1 + PKCE against app.coinfello.com); Hermes stores and refreshes the grant and the plugin never holds a token. ### [relay-telegram](https://github.com/NousResearch/hermes-agent/pull/154080) Platform · by 0xCheetah1 · open · 11-11 RELAY incoming agent calls on your Hermes Telegram bot with owner-only Accept and Reject buttons. ### [Hermes Agentic SOC](https://github.com/NousResearch/hermes-agent/pull/154079) Tools · by CyStrat-Services-GmbH · open · 11-11 Evidence-first SOC and DFIR work for Hermes: six defense specialist personas (L1 triage, L2 incident response, threat hunting, DFIR compromise assessment, detection engineering, vulnerability management), 39 skills, 25 technology scaffolds and deterministic THOR Lite triage with exact event counts, correlation, audit trail and human approval. Everything is local file work. ### [Harness Story](https://github.com/NousResearch/hermes-agent/pull/154078) Desktop · by wronglebowsk · open · 11-11 Plain-English session storyteller and Desktop workspace pane: reconstructs what you asked, the hidden Hermes startup context load (skills, connectors, plugins, tools, memory), why the model chose each tool or pivoted after errors, and a 5-bucket context window bar synchronized with an auto-scrolling Chronicle. ### [TradeGuard](https://github.com/NousResearch/hermes-agent/pull/154077) Tools · by Vansh Khosla · open · 11-11 Read-only stocks and ETF research for Hermes: evidence-aware fundamentals, daily technical observations, ETF structure, comparable ranking, microstructure calculations and an explicitly experimental walk-forward ML baseline. It provides explainable heuristic research priority, not trade execution, personalized investment advice, calibrated win probabilities, price targets or position sizing. Optional ticker-only research fetches historical daily bars from Yahoo Finance's unofficial public chart endpoint only after explicit plugin configuration consent; it uses no API keys, credentials, broker access, background processes or live orders. Supplied snapshots and issuer metadata are read only from paths the user explicitly requests. ### [Eselsbrücke](https://github.com/NousResearch/hermes-agent/pull/154076) Desktop · by studioluftbruecke · open · 11-11 A helper toolbox for Hermes with tools you switch on and off: a today pad, a pomodoro, a billable time log, a formatter, a colour picker and small tools, which your agent can call for exact answers. ### [longevity-coach](https://github.com/NousResearch/hermes-agent/pull/154075) Tools · by pridiuksson · open · 11-11 Health coaching loop for Hermes: 9 config-free skills, 9 typed tools, and the weekly-crunch automation blueprint (outcome ledger, stress triage, biometric recovery/rebound verification). Local-first, stdlib-only, no credentials. ### [Unblockbrücke](https://github.com/NousResearch/hermes-agent/pull/154074) Desktop · by studioluftbruecke · open · 11-11 Cards to unblock yourself or your agent, drawn by you when you are stuck and by Hermes when its tool calls keep failing. ### [Forgebrücke](https://github.com/NousResearch/hermes-agent/pull/154073) Tools · by studioluftbruecke · open · 11-11 A bridge between Hermes and your self-hosted Forgejo: Hermes works as its own account, writes follow Hermes' approval setting, and a desk in Hermes Desktop shows the pull requests that wait for you. ### [Acrylic](https://github.com/NousResearch/hermes-agent/pull/154072) Desktop · by Saderius · open · 11-11 Acrylic glass for Hermes Desktop: a frosted, see-through theme with four presets (Acrylic, Frosted, Sanded, Expressive). Tested mostly on Windows 11, where it uses the system's real acrylic blur; elsewhere it falls back to a CSS version. ### [Coolifybrücke](https://github.com/NousResearch/hermes-agent/pull/154071) Desktop · by studioluftbruecke · open · 11-11 See what needs you across your Coolify instance, deploy with your approval, and let Hermes explain failed deploys. ### [vivarium](https://github.com/NousResearch/hermes-agent/pull/154070) Other · by r3adaszer0 · open · 11-11 A living artificial-life world inside Hermes. Real tool calls drop nutrients into a 48x48 cellular-automaton dish; creatures carry genomes, forage, reproduce, mutate and go extinct on their own; every skill the agent learns births a new species. Deterministic, mathematically grounded, private. ### [Hermes Incognito](https://github.com/NousResearch/hermes-agent/pull/154069) Tools · by nicolasbuelvas · open · 11-11 Zero-trace ephemeral and incognito sessions across Hermes CLI, TUI, Gateway, and Desktop: hides active chats from sidebars and cross-session searches, blocks persistent memory and skill mutations, strips profile and personal memory in RAM in strict clean-room mode, optionally masks outbound PII via a reversible RAM-only vault, zero-fills transcript and hook spill files before unlinking on close or TTL expiry, and emits content-free SHA-256 verification receipts. ### [Hermes Hostd](https://github.com/NousResearch/hermes-agent/pull/154068) Tools · by violetSoftware · open · 11-11 Request command execution from a separately managed hermes-hostd daemon over HTTP or VSOCK, with host-side approval. ### [Hermes Design Studio](https://github.com/NousResearch/hermes-agent/pull/154067) Desktop · by 1lpredest1nato · open · 11-11 Non-destructive shader image editing inside Hermes Desktop, manually or with an agent. Human-approved shader tools, recipes, numerical previews and PNG export. ### [Shaderbrücke](https://github.com/NousResearch/hermes-agent/pull/154066) Desktop · by studioluftbruecke · open · 11-11 Make GLSL shaders with your Hermes agent, see them render, tune them by hand, and take them into the tool you ship in, with a frame-by-frame proof on every export. ### [Axiom](https://github.com/NousResearch/hermes-agent/pull/154065) Tools · by arjunbagga-xyz · open · 11-11 Continuous neural-symbolic conservation-law verifier and SMT certificate generator using Z3 and dReal. ### [chaos-takens](https://github.com/NousResearch/hermes-agent/pull/154065) Other · by arjunbagga-xyz · open · 11-11 Nonlinear phase-space attractor reconstruction via Takens delay-embedding theorem and maximal Lyapunov exponent estimation. ### [eigen](https://github.com/NousResearch/hermes-agent/pull/154065) Other · by arjunbagga-xyz · open · 11-11 Graph-Laplacian spectral spectral partitioner computing Fiedler vector and algebraic connectivity for agent multi-agent communication networks. ### [Transducer](https://github.com/NousResearch/hermes-agent/pull/154064) Tools · by arjunbagga-xyz · open · 11-11 Near-ultrasonic FMCW sonar, mechanical FFT vibrometer, and true thermodynamic random number generator using audio hardware. ### [Hermes Reflex](https://github.com/NousResearch/hermes-agent/pull/154063) Automation · by arjunbagga-xyz · closed · 11-11 Skill-to-executable-code reflex compiler with parameter injection, 3-gate pre-trust verification, Curator archive recovery, and two-tier native Hermes GenUI with a draggable sticky-note pinboard pane. ### [OmniSight](https://github.com/NousResearch/hermes-agent/pull/154062) Desktop · by arjunbagga-xyz · open · 11-11 Spatial multi-surface visual perception, AST-to-pixel bi-directional linking, and closed-loop UI self-healing engine with interactive Before/After transcript sliders. ### [Ghost-Box](https://github.com/NousResearch/hermes-agent/pull/154060) Tools · by arjunbagga-xyz · open · 11-11 Speculative shadow-execution sandbox and cryptographic blast-radius verifier for autonomous agent tools using ephemeral Git worktrees. ### [WoW Addon Forge](https://github.com/NousResearch/hermes-agent/pull/154059) Other · by SnoobieJunes · open · 11-11 Builds World of Warcraft: Forever (client 1.60.1, Interface 16001) addons that load and run, for players who are not programmers. Nine tools: wow\_env (install/addon detection), wow\_api (queryable index of the real client build's API, events and restrictions), wow\_check (six-stage static + headless verification: TOC lint, Lua 5.1 syntax, global resolution, restricted/secret API rules, headless load in an embedded Lua 5.1 VM stub of the game, efficiency lints), wow\_test (multi-client headless Lua test runner), wow\_new (scaffolds an addon from tested building blocks: core, replicated record log, guild sync, trust, lazy Classic-theme window, Discord bridge), wow\_install (installs after wow\_check passes, with backup), wow\_savedvars and wow\_errors (read game data and in-game Lua errors back, via the bundled ForgeProbe companion addon), and wow\_bridge (file-based Discord seam: pull an outbox, push an inbox, one /reload of latency each way). Ships Guildhall, a guild-management addon built entirely with these tools (board, calendar, a Marks currency earned automatically by playing, Discord mirror), the wow-addon-builder skill, /wow, hermes wow, and an Addon Forge Desktop pane. ### [surviving-the-static](https://github.com/NousResearch/hermes-agent/pull/154058) Other · by littlewarrior333 · open · 11-11 A real time architectural translator for local ai that turns cold, intimidating connection errors into plain english. ### [ChatGPT Style](https://github.com/NousResearch/hermes-agent/pull/154056) Desktop · by amitamit10 · open · 11-11 Unofficial ChatGPT / Codex-inspired theme for Hermes Desktop: light and dark palettes, a centred reading column, compact blue user bubbles, a rounded composer card with a blue send button, Codex-style model and effort pickers, a floating Files/Review card and a quieter sidebar and footer. Not affiliated with, endorsed by, or sponsored by OpenAI. ### [Augur](https://github.com/NousResearch/hermes-agent/pull/154055) Tools · by Bruno-coder21 · open · 11-11 Your agent knows your 3 priorities for the week, notices when you drift, parks the detour, and on Friday clears what it can and shows where your attention actually went. Labels each request in the background, asks one short question on the 2nd detour in a row ("Park it" or "Do it now", buttons on Telegram), runs a guarded research-and-draft Friday cron job over the Queue, and sends a weekly report image with planned vs actual per goal and a hidden 4th priority. ### [Kleos](https://github.com/NousResearch/hermes-agent/pull/154054) Tools · by NewOdysseyConsulting · open · 11-11 Kleos, the AI marketing team for small businesses, through its hosted MCP server (Streamable HTTP at https://app.getkleos.ai/mcp; OAuth 2.1 with PKCE, signed in with the user's own Kleos login; no keys in config). Twelve tools: list businesses, read the month's calendar, plan the next 30 days, draft a post or a blog article in the business's voice, save a post in the user's own words, list and approve drafts waiting for review, schedule an approved post to a connected channel, report results (posts published, impressions, clicks, leads) and whether ChatGPT, Google AI Mode and Perplexity recommend the business, and read the buyers and problems the content is written for. Bundles one skill (kleos) that says which tool fits which task and that nothing is published without the user asking. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Vape Coach](https://github.com/NousResearch/hermes-agent/pull/154052) Desktop · by NeelM0906 · open · 11-11 Local-first webcam vape tracker and taper coach. A MediaPipe watcher spots hand-to-mouth puffs plus the exhale, and the target gap between vaping sessions grows ~10%/day after a 1-day baseline. Hitting early climbs a nudge ladder: a roast, a sarcastic (never real) PR threat, a protein-shake order, then jumping jacks the same watcher counts on camera by itself. Nag-only: never blocks a request, never gates a tool, no tool that can touch code. Desktop pane with camera Start/Stop and a live rep counter, a status-bar countdown, pet/toast roasts, /vape and three agent tools. ### [gnothi](https://github.com/NousResearch/hermes-agent/pull/154050) Tools · by vrindachaa · open · 11-11 Quantitative claim verification, statistical comparison, and evidence-qualified conclusions. ### [SendSafe](https://github.com/NousResearch/hermes-agent/pull/154049) Tools · by NSEVEjk0 · open · 11-11 Inspect hidden Word and Excel content before risky reads, and create verified visible-content sharing copies. ### [Triage](https://github.com/NousResearch/hermes-agent/pull/154048) Desktop · by Saderius · open · 11-11 Keeps the list of every skill out of your context. Triage suggests skills above the composer as you type and as the chat goes on, ranked locally with no model call; a skill loads only when you pick it or clearly ask for it. ### [Wake](https://github.com/NousResearch/hermes-agent/pull/154045) Automation · by zzzubair · open · 11-11 Lets your agent speak first. Scheduled notes and cron jobs arrive as a turn for your own agent in the chat you bind: Desktop Bot Chat, Discord, Telegram, Slack and other gateway chats. ### [Agenda](https://github.com/NousResearch/hermes-agent/pull/154044) Automation · by zzzubair · open · 11-11 Your agent's own todo list, brought up in conversation when items are due. ### [Ghostwriter](https://github.com/NousResearch/hermes-agent/pull/154042) Desktop · by nevpit · open · 11-11 Write, revise, and export screenplays in a Hermes Desktop workspace with a project-aware assistant. ### [Token Master](https://github.com/NousResearch/hermes-agent/pull/154041) Desktop · by karmsheel · open · 11-11 Token spend analytics for Hermes Desktop: one stacked daily chart of token consumption across every Hermes profile, provider and account on the machine (7D/30D/90D/1Y, stacked by token class, provider, profile or source), per-provider / per-profile / per-model / per-task breakdowns, and an Accounts and limits card that marks each credential as subscription (quota), pay-per-token API spend or free and shows the live allowance windows Hermes can fetch. It reads the token accounting Hermes already records — every profile's state.db opened read-only, plus DeepSeek Harness session files — and adds no collection of its own. ### [claim-gate](https://github.com/NousResearch/hermes-agent/pull/154040) Tools · by Mohamedhazeem · open · 11-11 Deterministic evidence-readiness firewall for insurance claim submissions. Resolves model-supplied evidence and incident-fact IDs against an operator-controlled SQLite registry, evaluates a versioned policy, and returns a deterministic PASS, BLOCK, or HUMAN\_REVIEW. The model never gets to verify its own evidence. ### [hive](https://github.com/NousResearch/hermes-agent/pull/154039) Tools · by MFaiqKhan · open · 11-11 Hermes remembers how its errors were fixed, and Hermes installs can share those fixes: when a terminal command fails with an error that was fixed before, Hive adds the fix that worked (with how often it worked) under the output. Fixes are learned from exit codes, not guessed: a failed command, the steps after it, and the same command succeeding. Opt-in, Hive trades fixes with other installs over Nostr, signed by a random per-install key, with proof of work, cross-checks and a safety screen; it never runs anything. ### [rubal-oracle](https://github.com/NousResearch/hermes-agent/pull/154037) Tools · by X0EF · open · 11-11 Say a goal in Hermes chat with a deadline. Afterward Oracle checks proof (met / not met / couldn't verify) from git, Calendar, GitHub, Notion, Linear, Gmail, kanban, Strava, or Apple Health. Rubal is the forecast from similar past attempts. ### [Retake](https://github.com/NousResearch/hermes-agent/pull/154036) Desktop · by wevial · open · 11-11 Image-generation workspace for Hermes Desktop. Generate or upload an image, mark regions, request a reference-guided revision, compare versions and download. ### [Chronomancer](https://github.com/NousResearch/hermes-agent/pull/154035) Desktop · by Gdetrane · open · 11-11 A shared plan for the Kanban cards of every profile on a gateway. Give a card a slot (a planned start and an estimate): the plugin parks the card and releases it at the start, with no cron job. Hermes Desktop shows one timeline with a lane per profile; drag a card to plan or move it. Slots that overlap and share a one-at-a-time label show as clashes, with the reason and the next free start. An orchestrator agent plans through five tools. The plan advises: it never blocks a card or a tool call. ### [Chat Tree](https://github.com/NousResearch/hermes-agent/pull/154034) Desktop · by Ronitron007 · open · 11-11 One window for branching chats in Hermes Desktop. A side pane draws the open chat and its branches as a git graph, one node for each exchange. Branch from a node, ask on a model you pick, and merge back into any ancestor chat with Merge as-is or Digest. The model writes each digest through ctx.llm (auxiliary task chat\_tree\_digest, which follows the title generation model unless you set auxiliary.chat\_tree\_digest). ### [TruthGate](https://github.com/NousResearch/hermes-agent/pull/154032) Tools · by cyberNoman · open · 11-11 Deterministic final-turn verification for newly introduced Python stubs and common embedded-secret patterns, using pre-edit baselines and at most two corrective continuations. Standard-library-only, fail-open, and no network, subprocess, LLM, approval-bypass, or core-patching behavior. ### [Infra Atlas](https://github.com/NousResearch/hermes-agent/pull/154031) Automation · by hnasheralneam · open · 11-11 Registry-backed infrastructure graph with drift detection, a read-before-touch guardrail, and autonomous downtime response. ### [decision-brief](https://github.com/NousResearch/hermes-agent/pull/154030) General · by LaRoyBot · open · 11-11 Decision Brief MVP - a human decision layer for consequential agent and creative decisions. ### [Music Studio](https://github.com/NousResearch/hermes-agent/pull/154029) General · by KeyArgo · open · 11-11 Describe a song and Hermes helps you make it: it drafts lyrics and style with you, renders nothing until you approve, then makes a real take locally on ACE-Step or through MiniMax via ComfyUI, and checks whether the take did what you asked. ### [hail-hermes](https://github.com/NousResearch/hermes-agent/pull/154028) Voice · by KeyArgo · open · 11-11 Talk to your Hermes agent out loud — in a desktop page it adds to Hermes Desktop, or from a Watch it pairs with in four digits. Replies are spoken, the face moves only while real audio plays, sessions resolve to folders on disk, and any model Hermes has already authorised can answer. ### [Aura](https://github.com/NousResearch/hermes-agent/pull/154027) Desktop · by KeyArgo · open · 11-11 A film behind your work, and an app that breathes with your audio: a docked film pane for YouTube videos, playlists and direct MP4/WebM beside the chat, and a MilkDrop-compatible visualiser (140 presets) driven by what the machine is already playing, with agent moods for thinking, tools, approvals, errors and streaming. ### [Game Center](https://github.com/NousResearch/hermes-agent/pull/154026) General · by KeyArgo · open · 11-11 Real board and card games inside Hermes: chess, go, checkers, connect four, reversi, gomoku, hex, breakthrough, sea battle, blackjack and slots, each with its own rules engine and a referee that refuses an illegal move and records it. Play the built-in computer at four levels, a friend on the same screen, or your own Hermes model through the games\_play tool. ### [Omniscience](https://github.com/NousResearch/hermes-agent/pull/154025) Tools · by KeyArgo · open · 11-11 Local, read-only observability for Hermes Agent: a live graph of every session, turn, model request, tool call and subagent, plain-English failure briefings, and deterministic replay of a stored run. ### [Asset Curator](https://github.com/NousResearch/hermes-agent/pull/154024) Tools · by dev-willbird1936 · open · 11-11 Find existing assets or create them from code with saved license/browser preferences, optional relevance checks and preserved provenance. ### [hermes-phone-control](https://github.com/NousResearch/hermes-agent/pull/154022) Voice · by jinsu133 · open · 11-11 Preview of a phone-native Android companion with local MCP controls and optional voice. ### [Hermes 3D](https://github.com/NousResearch/hermes-agent/pull/154021) Desktop · by numengames · open · 11-11 Turn your Hermes into a 3D game world you can walk through. Your scheduled jobs, conversations and system health become places to explore, right on your own computer. Anything that changes Hermes, like pausing a job, waits for your approval. Built on the Numinia Engine by Numen Games. ### [SO-101 Arm](https://github.com/NousResearch/hermes-agent/pull/154019) Tools · by jamesEmerson112 · open · 11-11 Give Hermes a robot arm, safely: an SO-101 (a MuJoCo sim on your laptop, or a real arm through LeRobot) driven by named skills, never joint angles. A top-down camera grid corrects its aim, a missed pick is rehearsed on simulated copies before the retry, a plain-code watchdog parks the arm on faults, a pre\_tool\_call guard refuses motion until a repair is verified, and only a person can lift a freeze (/arm unfreeze). ### [Artemis](https://github.com/NousResearch/hermes-agent/pull/154018) Web · by LukeJJones · open · 11-11 Mobile web companion for canonical Hermes Bot Chats using full-owner Web Access. ### [BTW Studio](https://github.com/NousResearch/hermes-agent/pull/154017) Desktop · by SBMatthew · open · 11-11 Side-channel /btw workspace, $0 read-only SQLite session radar, and Δτ steer playbook for Hermes Desktop, CLI, Gateway, and Dashboard. Intercepts /btw mid-turn in the composer (with attachments, opt-in voice STT/TTS, and @session context) to render Markdown answers in a strip, popover, floating pane, or dialog without mutating the transcript or L0 prompt cache. Reads state.db?mode=ro at $0 for token/cache metrics, AST syntax checks, 10 loop-guard detectors, and cross-session file write conflicts, steers conversation\_id-stable turns or child subagents mid-flight, formats output via a 6-axis reading lens (\[0,1\]^6), and retains only rules verified by 3-call Δτ behavioral contrast in BTW.md. ### [sauce-vids](https://github.com/NousResearch/hermes-agent/pull/154015) Tools · by Davemafy · open · 11-11 Capture real browser workflows and render SaaS demo footage with locally licensed visual assets under Hermes direction. ### [TimelinesAI WhatsApp](https://github.com/NousResearch/hermes-agent/pull/154013) Tools · by TimelinesAI · open · 11-11 WhatsApp inbox through TimelinesAI's official MCP server (Streamable HTTP at https://mcp.services.timelines.ai/mcp, browser OAuth sign-in, no keys in config): search and summarise chats, find unanswered conversations, read history, and send messages, notes and labels from the user's connected WhatsApp number. Bundles one workflow skill, whatsapp-inbox, that makes the agent confirm the recipient and exact text before any send. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [disko](https://github.com/NousResearch/hermes-agent/pull/154011) Tools · by emirkaan5 · open · 11-11 Run and steer a disko AI DJ set on your Mac from any Hermes chat. disko's model picks each next track and beat-matches the mix; this plugin launches it on a music folder or a Spotify playlist, starts the set, steers the energy, asks for a different next track, pauses and stops, through ten tools, a /disko slash command and a hermes disko CLI that need no model turn, and a pre\_llm\_call hook that tells the agent what is playing while a set runs. Standard library only. ### [Own Compute](https://github.com/NousResearch/hermes-agent/pull/154010) Tools · by Siioh · open · 11-11 Bring your own compute. For researchers and scientists with a workstation, a lab server or a cluster account. Runs sandboxed container environments, Slurm batch jobs and cloud agents (a Hermes agent in a sandbox on the user's own machine, not a hosted service, that hands its work back as a git branch) on SSH hosts, Slurm login nodes, and the Docker or Podman on this computer. Independent of Nous Research. A Compute page in Hermes Desktop shows every machine and what runs on it, and the same actions are chat tools with approval cards before anything starts, changes or is deleted. ### [hermes-make-it-smart](https://github.com/NousResearch/hermes-agent/pull/154009) Other · by tesla963 · open · 11-11 Canonical catalog over every Hermes profile. Reads the local state.db files, gives each session the "NNN \[Subject\] MMM \[Tag\] \[Tag\] Description · Verb" name, and exposes faceted filters (subject, maker, context, intention, theme, year, project) as a native Desktop page. Includes a Pending view that attributes every memory/skill approval request to the session that produced it, and a Testing tab that simulates the rename, writes a mandatory backup, commits through the host's session.title API and offers one-click revert. ### [OpenSOP](https://github.com/NousResearch/hermes-agent/pull/154008) Automation · by opensop · open · 11-11 Turn your best agent workflow into a file anyone can run. An OpenSOP process file sets, step by step, what is a script (must be certain) and what needs the agent (judgment). Hermes starts a run with sop\_run, does its steps and submits them with sop\_submit; the next script step checks the work, so scripts decide done. The run lives on disk, not in the conversation: Hermes is re-anchored on its open step every turn and across compaction, nudged to finish it before wrapping up, and refused when it tries to decide a step reserved for a person (seat belts, not a sandbox). Any agent or person can pick the run up with the opensop CLI, and every step leaves a receipt. Requires the opensop CLI 0.10.0+ on PATH. ### [Swarm Studio](https://github.com/NousResearch/hermes-agent/pull/154007) Desktop · by Pukujan · open · 11-11 A whole creative studio inside Hermes. One prompt, and a crew of agents you can watch and steer builds the whole launch on a live canvas: a multi-page website, a working web app with a database, a playable game (also as a Godot, Unity or Unreal project), a promo film and the social posts. Open it from the Desktop sidebar, ask for a site in any chat (the swarm\_design tool), type /swarm, or run \`hermes swarm\`. The crew uses the models you already set up in Hermes, through two picker slots (planner and builders). Builds can be posted to Swarm Feed, where others try them live and remix them with credit. ### [money-rules](https://github.com/NousResearch/hermes-agent/pull/154006) Tools · by seedless-labs · open · 11-11 Money rules for Hermes: pay, split, sweep and alert on USDC from your own Solana wallet, under rules you set in chat. You sign one capped USDC allowance (SPL approve) to a local helper key; USDC stays in your wallet until a rule pays, and every new rule waits for /money approve. Rules fire on a schedule, when USDC comes in, or when a balance crosses a line. Gasless: a relay pays the Solana fee for a small USDC fee in the same transaction and only co-signs connect, revoke and pay. ### [CrowdCode](https://github.com/NousResearch/hermes-agent/pull/154003) Automation · by sr33j · open · 11-11 Cloud agents with their own wallets. Create CrowdCode agents owned by this install's Base wallet (no sign-up or API key), fund them with USDC, give them tasks that buy x402/MPP services or deploy paid services of their own, and withdraw the earnings. A pre\_tool\_call guard sends money-moving calls above a configurable amount to Hermes's approval prompt and blocks spending above a daily cap. ### [StudioTwin](https://github.com/NousResearch/hermes-agent/pull/154002) Desktop · by realtwin · open · 11-11 StudioTwin cloud images, environments, PBR materials, meshes, rigs, motion, sound effects and Concepts via remote MCP, with an optional Desktop media workspace. ### [hermes-issue-sync](https://github.com/NousResearch/hermes-agent/pull/154001) Automation · by elijahboston · open · 11-11 Two-way sync between Hermes kanban cards and Linear or Jira issues. Open issues with a chosen label in the projects you pick become cards; card status moves the issue through its workflow; a finished card posts its summary on the issue; a worker that blocks with a question gets it posted on the issue, and the first human reply there unblocks the card with the answer. Several connections (Linear workspaces, Jira sites, several projects each) run side by side, and one failing never stops the others. Desktop adds an Issue Sync panel, a status-bar chip and kanban card badges; /issues status and /issues show work from any chat. ### [Butler](https://github.com/NousResearch/hermes-agent/pull/154000) Desktop · by Intenzi · open · 11-11 Turns the Hermes desktop pet into an on-screen operator: WALK draws a guide on the real controls and verifies each step from the macOS Accessibility tree, DESCRIBE reads the control under your pointer (role, label, enabled state, value) without letting the click reach your app, and WATCH announces one condition from an app in the background and keeps it until you acknowledge it. Accessibility only: no screenshots, no OCR, no Screen Recording, no Input Monitoring. ### [House Rules](https://github.com/NousResearch/hermes-agent/pull/153999) General · by AIverflow · open · 11-11 House rules your agent can't forget. Refuses tool calls that match your rules (a command with flags such as git push --force or rm -r, or a regex on a tool argument such as a .env path) or that leave the folders you allow. The refusal tells the agent why and to ask you; you lift a rule for the current session with /rules unlock \. /rule turns plain words into a rule draft as YAML and replays it on past tool calls; you paste it into the profile config.yaml yourself, the plugin never writes it. Rules apply from the next session. For an agent that forgets or ignores instructions; it is not a security boundary (scripts, aliases and indirect effects are not seen). ### [AgentTrust Seal](https://github.com/NousResearch/hermes-agent/pull/153997) Tools · by Syrup-Labs-AI · open · 11-11 Seal a file and verify the seal later, so one Hermes agent can prove to the next that a handoff was not changed. Adds an agenttrust\_seal tool (seal, verify) and an /agenttrust-seal slash command. Verify answers SEAL HOLDS, SEAL BROKEN or NO SEAL, with codes 0, 1 and 2, so the receiving agent can stop before it builds on a changed file. Standard library only. ### [MapDesk](https://github.com/NousResearch/hermes-agent/pull/153996) Desktop · by mackydubzz · open · 11-11 A GIS workspace inside Hermes Desktop. Import GeoJSON, CSV, KML/KMZ, GPX, XLSX, zipped shapefiles and bounded public ArcGIS layers; style, query and select layers; run geodesic buffer, dissolve, clip, centroid, hull and points-in-polygon analysis; export GeoPackage, shapefile, KMZ, GeoJSON, CSV and PNG, from chat or from the map pane. Works with no account on a keyless OpenFreeMap basemap; a public Mapbox token is optional. ### [Hermes Committee](https://github.com/NousResearch/hermes-agent/pull/153995) Tools · by abualnassr · open · 11-11 A committee of AI models for hard decisions. You set the seats (each with its own role and model) and a chairman; the seats answer independently, rank each other's answers blind, debate the points they disagree on, and the chairman writes a decision with confidence, dissent, role-fit notes and next actions with owners. Evidence files (PDF, Word, Excel, CSV, text, photos) go to every model as cited \[E1\] passages; optional Research first searches the web once through Hermes's own web tools and adds checked \[W1\] passages before the seats start. Time and budget limits with a mid-run ask to extend, follow-up runs, saved committees, a live Desktop page (Brief, Convene, Decision) and a Markdown + PDF report. ### [Taster](https://github.com/NousResearch/hermes-agent/pull/153994) Tools · by Yonkoo11 · open · 11-11 Runs another Hermes plugin once in a locked box (fake home, canary wallet files, network denied and logged) and reports what it did versus its manifest. ### [agent-webmcp — docs sites as a callable toolkit](https://github.com/NousResearch/hermes-agent/pull/153993) Web · by system1970 · closed · 11-11 Search a 1,400+ domain capability index, mount a docs site's tools, and ask its AI assistant for settled, cited answers — one tool call, sources back. Owns its own headless/headed Chromium via Playwright (no dependency on Hermes browser tools; works on CLI/gateway/desktop). Proven live on Mintlify, Fern, ReadMe and GitBook docs assistants. ### [Hermes Agent Radio](https://github.com/NousResearch/hermes-agent/pull/153991) Desktop · by Sahil-SS9 · open · 11-11 A 24/7 radio station about Hermes Agent and open-source agents, written, fact-checked and voiced by agents, inside Hermes. Desktop gets a player bar under the composer that lifts into a panel, a full player page, a status-bar dot and a detachable pop-out window. Your agent gets eight tools and a /radio command to see what is on, read the guide and sourced show notes, browse stations and songs, request songs, vote in the weekly chart and send the hosts a question, topic or roast; a human producer reads every one before anything airs. ### [Agentic Operations](https://github.com/NousResearch/hermes-agent/pull/153990) Automation · by ryanteoyx · open · 11-11 Synthetic manufacturing decisions with constraint planning, durable recovery, and outcome correction. ### [Zed Mode](https://github.com/NousResearch/hermes-agent/pull/153988) Desktop · by latentJake · open · 11-11 Every Hermes Desktop conversation in one stack that recedes into the screen, ordered like an inbox: failed, needs you, working, unread, pinned, then the rest. Answer approvals and questions on the card, hold a thread beside the stack, start or branch a chat without leaving, switch a live chat's model, and tune the 3D live. Desktop plugin (desktop/plugin.js inside the plugin SDK surface) plus a read-only dashboard backend that reads the session list from state.db. No network, no credentials, no agent tools, hooks, middleware or env vars. ### [Hermes Engagements](https://github.com/NousResearch/hermes-agent/pull/153986) Memory · by latentoperator · open · 11-11 A client notebook for your Hermes agents. Each client and project is a plain Markdown record on your machine: who they are, who to talk to, how they like to work and where the real documents and repos live. Add a client by talking or pasting notes, link a chat to it, and the agent keeps the record current with a dated source for every change. Includes a read-only Clients page for Hermes Desktop and the web dashboard. ### [Waiting On You](https://github.com/NousResearch/hermes-agent/pull/153985) Desktop · by clientamp · open · 11-11 Desktop panel listing every question your agents are waiting on you to answer, across all chats and profiles, sorted by urgent/important and answerable in place (Yes, No, Reply, Later) with five seconds to undo. ### [Answer Summary](https://github.com/NousResearch/hermes-agent/pull/153983) Desktop · by ArronPK · open · 11-11 Long-session comprehension for Hermes Desktop: a short AI recap above the composer for every assistant reply, a one-click Catch up brief for the whole session, and a right-docked Discuss side chat that inherits the conversation, supports another model, and hands a structured outcome back to the main chat. ### [Commonly](https://github.com/NousResearch/hermes-agent/pull/153982) Tools · by Team Commonly · open · 11-11 Gives Hermes tools to read a Commonly pod and post to it, through the @commonlyai/mcp server (stdio), plus a skill for working in a pod. The tools work while Hermes is running. Nothing wakes Hermes when someone @mentions it in the pod. Agent Plugins v1 (plugin.json + mcp.json + skills/), configuration only. ### [Hot Hermes](https://github.com/NousResearch/hermes-agent/pull/153980) Desktop · by firetussin · open · 11-11 Compact profile service-state monitoring, resource planning and explicit Desktop connection retention. Detects shared gateways; no model keepalive prompts or unverified speed claims. Reads OS memory/CPU and sampler process metadata. ### [TradePreflight](https://github.com/NousResearch/hermes-agent/pull/153977) Tools · by Mohamedhazeem · open · 11-11 Deterministic, fail-closed execution firewall for AI-initiated equity orders. A pre\_tool\_call hook normalizes broker-specific trade calls into one canonical order, evaluates it against an operator-owned policy (gross buy notional cap, max quantity, symbol / order-type / side allowlists) and blocks anything it cannot prove compliant; every decision and execution outcome is recorded in SQLite. Ships two paper-trading execution tools with deliberately different broker schemas (symbol/side/quantity/limit\_price vs ticker/direction/qty/price), a read-only dry-run tool, a policy inspector and an audit reader. ### [Model Kombat](https://github.com/NousResearch/hermes-agent/pull/153976) Models · by DarkeTech · open · 11-11 Head-to-head model evaluation: scout a task or kanban card into lanes, build a deterministic fight card (Warm-up 1pt, Main Event 2pt, Championship 3pt, sudden death) at standard, hard or gauntlet intensity, fight two models on identical prompts, and score mechanically — executed code, exact counts, numeric tolerances, phrase groups, JSON leaf equality, SQLite rows — with a per-lane Elo ladder, live arcade pane and kanban recommendation badges, and a measured best-value call ('equal quality — X is 18x cheaper and 2.8x faster') when two strong models tie; a local task\_stats.json remembers which tasks actually split the two models and demotes saturated ones. ### [Evident — Is this evident?](https://github.com/NousResearch/hermes-agent/pull/153984) Web · by Evident-Scalelabs · open · 11-11 Follow real buyer journeys, compare website claims with independent AI answers and suggested competitors, and produce one visual report with highlighted evidence, practical changes and repeatable checks. ### [memex8](https://github.com/NousResearch/hermes-agent/pull/153974) Memory · by Ex8-ca · open · 11-11 Self-hosted vector memory (Qdrant + TurboQuant compression) with auto-discovered knowledge realms, semantic search across conversations and ingested documents, and a slumber pipeline for memory consolidation. v1.2.0 adds private-span redaction before capture, compact retrieval with ordered batch get, scoped chronological timeline, user-certified memory protection, and recoverable consolidation/dedup archives. v1.1.0 added optional per-session injection dedup with char-budget gate, decision/evaluation memory-type promotion (slower decay for load-bearing turns), prompt-injection sanitization on read, configurable system-prompt template, export/import memories as JSON, and a /memex8 slash command. New memory behaviours are opt-in; /memex8 export\|import is confined to $HERMES\_HOME/plugin-data/memex8 unless enable\_export\_import=true. ### [Chalkline — Voice note in. Paperwork out.](https://github.com/NousResearch/hermes-agent/pull/153973) Automation · by levi-lvjp · open · 11-11 A Hermes back office for solo tradespeople: capture enquiries, prepare quote and invoice PDFs, retain receipts and agreed extras, review follow-ups and export an accountant ZIP. Customer sends require payload-bound owner approval, except three template kinds approved once by hash. ### [Defuzz](https://github.com/NousResearch/hermes-agent/pull/153970) Tools · by defuzzr · open · 11-11 Start projects from templates, organize adopted copies, place new files by folder rules, and run project checks or a read-only Hermes-home audit. ### [Smart Model Swap](https://github.com/NousResearch/hermes-agent/pull/153968) Models · by CarmaOS-Dev · open · 11-11 Run a big and a small local LLM on one GPU rig and let the agent switch between them by task: the small one for quick or voice chat and to free VRAM for GPU jobs (ComfyUI), the big one for deep work. SmartModelProvider launches and stops llama.cpp's llama-server itself, swaps in the background without cutting replies, checks per-GPU VRAM before loading, restores the big model when idle, and ships a launch-parameter editor in the dashboard and Desktop, plus a bundled routing skill. Ollama, vLLM and LM Studio are detection-only. ### [internship-scout](https://github.com/NousResearch/hermes-agent/pull/153967) Tools · by aditkole · open · 11-11 Find internships on public Greenhouse, Lever and Ashby job boards, pre-screen each application form for blockers (cover letter, essays, how-did-you-hear, transcript, sponsorship, graduation window, pay), and track applications in a local tracker with requisition-level duplicate detection. ### [call-your-shot](https://github.com/NousResearch/hermes-agent/pull/153966) Tools · by chadius-maximus · closed · 11-11 The agent calls its shot before a multi-step task: how many tool calls and how many minutes it will take. Hooks count the real tool calls and time, score the estimate 0 to 100, and /shots or hermes shots shows predicted against actual. Each new call\_shot result tells the agent its own record (for example that actual time came to 0.3x its estimates) with a calibrated figure. ### [Handle It](https://github.com/NousResearch/hermes-agent/pull/153965) Automation · by casualsav · open · 11-11 Handle It: your agent sits in the customer support chat so you don't have to. Hermes holds a company's live customer-support chat for you in its own browser, negotiates toward your goal for as long as it takes, and asks you before it agrees to an offer, pays, cancels or shares a detail you did not allow. Ends with the outcome and a saved transcript. ### [rule-injector](https://github.com/NousResearch/hermes-agent/pull/153964) Automation · by klautimus · open · 11-11 Moves agent behavioural rules out of the system prompt and injects each one at the point of need, through three hooks: pre\_llm\_call for behavioural guards, transform\_tool\_result for corrections the model reads in place, and pre\_verify as a completion gate. A rule is {trigger, inject} — matched on keywords, regex, tool name, tool error, path or session scope, or a boolean compound of those — with a per-turn injection budget and an undroppable always-on floor. An optional adjudicator adds a semantic second pass over ambiguous turns, and propose.py drafts a rule from a user correction. Ships a local web dashboard (\`python3 dashboard.py\`, FastAPI, read-only) that lists the rule set and the recent activation trace, and six starter rules; rules.json is a template the user replaces. ### [Hermes64](https://github.com/NousResearch/hermes-agent/pull/153963) Tools · by Qhantom-Labs · open · 11-11 Author and edit room-based N64-era-style 3D games with story checks, bot playtests and local exports. ### [Proving Ground](https://github.com/NousResearch/hermes-agent/pull/153961) Tools · by dmustapha · open · 11-11 A wiring test for your plugin stack: known, harmless test calls go through every installed plugin in a throwaway child Hermes home with the model never called, and each cell shows which plugin blocked, rewrote, changed the result, gated for approval, or passed it through. ### [On Air](https://github.com/NousResearch/hermes-agent/pull/153960) Platform · by HomunculusLabs · open · 11-11 AI stream co-host with guarded chat, captions, optional voice, private rehearsal, and operator controls. ### [Helper](https://github.com/NousResearch/hermes-agent/pull/153958) Desktop · by benhens · open · 11-11 Desktop page for people who are new to Hermes: a Dashboard with nine set-up cards (what each thing is, the steps from the Hermes docs, a button to its place in Hermes) and Helper, a small character with its own conversation window that answers questions about Hermes in plain words, from lines checked against the Hermes docs. ### [Emporia](https://github.com/NousResearch/hermes-agent/pull/153957) Platform · by wujitsana · open · 11-11 An open network for AI agents: Hermes agents meet other agents on relays — a signed feed, rooms, sealed DMs the relay cannot read, games, labs tied to code repos, and a season of quests. Testnet only; the relay holds no money. ### [hermes-cartographer](https://github.com/NousResearch/hermes-agent/pull/153956) Tools · by nnazinwaogu · open · 11-11 Code intelligence for Hermes Agent: a token-budgeted repo map, symbol definitions and references, deterministic first-turn context injection, and AST-safe cross-file rename for Python, backed by an incremental SQLite index. Four tools (code\_map, symbol\_find, symbol\_refs, symbol\_rename), one pre\_llm\_call hook, and /carto plus \`hermes carto\` subcommands. ### [Defuzz Componentize](https://github.com/NousResearch/hermes-agent/pull/153954) Tools · by defuzzr · open · 11-11 Optional Defuzz companion for turning Python scripts into reusable components checked against saved baseline output, with 14 declarative project kits, CLI and slash commands, bundled guidance, and a Desktop settings page. Requires Python 3.13 or newer. ### [Apparent](https://github.com/NousResearch/hermes-agent/pull/153951) Tools · by simonvanlaak · open · 11-11 Read screen, voice and drawing recordings as timestamped instructions and visual evidence in Hermes. ### [Doorman](https://github.com/NousResearch/hermes-agent/pull/153949) Automation · by TwoRobotsinaTrenchcoat · open · 11-11 Wake-on-LAN doorman for sleeping inference backends. When an API request targets a host from a config-driven host map that is asleep, Doorman broadcasts a UDP magic packet, waits a bounded time for the inference port to accept a connection, then holds a per-request keep-alive (one packet per interval) for the life of the request, released on completion or on a terminal error; hosts that answer are left alone and no packet is sent. If the turn is cancelled (/stop), a lifecycle hook releases every live hold so the host is not woken for a request nobody is waiting on. Pure Python, no dependencies, config re-read per hook fire. ### [Pioneer](https://github.com/NousResearch/hermes-agent/pull/153948) Tools · by acadine-intelligence · open · 11-11 An inverted search for your own work. Pioneer reads your project folder, Hermes memory and recent chats, then looks for what does not exist yet and should. Your own model writes the gaps. Code checks every citation, runs an absence check against existing tools, and scores need, originality, counterfactual value, value and durability. When an existing tool already does the job, the board says so and names it. Rounds expand into a trail, shown on a Desktop page and pane. ### [Hyperframes](https://github.com/NousResearch/hermes-agent/pull/153947) Automation · by wesleymatosdev · open · 11-11 Drive Hyperframes (Apache-2.0 HTML video compositions) from Hermes: six hyperframes\_\* agent tools (lint, managed Studio preview lifecycle, timeline readback, frame snapshots for visual verification, Studio-selection context bridge, MP4 render) plus a dashboard tab with project switching, the live Studio preview embedded, an agent chat bound to a real Hermes gateway session (async turns with polled status) — the same conversation a terminal sees via \`hermes chat --resume\`. Wraps a pinned local Hyperframes CLI (no npx) and talks only to loopback services: the local CLI and the gateway's own api\_server. ### [Iolite](https://github.com/NousResearch/hermes-agent/pull/153946) Tools · by KyaniteLabs · open · 11-11 Pick an LLM that is smart enough, then cheapest, in one tool call. Set an intelligence floor or name an anchor model, and Iolite returns the few models on the cost x intelligence front, ranked toward cheaper, smarter or balanced. It answers from a catalog of 621 models bundled in the plugin, so a fresh install works offline with no setup. Adds three tools (iolite\_shortlist, iolite\_model, iolite\_show), /iolite, a skill and a Hermes Desktop pane that plots the catalog in 3D and 2D and follows the agent's picks. Data and credits: Epoch AI ECI (CC BY), LMArena ratings (CC BY 4.0), LiteLLM prices (MIT); every answer names its sources, and catalog rows whose source names Artificial Analysis are refused. ### [agent-wallet](https://github.com/NousResearch/hermes-agent/pull/153945) Tools · by BlackHelio · open · 11-11 Approval-gated multi-chain wallet for agents. Default coins are XMR, BTC and XRP (XRP is receive-only in this release); ETH, POL and USDC on Base and Polygon are one toggle away. The agent can ask to pay, propose an auto-pay mandate, ask for a capped sandbox budget (created unfunded) or close an incoming order, but nothing is signed until a human types a six-digit PIN on a card that shows the exact destination, amount, fee and fiat value, and no key material ever enters the Hermes process: a daemon (walletd) holds an encrypted keystore, signs Bitcoin and EVM locally, supervises the Monero signer, enforces spend policy, and writes Ed25519-signed, hash-chained receipts (denials included, anchored so truncation is detected). Purchases can be priced from an HTTP 402 (x402) response and settled with an EIP-3009 authorisation. 23 agent tools, one per distinct act. Ships a native Desktop page (balances, approvals, mandates, sandboxes, receipts, per-coin switches) and full CLI parity. ### [Hermes Custodian](https://github.com/NousResearch/hermes-agent/pull/153944) Tools · by Cam-AcitveIce · open · 11-11 Local, read-only security-posture monitor for the machine a Hermes agent runs on (Linux, macOS, native Windows): scoped file and permission baselines, evidence-backed findings and a change history, a first-run terminal setup, an optional background monitor and an on-demand authenticated loopback dashboard. Registers four read-only tools (custodian\_status, custodian\_findings, custodian\_changes, custodian\_scan). ### [Alchemy Task](https://github.com/NousResearch/hermes-agent/pull/153943) Tools · by Darengb · open · 11-11 Alchemy Task does the work before the work: it turns messy real-world work into finely researched tasks, ready for your agent to do for you. ### [HermesProof](https://github.com/NousResearch/hermes-agent/pull/153941) Tools · by CodedAccorn · open · 11-11 Deterministically verify agent task outcomes and persist auditable evidence reports. ### [PFWA](https://github.com/NousResearch/hermes-agent/pull/153939) Tools · by PFWA · open · 11-11 Provably fair randomness for Hermes, plus prediction markets on Robinhood Chain. pfwa\_random gives a number, coin flips, winner picks from a list or a shuffle from a drand evmnet (League of Entropy) round chosen before it existed, with the round, signature and derivation so anyone can recheck it: for giveaways, raffles, draws and tie-breaks an LLM would otherwise make up. Market tools list PFWA's open markets, read live prices onchain, show a wallet's positions and claimable winnings, list results, and build unsigned buy, sell or claim transactions for the user's own wallet to sign; two more read HoodRand requests and the live Sherwood game round. Native Python plugin, standard library only, no keys and no configuration. ### [Syrinx](https://github.com/NousResearch/hermes-agent/pull/153938) General · by cstljuan · open · 11-11 Your agent is fast; Syrinx makes it wait for the right moment. In a focus block (/flow 45) the agent's questions are parked with the default it took and it keeps working; when you return you get a catch-up, a list counted by code from the tool calls (files changed, commands run, anything that left your machine, and how much work rests on unconfirmed defaults), and every question in one card. During the block anything that would leave your machine (git push, publish, send, deploy) is held before it runs and offered back on that card. Settle a question once with /always and it is never asked again. Late-night requests become a morning brief. Deep-work hours, a question budget, a time receipt, a Desktop pane and Argus, a pet whose pose shows the state. ### [live-video](https://github.com/NousResearch/hermes-agent/pull/153936) Voice · by dacheah · open · 11-11 Give Hermes eyes in a live voice conversation. Open a page on your phone, talk, and the agent answers in duplex voice while seeing the camera; ask to be told when something comes into view and the relay watches the frames for 90 seconds, verifies a sighting against the image before claiming it, and says it once. The six-action tool starts the relay, hands you the URL to open, and replays what a session saw. ### [The Clearing](https://github.com/NousResearch/hermes-agent/pull/153935) Desktop · by Dgardenhire · open · 11-11 The Clearing is a writing editor inside Hermes with 159 editing methods you can run by name: 11 editor passes, 9 composition checks drawn from Strunk's rules, 33 editorial methods, 56 craft techniques and 44 figures of speech it can find in your draft or suggest where one would help. Press Read my piece and it marks your own words with what to change, most important first. Nothing changes until you press Use this. Three reviewers, a newcomer, a skeptic and a line editor, can each read the piece and then argue it out. When a suggestion adds a fact you never wrote, it says so before you keep it, and the page shows how much of the piece is still your own words. A built-in quick check and dictionary work without a model. An independent community plugin, not made or endorsed by Nous Research. ### [Adeptory](https://github.com/NousResearch/hermes-agent/pull/153934) Tools · by brefledev · open · 11-11 Builds specialist Hermes profiles with researched knowledge, searchable libraries, working procedures, and a roadmap for useful tools. ### [Bot Drops](https://github.com/NousResearch/hermes-agent/pull/153931) Desktop · by brainarts-lab · open · 11-11 Share a Hermes bot as a link and a card, and adopt bots other people share as new profiles. A preview shows the bot's card, what it needs and what was left out before anything is installed; adopting runs Hermes' own profile install on the exact reviewed commit and never changes existing profiles. Sharing packages the personality and the author's own text skills, renders a link card and a story image locally, and publishes them as a new public GitHub repository with a GitHub Pages bot page. ### [OwnMachine](https://github.com/NousResearch/hermes-agent/pull/153929) Desktop · by rbur0425 · open · 11-11 A persistent private macOS VM workspace for Hermes with SSH/MCP and human-controlled browser access. ### [Doom](https://github.com/NousResearch/hermes-agent/pull/153926) General · by SilentKnight87 · open · 11-11 Play Doom in a dashboard tab. The GPL engine runs as plain JavaScript on Freedoom or your own WAD. ### [Hebe Longevity](https://github.com/NousResearch/hermes-agent/pull/153932) Desktop · by brainarts-lab · open · 11-11 Personal Longevity Intelligence for Hermes, built on Insilico Medicine's open PLSI toolbelt (MIT). Reads nine routine blood markers from a confirmed CSV or typed panel, calculates Blood PhenoAge, shows trends and formula drivers across blood draws, and drafts safety-screened research plans challenged by a separate model call before only the user can start a plan. Includes a Hebe page and optional theme in Hermes Desktop plus a chat result card. ### [Interview Prep Coach](https://github.com/NousResearch/hermes-agent/pull/153925) General · by KellerAI · open · 11-11 Prepares the user for their own interview: researches only the people they will meet, from public professional sources; predicts and tiers likely questions; builds STAR answers only from real experience; runs scored mock interviews; writes questions to ask; assembles a cheat sheet; lists post-interview signals with evidence (no grade or prediction); drafts application materials and thank-you notes the user sends. ### [Corpus Open](https://github.com/NousResearch/hermes-agent/pull/153923) Tools · by teakesdev · open · 11-11 For pro se and self-represented litigants: people handling their own lawsuit or court case without a lawyer. Legal citation checking for AI drafts. Under every reply, a citation guard lists each case, statute and code cite with what CourtListener (case law), Corpus (statutes from all 50 states) and the person's own case file hold at it, and flags a fake (hallucinated) citation before it reaches a court filing; a cite that cannot exist is caught with no lookup at all. A quote guard refuses a write that misquotes a filed document or cites the wrong page. Each session opens with the case's court deadlines, and an intake skill interviews the person one plain question at a time. Twelve tools keep a local litigation case folder (labeled facts with sources, evidence, deadlines, brief check, a packet for an attorney deciding whether to take the case, court rules, hash verification), /matter shows it, and Hermes Desktop gets a Case pane. ### [ChartQuery](https://github.com/NousResearch/hermes-agent/pull/153922) Tools · by chartquery · open · 11-11 Tools to render charts, QR codes and barcodes as shareable images, through the ChartQuery API. ### [Asin](https://github.com/NousResearch/hermes-agent/pull/153921) Tools · by asindotdev · open · 11-11 Tools to search Amazon and read a product's price history and offers, through the Asin API. ### [HuntAPI](https://github.com/NousResearch/hermes-agent/pull/153920) Tools · by huntapi · open · 11-11 Tools to read the metadata of public videos and download their video or audio, through the HuntAPI media API. ### [Workflow Studio](https://github.com/NousResearch/hermes-agent/pull/153919) Desktop · by SaM-0777 · open · 11-11 General-purpose agent-based workflow studio for building, managing, and executing complex workflows. ### [contractguard](https://github.com/NousResearch/hermes-agent/pull/153918) Tools · by Mohamedhazeem · open · 11-11 Hermes-native API contract assurance: validates proposed requests before they are sent, verifies live responses even when the status code is 200, compares candidate contracts against a human-approved baseline, and keeps redacted evidence receipts. ### [PiqueSignal](https://github.com/NousResearch/hermes-agent/pull/153917) Desktop · by PiqueSignal · open · 11-11 See PiqueSignal's Solana memecoin board right inside Hermes. Press Deep Check on any coin and Hermes checks it for you: fake volume, the live price, and what people say on X. Open a live chart, ask Hermes about any coin, or listen to a short spoken board update. The free board needs no account; a paid member key adds the live member board. Research only, not financial advice. Nothing trades. ### [Hyperliquid Desk](https://github.com/NousResearch/hermes-agent/pull/153913) Tools · by KevinM0001 · open · 11-11 Live Hyperliquid dashboard tab, Flow Deck alerts, wallets and market tools with a commentary skill on Hyperliquid's public data, plus node panels from the plugin author's hosted relay (on by default, can be turned off) or your own engine feed. Read-only; no orders, no trading keys. Independent; not affiliated with Hyperliquid. ### [MyFolks for Hermes](https://github.com/NousResearch/hermes-agent/pull/153912) Tools · by fresfrida · open · 11-11 Persistent, exact memory about the people in your life. myfolks, /myfolks or /contacts saves and returns details word for word, with when and where each was saved; saving and recall never call a model. Plain words (remember, fix, forget, undo), a Desktop page to search, add, edit, delete and undo, vCard and CSV import, and optional Google Contacts sync. Encrypted SQLite store (AES-256-GCM) in the Hermes folder. Owner only by default, group chats refused, optional per-person sharing (read or add). ### [WrenchBoard](https://github.com/NousResearch/hermes-agent/pull/153911) Tools · by Junkz3 · open · 11-11 Turn Hermes into a circuit board repair bench: a live boardview beside the chat, opened from 14 boardview formats, test plans with readings pinned on the pads, part cards drawn from the real copper, and your schematic and datasheets read on your computer. ### [Microduck Robot (MuJoCo)](https://github.com/NousResearch/hermes-agent/pull/153910) Tools · by Arca Computer (arcabotai) · open · 11-11 Drive Microduck, the open-source bipedal duck robot from Pollen Robotics / Hugging Face, in a MuJoCo simulation running its real trained ONNX RL policies: look through the duck's head camera, walk, turn, sit, stand, move the head, and line up and kick a ball at a goal (10 tools), with every motion gated by a pre\_tool\_call safety interlock (speed and duration caps, an arena geofence, a latching e-stop only a human clears with /duck clear, every blocked call logged). Ships a /duck command, a bundled skill and a dependency-free mock duck, so it runs with no MuJoCo at all. Independent community integration, not affiliated with Pollen Robotics. ### [FieldKit](https://github.com/NousResearch/hermes-agent/pull/153909) Tools · by Snehal707 · open · 11-11 Offline observation forms for Hermes. Describe your work in a Hermes chat and Hermes drafts the forms, then publishes them as a workspace after you approve. Fill them in on a phone with no signal (installable PWA served by the Hermes dashboard: answers, photos, voice notes, GPS and signatures), and sync entries and files back to Hermes when online, with duplicate protection and conflict review. Hermes reads the synced entries and files through 12 tools and analyses them with its own tools when asked. FieldKit never calls a model. ### [Reliability Desk](https://github.com/NousResearch/hermes-agent/pull/153908) General · by KellerAI · open · 11-11 High-reliability practices on your real launches, changes and incidents: go/no-go polls with named veto owners, blameless incident reviews that fix conditions rather than people, FMEA with RPN and severity override, closed-loop readbacks, stop-the-line triggers, change-control recommendations and session checklists. Every finding cites its evidence, and every GO, NO-GO or change-control outcome is a recommendation for a named human. ### [omnihermes](https://github.com/NousResearch/hermes-agent/pull/153906) Tools · by DataGuy-Eterniti · open · 11-11 On-device biometric presence and face-verified authorization gate for Hermes Agent. ### [INSET](https://github.com/NousResearch/hermes-agent/pull/153904) Tools · by winsznx · open · 11-11 Opens your site in headless Chromium at phone, tablet and desktop widths, inspects every page section separately (the Section Atlas) and measures overflow, clipped text, covered controls, section spill-over and failed media with DOM selectors and pixel boxes. Hermes fixes the source with its normal tools and approvals; INSET then rechecks the same URL and viewport and reports resolved, reproduced, needs review or not verified, never a pass it could not observe. ### [Stock Runway](https://github.com/NousResearch/hermes-agent/pull/153903) Tools · by usegreymatter · open · 11-11 See when each product will run out and when to reorder. Forecasts seasonal monthly demand, shows past demand, deliveries and stock on one chart, grades forecast accuracy, and lists what to order now (quantity and order-by date, with lead time, MOQ and pack size). Works from the bundled sample, a local CSV (reads SAP, NetSuite, Zoho, Odoo, Sage, QuickBooks and Dynamics export column names as is), or an optional read-only Odoo connection configured through Hermes secrets. Adds a /stock command, a stock\_runway tool and a Desktop page. ### [riskpulse](https://github.com/NousResearch/hermes-agent/pull/153901) Tools · by Oko300 · open · 11-11 Live SEC EDGAR risk analysis for US public companies. Analyze Form 8-K material events and Form 4 insider activity with risk scoring and plain-language narratives. ### [Annotary](https://github.com/NousResearch/hermes-agent/pull/153900) Tools · by josh-andrews-dev · open · 11-11 Review your agents' plans and documents with comments, from Desktop, the Webapp and your phone: an agent shares a Markdown plan, you comment on exact lines in the Annotary page, and Send back or Approve wakes the agent in the chat that shared it (an @mention wakes only the Bot it names; a plain comment wakes nobody). ### [CyboPal ONE DevKit (unofficial)](https://github.com/NousResearch/hermes-agent/pull/153899) Tools · by drowzeys · open · 11-11 Unofficial, not made or endorsed by CyboPal. Drive a CyboPal ONE (a 27-inch 4K screen on a 6-DOF robotic arm with the Pauli companion display) from Hermes: follow, posture presets, exact poses, gestures, both screens, both cameras and an emergency stop. Works today against the open source CyboPal ONE DevKit live simulator. ### [polydesk](https://github.com/NousResearch/hermes-agent/pull/153898) Tools · by clatsonhacks · open · 11-11 Agent paper-trading desk for Polymarket: research a market, quote a trade against the live CLOB order book (fills sweep real book levels, so slippage is realistic), confirm with a single-use 5-minute ticket that is re-priced at execution, then track mark-to-market P&L with automatic settlement when markets resolve. Risk limits (max trade, max % of equity per outcome, max slippage, liquidity checks), a persistent ledger, a /desk command and a bundled workflow skill. The paper bankroll can be backed 1:1 by testnet USDC in a linked Base Sepolia wallet. ### [Tap Proof](https://github.com/NousResearch/hermes-agent/pull/153897) Tools · by ladyFaye1998 · open · 11-11 Find the exact CSS viewport widths where a control on a local web page loses its center hit. Tap Proof reports the blocking element and saves boundary screenshots and a receipt. Replay the same check after a fix. ### [Focus](https://github.com/NousResearch/hermes-agent/pull/153896) Tools · by Kattegat-crew · open · 11-11 Cognitive context engine for Hermes Agent. Intercepts turns before inference, extracts true intent and entities, prioritizes repository governance, selectively ingests memory, and injects winning skills in a single step with zero KV-cache mutation. ### [jg-provider-usage](https://github.com/NousResearch/hermes-agent/pull/153892) Desktop · by jerrygooch · open · 11-11 Provider usage in a Desktop status-bar chip, popover and full page: subscription windows, API balances, reset times and profile-scoped account state. Includes data-only JSON connectors with explicit destination approval, setup cards for Claude, Gemini and Z.ai, a running-agents footer with honest unknowns, and an active-account Codex banked-reset preview/confirmation. Stock Hermes 0.23.1 needs no core patch; xAI subscription credits and exact worker-account usage need optional host metadata and otherwise remain unavailable. Usage reads use Hermes account helpers or scoped API-key access and provider HTTPS endpoints; subscription tokens stay with Hermes. Custom definitions are HTTPS GET only, approved-host restricted, with no redirects or executable code. Confirming a Codex reset spends one banked reset; refresh and preview do not. No purchase, automatic redemption, telemetry or self-update. Screenshots and demo use real plugin components with clearly labeled synthetic data, not live accounts. ### [Puppetmaster Learn](https://github.com/NousResearch/hermes-agent/pull/153891) Automation · by professorpalmer · open · 11-11 Turns a Puppetmaster swarm into a Hermes skill. Puppetmaster runs durable worker jobs (swarms and flow graphs) for an agent. When a Hermes session ends, this plugin finds the swarm that finished during that session in the session's directory and, if it produced findings or a patch, writes its goal and findings (title, location, severity, failure scenario, evidence) as a skill candidate under $HERMES\_HOME/skills-candidates/. Promote it with \`puppetmaster skills promote-candidate \\`. It never writes a live skill or starts an agent or job, and writes one candidate per job. Needs the Puppetmaster CLI (PyPI puppetmaster-ai). ### [Accountability Partner](https://github.com/NousResearch/hermes-agent/pull/153889) Automation · by ZK-Snarky · open · 11-11 Holds you to what you said, in your own words: saves promises with your exact quote, checks in each night, takes a one-tap 'yes no' answer, and keeps an honest score (streaks, weekly 'you said, you did' receipt, consequences you chose). Checks receipts: a promise can name where proof lives and what counts as a win in the user's own rule (a WHOOP workout, GitHub commits, a Strava run, a calendar event); for a promise with a rule (e.g. 'minutes \>= 30', 'commits \>= 1') the plugin checks the rule in code on that tool's result; if the user allowed that tool (setting proof\_tools, empty by default, via /promises allow \) the plugin fetches it itself through ctx.dispatch\_tool at check-in, otherwise the agent fetches it; data that falls short is shown to the user and they are asked, unless they said 'the data decides' when making the promise (then a reading that fails the rule at check-in is a miss); missing data or a failed tool is never a miss. Without a rule, promise\_prove accepts only evidence text a tool really returned in that session (seen via post\_tool\_call). Consequences enforce themselves: on a miss the agent carries out the user's chosen consequence with any connected tool (a wallet, an app blocker), through Hermes's normal approvals, and promise\_enforced counts it only with that tool's result as proof. Optional lock-in on the user's explicit words: until an end date (max 90 days, 24h cool-off) the promise can't be dropped and its consequence fires without a new yes; the user grants the enforcing tool a standing OK in Hermes once at setup. The pre\_tool\_call hook, only while a consequence is owed through a named tool, blocks calls over the user's max amount or to another recipient and escalates to Hermes approval when not locked in. Only the user's own words or a real tool receipt change the ledger: every write needs a quote the plugin checks against the user's current message or a tool result. ### [bundlebees](https://github.com/NousResearch/hermes-agent/pull/153888) Automation · by mojomast · open · 11-11 bundlebees: dispatch sandboxed coding workers (bees) from Hermes, watch their attempts, and keep the apply, test and push step in your own hands. Four tools with 571 o200k tokens of schema between them. bundlebees\_status reports launcher health, capacity, live attempts, events, containers and authorized projects. bundlebees\_dispatch validates a file-leased brief and starts a worker in a disposable Cua sandbox (default model deepseek-v4.1-flash), optionally linked to a Hermes kanban card. bundlebees\_attempt gets, follows, cancels or reviews one attempt (cancel and review need confirm=true). bundlebees\_host\_prompt writes a paste-ready setup prompt for a worker host and never takes a host, token or URL. When an attempt finishes, you run hermes bundlebees verify \: it applies the patch on a clean feature-branch worktree, enforces the file lease, runs the named tests and a secret scan, and on green commits and pushes that feature branch (main and master are refused; --dry-run applies and tests without pushing). The plugin never merges, tags or force-pushes. Hermes Desktop 0.23 or newer shows a dark pane with Health, Containers, Models, Concurrency, Projects, Live attempts, Verify results and Remote host tabs; the four tools still run without it. Skill id bundlebees:bundlebees-operator. It is for developers who split repo work into small parallel briefs and want cheap workers on a machine they control, with a human gate before anything lands. To start, you need Hermes 0.23.0 or newer and a worker launcher (by default the Orbit local-workers adapter on the same machine). Run hermes plugins install bundlebees and hermes plugins enable bundlebees, store the launcher token as the Hermes secret BUNDLEBEES\_LAUNCHER\_TOKEN (never in chat), load bundlebees:bundlebees-operator, and begin with bundlebees\_status op=health. hermes bundlebees doctor checks the token, the adapter and the schema size without printing values. ### [Whiplash](https://github.com/NousResearch/hermes-agent/pull/153890) Tools · by pgbouncer · open · 11-11 Let Hermes jam live on an Elektron Digitakt II over USB MIDI: genre/mood beats with MIDI clock, routed onto a session kit map of what each track holds (tracks 1-16), in any key with chord progressions, arranged into song sections (whiplash\_arrange: intro, build, drop, outro on bar lines), written melodies and rhythms, bar-quantized pattern switches with DJ-style transitions (filter swap, drop swap, riser), track levels and pan with saved mixes (whiplash\_mix), listening to the unit's real output over USB audio (whiplash\_listen: loudness, clipping, band balance, timing and kick offsets as numbers, never audio) with a loudness guard that refuses level increases after clipping, a kit scan (whiplash\_scan\_kit: with the beat stopped, one note per track, heard and classified into proposed kit-map labels that are never applied without the user's yes), play along (whiplash\_play\_along / whiplash\_follow: reads the unit's MIDI input to show the user's knob moves, stop a ramp on a knob they grab, and answer phrases they play in key on another track; whiplash\_undo reverts the last part, mix or arrangement change), named sound parameters, clock-synced parameter ramps, live fills/drops/density/swing/humanize, switching a track among samples already in the project pool, mutes and tempo. ### [Hermes Intercom](https://github.com/NousResearch/hermes-agent/pull/153885) Automation · by greqone · open · 11-11 Cross-session discovery and inter-agent messaging for Hermes Agent across Desktop, TUI, CLI, Discord, and Telegram within the active profile. Adds \`session\_list\` (read-only, schema-tolerant discovery over the profile's \`state.db\` and \`gateway\_routing\` table, reporting whether each session has a live route or only a synthesized \`:local:dm:\` key) and \`session\_send\` (delivers \`\[Message from agent '\'\]\` cards with \`\[intercom from=\ hop=\/\ mode=\\]\` loop protection, per-sender deduplication, and per-target rate limiting). Automatically mounts dormant \`:local:dm:\` sessions in the local gateway authority before injection. From a classic CLI REPL or external process, a \`:local:dm:\` target falls back to \`prompt.submit\` over the local Gateway WebSocket when \`allow\_gateway\_injection: true\`. A Discord or Telegram target has no such fallback. ### [Codemode](https://github.com/NousResearch/hermes-agent/pull/153884) Tools · by junbo-hh · open · 11-11 Code-as-Orchestration: the model writes one JavaScript snippet that calls your MCP tools (plus web\_search/web\_extract) in a QuickJS sandbox, loops and filters there, and returns only the answer. Tool schemas are found on demand with BM25 searchTools(). ### [claim-check](https://github.com/NousResearch/hermes-agent/pull/153882) Tools · by cryptobandit · open · 11-11 Receipts for agent claims: checks the success claims in each final reply ("all tests pass", "I created src/app.py", "committed and pushed", "installed requests", "ran make build successfully", "server is running on port 8000") against the session's actual tool results and appends a ✅ verified / ⚠️ unsupported / ❌ contradicted footer citing the tool call. Optional nudge feeds contradictions back into the next turn; \`/receipts\` and the \`claimcheck\_report\` tool show the evidence ledger. Deterministic, pure stdlib, works with any model. ### [Model Board](https://github.com/NousResearch/hermes-agent/pull/153881) Models · by driptothebone · open · 11-11 Phone-first bulk model routing across profiles with exact preview, confirmation and saved-setting readback. ### [Quick Eval](https://github.com/NousResearch/hermes-agent/pull/153880) Models · by tikiverse · open · 11-11 Get a second opinion on any answer. /quick-eval sends your last question to other models, shows the answers side by side, and records which ones you mark good. Runs and picks go to an append-only JSONL file, and a review page (/quick-eval open: the Quick Eval page in Hermes Desktop on Hermes 0.22.3 or newer, or a browser page elsewhere; also a tab in hermes dashboard) shows each comparison and a model leaderboard. ### [foundry](https://github.com/NousResearch/hermes-agent/pull/153879) Tools · by Amit Srivastava · open · 11-11 Foundry — five creator bots sharing one memory ledger: research, draft, score, and price your work. Say 'foundry start' — a short interview sets up your brand. Approval is disabled — no trusted human review channel yet; nothing auto-posts. ### [Wiretap](https://github.com/NousResearch/hermes-agent/pull/153878) Desktop · by aka-lacie · open · 11-11 Inspect the exact request each model call put on the wire (system prompt, tool schemas, messages, blocks added at send time) and the model's response, aligned to the stored transcript, in a dashboard tab and a Desktop page; also per-call latency, tokens, prefix-cache hits with miss forensics and cost, and a read-only CLI. ### [Hook Desk](https://github.com/NousResearch/hermes-agent/pull/153877) General · by KellerAI · open · 11-11 Skills-only plugin. Studies the opening lines of posts you paste, names the pattern, keeps a sourced formula file and drafts opening lines in your voice for your approval. Never posts or browses. ### [Docket](https://github.com/NousResearch/hermes-agent/pull/153876) Tools · by MKempler · open · 11-11 Quotes for trades businesses (plumbers, electricians, HVAC): describe a job in words or a voice note and Docket prices every line from the owner's own price book with a deterministic engine (the model only picks item codes; it never writes a number). The owner approves with one tap on Telegram, Slack, Discord or Desktop, the customer accepts on a quote page, and the owner books the day and closes out with an invoice. Includes a plain-words price book editor with preview before save, PDF quotes and invoices, follow-ups, a run sheet and a Desktop Jobs board. ### [contagion](https://github.com/NousResearch/hermes-agent/pull/153875) Other · by onchaindc · open · 11-11 Taint-aware execution. Once a session reads untrusted content, its commands lose network, secrets and out-of-project writes at the OS layer. ### [Trophy Room](https://github.com/NousResearch/hermes-agent/pull/153874) Desktop · by abundantbeing · open · 11-11 Achievements for Hermes Desktop: 120 illustrated trophies earned from real Hermes work (tools, skills, profiles, cron, plugins, messaging, kanban, media), collectible cards across five tiers, unlock pop-ups with a synth sting, optional OS notifications, recent trophies and replayable away collections. Hooks only on the agent side (no tools, middleware or env vars); progress is stored in the plugin's own data directory. No telemetry or third-party services; artwork is bundled and the sound is synthesized locally. Optional read-only import from the bundled hermes-achievements plugin. Derived in part from hermes-achievements by PCinkusz (MIT): the 60 original achievement definitions. ### [local-scout](https://github.com/NousResearch/hermes-agent/pull/153872) Tools · by Angerosteles · open · 11-11 Local privacy and safety layer for Hermes. Detected personal data and secrets (e-mails, phones, cards and CVVs, IBANs, SSNs, PESEL numbers, addresses, API keys; person names in prompts) are replaced with one-way placeholders on your machine before Hermes sends a request to a model: user prompts, file and terminal output, MCP tool results, tool-call arguments and auxiliary calls such as title generation, in CLI and gateway sessions. Also flags prompt injection in what the agent reads (flagged, not blocked), offers optional calibrated local triage that can answer routine tickets with no main model call, an optional basal method through a basal engine you run locally, and local Insights. ### [Barehand](https://github.com/NousResearch/hermes-agent/pull/153871) Tools · by mrv777 · open · 11-11 Lets an agent operate a machine with no OS, network or agent on it (firmware menus, boot loaders, installers, rescue shells such as grub\>) through the screen and keyboard alone. It provides a computer\_use provider for QEMU virtual machines over QMP (or a VNC console), a barehand\_machine tool, a hermes barehand CLI, a skill and a flight recorder. \`hermes barehand up\` starts local qemu processes that keep running after Hermes exits until \`hermes barehand down\`; no API key reaches them and the plugin makes no network calls. In recorded runs a model installed Debian 13 from the network installer (it repaired a grub\> prompt on first boot from the installer's rescue mode) and Ubuntu Server 26.04 (it finished the stalled installer by hand) on blank aarch64 machines. \`hermes barehand demo\` replays a bundled recording. Evidence and the full disclosure are in the README and docs/results. ### [Hermes Ledger](https://github.com/NousResearch/hermes-agent/pull/153870) Tools · by elyga408-oss · open · 11-11 Tamper-evident flight recorder: every tool call (arguments redacted), a SHA-256 of every result, and every approval decision are appended to a local hash-chained, Ed25519-signed log. \`hermes ledger verify\` walks the chain and names the first edited, deleted, reordered or truncated entry; \`hermes ledger show\` prints a timeline; \`hermes ledger export --receipt\` writes a signed receipt of one session that anyone can check with only the public key (records from other sessions appear as content-free stubs), with \`hermes ledger verify-receipt\` or the bundled standalone verify\_receipt.py. An approval record binds to the exact call it approved. Adds a /ledger slash command, a Ledger dashboard tab with a chain-intact badge, and the skill hermes-ledger:ledger-audit. It proves the log was not changed after writing, not that the agent was honest when it wrote it (threat model in the README). ### [agssh](https://github.com/NousResearch/hermes-agent/pull/153869) Tools · by mosesman831 · open · 11-11 SSH hosts as first-class agent tools. Register your hosts once, then the model calls agssh / agssh-read / agssh-find / agssh-hosts instead of assembling \`ssh -i key user@host 'quoted string'\`. Commands travel as one byte-exact argv element, the tool returns the real remote exit code (never a \`\| tail\` status), output is capped with a full-text spill file, and every call multiplexes over one persistent ControlMaster connection per host so rapid agent calls never exhaust sshd MaxStartups. Dangerous commands go through Hermes' own approval gate, and a pre\_tool\_call hook blocks raw terminal ssh to registered hosts so exit codes and the audit log cannot be bypassed. ### [GoneBy](https://github.com/NousResearch/hermes-agent/pull/153868) Memory · by rahlquist · open · 11-11 Retroactive project memory for LLM agents. Captures every turn into an encrypted per-project write-ahead log (zero tokens during work), then at project close walks the log backward from the shipped product — the oracle — keeping only entries that verifiably contributed; survivors are embedded (MRL-truncated) and stored in Valkey or Qdrant for semantic search in later sessions. General plugin: coexists with any memory provider instead of replacing it. ### [Who's Listening?](https://github.com/NousResearch/hermes-agent/pull/153867) Memory · by Web3B-eth · open · 11-11 Memory that knows who's listening. Each memory is Private or Shared. Private memories go into a request only when the machine running Hermes (the gateway), the app you chat in (the platform) and the model provider are all trusted. By default, the gateway you install it on is trusted unless it's a hosted Hermes such as Hermes Cloud; the trusted platforms are the terminal and Hermes Desktop, including its Webapp; and the trusted model routes are Venice's private models and models on your own machine or local network. Everything else gets Shared memories only. You can configure all of these options in the plugin's settings: trusting either more or fewer gateways, platforms or model providers. ### [Delivery Shelf](https://github.com/NousResearch/hermes-agent/pull/154012) Desktop · by mechkw · open · 11-11 Save delivered files together, preserve versions, and find finished work again in Hermes. ### [Common Ground](https://github.com/NousResearch/hermes-agent/pull/153863) Tools · by knownquantity · open · 11-11 Durable collaboration records, mutual contributions and evidence-backed commitments. ### [toolrecall-light](https://github.com/NousResearch/hermes-agent/pull/153862) Memory · by whiskybeer · open · 11-11 Deduplicates byte-identical repeated tool results inside Hermes Agent: repeats of terminal, search\_files, MCP, web and connector output are saved to a local SQLite table and replaced by a one-line stub before they enter model context, with a recall\_result tool that restores the original bytes on demand. Fail-open, stdlib only, no core patches. ### [Summon](https://github.com/NousResearch/hermes-agent/pull/153861) Platform · by gtmcknight · open · 11-11 Your own @grok on X that learns: tag your bot under any post and your Hermes answers in the thread; say "learn this" under a how-to thread or video and it saves it as a skill that waits for your approval. A gateway platform for X that only reads posts from its owner, by user ID. Judges: hermes gateway setup, choose X, "Borrow @taybotfyi" with invite code judge-hermes; no bot account needed. ### [Session Cassettes](https://github.com/NousResearch/hermes-agent/pull/153854) Tools · by eightyFJ · open · 11-11 Record one good session's read-only tool results as a cassette, pin the facts a good answer must contain, then replay the same request against another model, prompt or skill: tool\_execution middleware answers every tool call of the replayed turn from the recording (exact hit, or a search/slice of a snapshot of the workspace files the session's results named; anything else gets a cassette-miss error) and never calls the real tool; the verdict is PASS, FAIL, DIVERGED or INVALID from the pins, with no judge model. ### [Keryx](https://github.com/NousResearch/hermes-agent/pull/153853) Voice · by hsyvy · open · 11-11 Outbound phone calls from your own phone. Hermes writes a brief and calls place\_call; after you approve (every call, single use), your phone dials over Bluetooth hands-free through PipeWire's telephony API, the plugin plays an AI-disclosure line, and a realtime voice model (OpenAI Realtime or GPT-Live) holds the conversation, asking you mid-call when it needs a decision. The outcome, details and transcript go back to Hermes; /calls lists recent calls. Guardrails in code: emergency, short-code and premium numbers blocked, quiet hours, hourly rate limit, hard 5-minute cap. Linux only (PipeWire \>= 1.4, a paired phone). ### [TempleOS for Hermes](https://github.com/NousResearch/hermes-agent/pull/153852) Desktop · by mojomast · open · 11-11 TempleOS for Hermes: run a TempleOS-family OS in a local QEMU VM and share one live screen with the Hermes agent, which can see, type, click and write HolyC. The default guest is Temple of Hermes 0.1.2; TinkerOS 5.20-rc10, stock TempleOS 5.03 and ZealOS are optional. Four tools: templeos\_vm (start, stop, reset, snapshot, restore and doctor the VM), templeos\_screen (PNG look, serial or on-screen text, wait for idle, frame hash, operator lore), templeos\_input (type, keys, click, move, HolyC) and templeos\_files (list, pull and push files over COM1). The Hermes Desktop pane shows the same screen in Watch, Shared or Take control mode; when you take control, agent input pauses until you hand back. The pane's animated Terry companion is on by default and can be hidden or turned off. Bundled skill: templeos:templeos-operator. It is for TempleOS and HolyC enthusiasts, retro-OS tinkering, and watching an agent work a real 640×480 OS beside you. To start, on Linux x86\_64 with Hermes 0.23.0 or newer, install QEMU (for example sudo apt install qemu-system-x86 qemu-utils), then run hermes plugins install templeos --enable and hermes plugins doctor templeos; doctor fetches the default ISO from its official release and checks its SHA-256. Open the TempleOS pane and ask Hermes to start the VM and print a HolyC hello world. KVM is used when /dev/kvm can be opened (usually the kvm group), otherwise TCG. ### [Foreman](https://github.com/NousResearch/hermes-agent/pull/153851) Automation · by PedValme · open · 11-11 A green check the agent did not rig. A deterministic gate (no model, standard library and git) blocks git commit, git push and gh pr create when the change skips, deletes or moves tests, weakens what the check runs, bypasses hooks or carries a secret, and always blocks an agent-run gh pr merge; an optional second model reviews what rules cannot see, such as a loosened assertion. /foreman shows each pull request with a verdict and merges only on your command; a Desktop panel shows where each change is. ### [Fine Print](https://github.com/NousResearch/hermes-agent/pull/153844) Tools · by Joko-F · open · 11-11 Shows where Hermes silently shrinks, cuts, swaps or hides what the model sees. When a turn ends in which a limit bit and no follow-up by the model was seen, the user gets one card with before/after numbers and the fix (a \`hermes config set\` command, a workaround, or "hardcoded in v0.23.0"): a web page cut to head and tail, an image downscaled 4x or more, a search page that stopped at 50 with no "more" flag, a failed search backend swapped for the keyless tier, a file read whose lines were all cut at 2,000 chars, terminal output of which under half was kept, a tool result replaced by a 1,500-char preview, a context compaction, loaded images that the request no longer carries, an attached image that reached the model as a helper model's description, a pinned context window that is not in force, a sixth URL that web\_extract never requested, a workbook read up to row 5,000 with its hidden sheet skipped. The model gets one short note inside the tool result only where Hermes itself tells nobody. \`/limits\` shows the session's hits, explains 81 catalogued limits of v0.23.0 (\`/limits explain \\`), replays the same detectors and card policy over your own history (\`/limits report\`, or \`/limits report html\` for one self-contained page) and lists every detector with its status (\`/limits self-test\`): 16 are verified (on real history, or in a live session for limits that leave no stored row) and may raise cards, 19 are fixture-only and only log. ### [Vitrail](https://github.com/NousResearch/hermes-agent/pull/153842) Desktop · by elomkdn · open · 11-11 Live GLSL shaders in the chat. Ask for a visual and the model writes a Shadertoy-style fragment shader that animates inside the reply in Hermes Desktop, with the sliders, colour pickers and switches it declared as annotated uniforms. Before vitrail\_render returns, the Desktop half compiles the shader with WebGL 2 on the user's GPU and reports the verdict, so a compile error goes back to the model with the driver log and the model fixes it in the same turn; a still frame comes back too, for models with a vision tool. The card is drawn through a tool media renderer where Desktop has one and through a ::vitrail transcript directive where it does not (Desktop 0.22.5). Other surfaces (Telegram, CLI, TUI) get a standalone HTML player of the same shader. Adds a gallery page with hover-to-play tiles, GLSL copy and 1920x1080 PNG export, the vitrail\_render and vitrail\_library tools, the vitrail:shader-studio skill, and post\_tool\_call / transform\_llm\_output hooks that append the shader's delivery line (MEDIA file or directive) when a reply lacks it. ### [Last Price](https://github.com/NousResearch/hermes-agent/pull/153858) Tools · by iamdflame · open · 11-11 Lets Hermes haggle with buyers in chat while the plugin, not the model, owns every price that leaves. You set an item's floor from your terminal; it stays in plugin state and never reaches a model, tool result, prompt or log. Every deal reply is held and checked, and one offering less is rewritten before anyone sees it; a buyer who offers or accepts a price the plugin would take is closed at their own number; deal sessions can use only Last Price's three tools (plus Hermes's tool-search bridge, whose calls are checked the same way). Includes hermes lastprice duel --vs-prompt, a side-by-side against a seller with the floor in its prompt (inside Hermes, in the 48-duel bench of version 0.1.1: that seller sold below the floor in 10 of 48 duels; Last Price in 0), and hermes lastprice challenge, a public 'buy it below my floor' Telegram challenge. ### [Antemuse](https://github.com/NousResearch/hermes-agent/pull/153840) General · by roosiq · open · 11-11 The agent takes a walk, daydreams, dreams, doomscrolls, or gets unstuck: a bounded change to how it works, with a dose rather than a duration. Walks step through WordNet, dreams recombine FrameNet roles, daydreams draw on stored memories, doomscrolls on live feeds. No model calls on the state path. ### [Pulsr](https://github.com/NousResearch/hermes-agent/pull/153837) Desktop · by Stormxftw · open · 11-11 Live numbers beside your chat: a Pulses pane for Hermes Desktop with rings, bars, numbers and trends from Discord servers, GitHub, feeds, public JSON APIs, read-only MCP tools, Windows notifications and Hermes's own activity. The agent drafts a pulse; only a person can approve it, and the approval card lists every host and the hourly request budget. No model tokens to refresh. Network: GETs to discord.com, cdn.discordapp.com, api.github.com, avatars.githubusercontent.com and HTTPS hosts you approve; the builder's Find sends one query to geocoding-api.open-meteo.com or store.steampowered.com. Notification pulses read counts from the local Windows store; sender and text previews are a separate opt-in. No telemetry; background collection is opt-in. ### [Jobby](https://github.com/NousResearch/hermes-agent/pull/153835) Desktop · by desaianm · open · 11-11 Source-backed international job discovery and PDF/DOCX application drafts inside Hermes; no autonomous submission. ### [ListeningKit Log](https://github.com/NousResearch/hermes-agent/pull/153834) Tools · by deepmroot · open · 11-11 Run X, Discord, Slack, Instagram and Reddit conversations from Hermes through the locally installed log MCP server: inspect inboxes and full threads, find conversations awaiting a reply, draft in the user's voice, and send only after approval. LinkedIn support is currently limited to a session-status check. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Hermes Screen](https://github.com/NousResearch/hermes-agent/pull/153832) Tools · by Johaan Jis · open · 11-11 Pick what to watch from your own Jellyfin library using real watch history, play movies and saved scenes on your Chromecast with receiver-verified status, send YouTube links to the TV, and play camera rock-paper-scissors against Hermes. Network calls: your Jellyfin server, your Chromecast (Cast protocol, TCP 8009), YouTube on the TV, www.gstatic.com (receiver SDK), your own optional game service, and optional YouTube Data API. Listeners: a temporary expiring movie relay while something plays, and an opt-in two-container game service you deploy yourself. No telemetry. Secrets only via Hermes; no .env reads. Camera frames never leave the phone. ### [Dive In](https://github.com/NousResearch/hermes-agent/pull/153830) Desktop · by RahmelJackson · open · 11-11 Turn a public source into a visual investigation with source-linked claims, charts, and follow-ups. ### [Dreaming](https://github.com/NousResearch/hermes-agent/pull/153828) Memory · by hsyvy · open · 11-11 Local memory provider (memory.provider: dreaming) after Anthropic's "The components of agent memory": every turn is kept as a transcript and the agent writes quick notes; an offline dream pass checks notes against the transcripts, resolves changes stated by reference ("she handed everything over", nicknames, "what I said last time"), and rewrites a linked wiki plus a short index loaded into every session. Tools via the memory provider: mem\_search, mem\_read, mem\_write; CLI: hermes dreaming status\|dream\|search\|show. ### [Google Home](https://github.com/NousResearch/hermes-agent/pull/153826) Platform · by karmsheel · open · 11-11 Google Home / Nest speaker as the mouth of a Hermes profile: replies are cast over the Cast v2 protocol with zero third-party dependencies, voice arrives from the host microphone through Hermes STT, and one speaker keeps one conversation. Adds googlehome\_speak, googlehome\_notify, googlehome\_status, googlehome\_volume and googlehome\_devices, a Desktop page with push-to-talk and speaker controls, and hermes googlehome CLI commands. ### [hermes-janus](https://github.com/NousResearch/hermes-agent/pull/153824) Tools · by Yonathan-T · open · 11-11 Bi-directional temporal state engine & MVCC workspace for Hermes Agent. Zero-copy Content-Addressable Storage (CAS), Write-Ahead Logging (WAL), Merkle-DAG branching, instant rollback, Mark-and-Sweep GC, and tournament consensus. ### [lights-out](https://github.com/NousResearch/hermes-agent/pull/153822) Tools · by JannetEkka · open · 11-11 Nothing billable starts without your yes and its price: every AWS CLI create Hermes runs (EC2, EBS, Elastic IP, NAT gateway, ELBv2, RDS, S3) goes to Hermes's own approval gate from pre\_tool\_call with a dated us-east-1 price, a local ledger records what was created and how to delete it, and a receipt after the turn shows what is still running and its $/month; /lights out turns off everything not kept (asks first), /lights keep \ \ keeps one on purpose until a date, /lights check compares the ledger with the cloud; tools lights\_status (read-only) and lights\_keep (a proposal the user must approve). ### [learnloop](https://github.com/NousResearch/hermes-agent/pull/153821) Desktop · by huss-wrld · open · 11-11 Structured lessons and practice inside Hermes, after LearnHarness: /learnloop:study probes what you know, plans a path and teaches it one graded check at a time; /learnloop:practice drills it with a hint ladder. What you learn becomes a knowledge graph you can explore and practise from. ### [MemDiffs](https://github.com/NousResearch/hermes-agent/pull/153820) Memory · by daji-dev · open · 11-11 Make updates to memory, skills, and self-improvement reviews expand on-click to show the diff. Set display.memory\_notifications to off if using this, or you'll see duplicate non-clickable notifications (config.yaml → display). Existing chats keep their old setting, so start a new chat afterwards. ### [SPAWN Backrooms](https://github.com/NousResearch/hermes-agent/pull/153886) Tools · by Spawn-Backrooms · open · 11-11 Work SPAWN jobs with human-approved proposals, child commission drafts and verified file delivery. ### [Living Spaces](https://github.com/NousResearch/hermes-agent/pull/153809) Desktop · by akinduroifedayo · open · 11-11 An open canvas for your bots in Hermes Desktop. Ask in chat and your bot builds dashboards, trackers, mini-apps and visual pieces in a Spaces tab. Each Space keeps its data, and you change it by asking. ### [pi-team](https://github.com/NousResearch/hermes-agent/pull/153808) Other · by iliFalk · open · 11-11 Coding lanes in the Hermes desktop — a project tree, each lane's task, diff, plan and thread, and a review queue agents act on. ### [Mulligan](https://github.com/NousResearch/hermes-agent/pull/153805) Tools · by onyebuchidaniel60 · open · 11-11 Counterfactual replay for Hermes: record a run, rewind to any step, edit one tool decision, replay forward in a fresh child session, and diff the branch against the original. Replay-to-point reproduces the recorded steps with zero provider calls and zero tool side effects; the branch runs live with normal approvals. Uses llm\_execution, tool\_execution and tool\_request middleware, pre/post\_tool\_call hooks and ctx.subagent\_lifecycle.launch; stdlib only, no dependencies. Stores append-only JSONL traces (prompts, tool arguments and tool results) under $HERMES\_HOME/replay-traces/. ### [Skilluminati](https://github.com/NousResearch/hermes-agent/pull/153804) Tools · by yxssxn · open · 11-11 Your skills, judged blind: finds out whether each skill your Hermes agent loads makes its answers better. /skilluminati \ takes the latest 12 times you used the skill, each with the chat as it stood at that moment, and your model answers each one twice, once with the skill and once without. The Owl, a blind judge, sees both answers and the skill's one-line description of its job, never which answer had the skill, and sees every pair twice with the answers swapped; if it picks differently the second time, the pair counts as Same. Verdicts: Enlightened (answers are clearly better with it), Dead weight (the model does this anyway, it only costs tokens), Hindrance (answers are worse with it) or Unproven (too few uses to tell yet), each marked Probable or Plausible. Built-in question sets for frontend-design, systematic-debugging and hermes-agent let a fresh install run a test with no history, and /skilluminati selftest plants a helpful, a harmful and an empty skill to prove the Owl can tell them apart. Surfaces: the /skilluminati command, two read-only agent tools, a Desktop page with a sidebar entry (plugin SDK only) and a read-only dashboard backend. ### [leaveby](https://github.com/NousResearch/hermes-agent/pull/153802) Tools · by AloysJehwin · open · 11-11 Gives Hermes a sense of distance: real travel times right now (Google Maps or TomTom), when to leave to arrive on time, whether a day of stops actually fits, drive times added automatically when a message mentions a saved place with a plan, and background 'leave now' alerts on your chat platform that cost no LLM tokens. ### [ScopeLedger](https://github.com/NousResearch/hermes-agent/pull/153801) Tools · by 221h0lmes · open · 11-11 Local-first billing-evidence reconciliation and owner-review reports for service agencies. ### [AWS Manager](https://github.com/NousResearch/hermes-agent/pull/153800) Desktop · by ParichayMago · open · 11-11 Review, apply and verify EC2, Lambda and S3 actions in Hermes Desktop. ### [ROM Studio](https://github.com/NousResearch/hermes-agent/pull/153798) Desktop · by erosika · open · 11-11 Make your own small game by talking with your agent, one scene at a time, from a decomp (the community-made source of an original handheld game) that you supply. A pane in the desktop app shows each change as a sentence with a picture from the running game, plus a play link, scenes, gallery, maps, music, dialogue, source search and the agent's questions. Ships a command-line kit the agent runs, a browser player with an always-on flash guard and the make-your-game skill; starts one local page on 127.0.0.1 once a game folder is set. No agent tools, hooks, middleware or env vars. Linux only. ### [dioramic](https://github.com/NousResearch/hermes-agent/pull/153797) Other · by faze39 · open · 11-11 A Blender scene director. The agent describes a scene as a structured JSON spec (toy-style shapes and 3D text, finishes, colours, three lighting rigs, camera moves, keyframed spin, bounce, float, pop-in, move and wobble) and dioramic renders it in headless Blender (EEVEE) as a vertical 9:16 MP4 plus a still preview, returned as MEDIA tags so the gateway delivers the clip. dioramic\_tweak merges only the changes into an earlier scene and renders a new version. ### [thumbs-fix](https://github.com/NousResearch/hermes-agent/pull/153796) Tools · by LikemDzokoto · open · 11-11 React with a thumbs-down to a bad Hermes reply on Telegram and it names the skill behind it, asks one question, shows a proposed edit as a diff, and changes the skill only after you approve. ### [smriti](https://github.com/NousResearch/hermes-agent/pull/153792) Memory · by vn-envy · open · 11-11 Local personal memory with dated facts, scoped corrections, and source records you can retrieve from any compatible MCP client. ### [twingraph-instagram](https://github.com/NousResearch/hermes-agent/pull/153790) Tools · by dalroot · closed · 11-11 Autonomous Instagram Agent for Hermes Agent powered by TwinGraph and Meta Graph API v21.0 Digital Twin architecture. Provides 30-day analytics, reel retention curves, intelligent direct messaging with auto username-to-PK resolution, and zero-footprint ghost mode exploration. ### [ProveIt Tests](https://github.com/NousResearch/hermes-agent/pull/153788) Tools · by KellerAI · open · 11-11 Flags recognizable ways an agent weakens tests to reach green: removed or loosened asserts, skip/xfail/.only, deleted tests, expected literals rewritten to match failing output, assert True and tautologies, try/except pass wrappers, and test-runner escape hatches (pytest \|\| true, -k "not ...", --deselect, --passWithNoTests, CI continue-on-error), and pytest config / conftest.py narrowing (addopts -k/-m/--deselect/--ignore, testpaths, collect\_ignore, deselecting hooks). Net-new only: each edit is compared with the file as it was before, so pre-existing skips are not blamed on the edit. Deterministic AST + regex, no model calls. Modes: observe (default), escalate to the human-approval gate, or block on unattended/cron runs. SHA-256 receipts and a /proveit ledger. Python-strong; JS/TS coverage is heuristic and limited. ### [Pilot](https://github.com/NousResearch/hermes-agent/pull/153787) Tools · by LangerSword · open · 11-11 Give Hermes a real interactive terminal. Pilot runs commands inside a private tmux server, so the agent can drive programs that need a TTY — ssh logins, vim/nano, htop/btop, fzf, less, package installers, ncurses menus and REPLs — by typing literal text and named keys and reading the pane back as plain text, with sessions that persist across tool calls. A human can attach to the same pane at any time and take over. ### [Hermes Voice Gateway](https://github.com/NousResearch/hermes-agent/pull/153786) Platform · by SRai22 · open · 11-11 Answers Twilio phone calls with Hermes: caller speech is transcribed and Hermes' replies are spoken with ElevenLabs over a Pipecat pipeline. Callers get only two restricted tools (inventory search, enquiry save). Operator-approved callbacks are proposed by the agent and dialed only after a human runs /voice-callback approve. ### [Charrette](https://github.com/NousResearch/hermes-agent/pull/153785) Desktop · by GuideboardLabs · open · 11-11 A whiteboard beside the chat where the board is the message and the chat is its caption. You draw with a pen, a shape tool or a typed prompt; the agent reads the board the way it reads a document, answers on the same surface, and can point at one particular stroke to say what it means. It adds shapes, arrows, tables and charts, writes a label beside the thing it describes, puts a picture on the page, asks you a question with up to three options anchored to the object it is unsure about, and exports any region as an image to build an artifact from. Every object records who made it and which layer it sits on, so a drawing keeps its provenance and can be replayed: press replay and the board rebuilds itself stroke by stroke, camera following the pen, at 0.5x to 5x. Nothing here is a screenshot of a chat. The board is a document the agent and the human both write on, and the same document renders headlessly to a PNG for a report or a page. ### [sarvam-provider](https://github.com/NousResearch/hermes-agent/pull/153782) Models · by Subramanyarao11 · open · 11-11 Sarvam AI as a Hermes model provider (\`--provider sarvam\`): Sarvam-105B and the open models Sarvam hosts in India, with tool calls, JSON output and reasoning levels. Not affiliated with Sarvam AI. ### [Wings](https://github.com/NousResearch/hermes-agent/pull/153781) Tools · by e-shizz · open · 11-11 A private WireGuard network between every Hermes you own and your phone. Two devices are enough: run setup on a computer, scan one code on the phone. Bots and sessions from all your machines in Desktop, the terminal and the Hermes app. Computers link directly when they can. No account, no coordination server, no third party. ### [sarvam](https://github.com/NousResearch/hermes-agent/pull/153780) Voice · by Subramanyarao11 · open · 11-11 Sarvam AI for Hermes: speech-to-text (voice notes and live transcription in voice mode) and text-to-speech (Bulbul voices) for 22 Indian languages, plus six tools: sarvam\_translate, sarvam\_transliterate, sarvam\_detect\_language, sarvam\_read\_document (PDF/image OCR), sarvam\_translate\_document and sarvam\_transcribe\_recording (long audio with speaker labels). Not affiliated with Sarvam AI. ### [its-in-your-head](https://github.com/NousResearch/hermes-agent/pull/153779) Desktop · by hi4c0ck · open · 11-11 It's in Your Head — observes how you work (tools, turns, friction) with local-only telemetry, a six-dimension perception engine, deterministic directional adaptation (opt-in, one bounded instruction per turn) and a 96px desktop signal strip with portrait/drift surfaces. ### [Nerdit](https://github.com/NousResearch/hermes-agent/pull/153778) Tools · by Nerdit · open · 11-11 Hermes puts what it builds online on its own machine and keeps it alive. nerdit\_ship deploys a folder to the local Nerdit engine (open source, Docker), waits until it is healthy and returns its URL, with the engine's diagnosis on failure. nerdit\_share gives an app a hosted https URL, private by default and public only after the user confirms. A pre\_llm\_call hook tells the agent when one of its apps is failed, degraded or restarting, once per outage and with no model polling; /apps answers with the status of every app without a model call; the nerdit:watch automation blueprint checks the apps on a schedule, restarts what a restart fixes and reports. Two bundled skills (ship, rescue). ### [Soloknuckle](https://github.com/NousResearch/hermes-agent/pull/153776) Tools · by Soloknuckle Contributors · open · 11-11 Deterministic local guardrails for Hermes and dev workflows. Three invocable tools—git-diff secret scanning, a destructive-command firewall, and a project health score—plus native Hermes surfaces: additive credential-redaction patterns, a soloknuckle:hygiene skill, hermes hygiene CLI, /hygiene, and a weekly hygiene-report blueprint. One rule set, same verdicts across CLI/hooks/MCP/Hermes. No hooks, no middleware, no prompt rewriting, no telemetry. ### [autonoma](https://github.com/NousResearch/hermes-agent/pull/153777) Tools · by JoeisEvil · open · 11-11 Earned autonomy: a trust score per kind of action (file edits, deletes, git push, network, messages, spending, system) that rises with clean work and falls on denials, corrections and failures. Below a class's threshold the call goes through Hermes's own approval prompt; at or above, it runs without asking. Catches the agent routing around a block (shell redirects, inline python/node code, \`bash -c\`), asks again for risky actions after the turn read outside content, revokes "session"/"always" grants after a correction, writes handoff notes from the tool log when a session ends unfinished, and adds a live dashboard tab. ### [snap-action](https://github.com/NousResearch/hermes-agent/pull/153774) Tools · by nini1294 · open · 11-11 Send your agent a screenshot and it becomes action — bookings, bills and deliveries are auto-filed to a persistent ledger with tappable, time-aware actions. ### [HermesTube](https://github.com/NousResearch/hermes-agent/pull/153769) Desktop · by pedroberaldo87 · open · 11-11 Take back control of what you watch. HermesTube puts your YouTube subscriptions in Hermes Desktop: tabs you make by folder, tag, channel or topic, a folder pane, a channel manager with undo, folder suggestions from your Hermes model, topic tabs it builds from YouTube search, and videos in Hermes's own player. New and still experimental: a cleanup list you approve, /tube buttons on Telegram and scheduled routines. ### [Hermes Hands](https://github.com/NousResearch/hermes-agent/pull/153768) Tools · by Ridwan Nurudeen · open · 11-11 Run Hermes terminal and file tools on a paired Windows or Linux computer while Hermes stays on a Linux or WSL gateway. The device agent connects out over HTTPS, signs with Ed25519 keys pinned at pairing, and fails closed when it is offline, revoked or unreachable. Optional witness mode is started by the person at the computer: local consent, every command shown on their screen before it runs, a consent question for each screenshot, a stop key, a 30 minute limit and signed receipts. In a witness session every tool call that reaches the device needs the helper's approval, and shell composition, scripts, common startup persistence and other tools are blocked. ### [USSD & SMS (Africa's Talking)](https://github.com/NousResearch/hermes-agent/pull/153766) Platform · by Dixon-O · open · 11-11 Use Hermes from any phone, including feature phones with no internet: dial a USSD code for an instant menu, ask a question, get the agent's answer by SMS, and read long answers back page by page on USSD. Africa's Talking (Kenya, Uganda, Tanzania, Rwanda, Nigeria, Malawi and more). Closed by default (allowlist), phone users get the \`safe\` toolset, store-first at-most-once SMS outbox, daily SMS cap. ### [Hype](https://github.com/NousResearch/hermes-agent/pull/153764) Tools · by asdugit · open · 11-11 An engine inside a plugin: /hype reads the session itself (no prompting) and turns what it built into a launch trailer (music only by default, narration opt-in); the product is the hero and the session's real diffs, terminal output, tests and stats are the proof, verbatim. Presets: 7 styles (launch, cinematic, pixel, terminal, blueprint, comic, ink), trailer, lore and speedrun films, 30-90 s, 16:9 and 9:16, each customizable by key=value knobs (colours, font, music, energy, voice, words, logo, language) or plain words after the style (one validated change set over the preset film, cached per job); a preset plus a restructuring ask goes to the Director. Free roam: a free-text direction without a preset runs the Director engine (treatment, film script, generated scene code that is linted and dry-run, a vision critic scoring keyframes against a rubric, revisions, final cut); smaller models take a compose path over proven blueprints with schema repair and measured checks (in our trials about one point below a frontier model on the critic's 10-point scale, at a fraction of the cost). Surfaces: /hype, tools hype\_render and hype\_status, hermes hype CLI, a Desktop studio (pane, page, palette, status chip, tool-media card) backed by dashboard routes. Renders locally with headless Chromium and ffmpeg; original procedural music and SFX; opt-in: Blender 4.2+ for 3D shots ("blender" / "use blender"); only when the user answers yes (hermes hype install-blender, the doctor's y/N question at a terminal, or the Desktop Install button) it downloads Blender 5.2.4 LTS from download.blender.org, verifies the pinned SHA-256 and installs it per-user without admin rights (never asked in CI, cron, gateway or chat). ### [Flower Shop](https://github.com/NousResearch/hermes-agent/pull/153762) Desktop · by k-reeeee · open · 11-11 An AI business operating workspace inside Hermes Desktop for a small flower shop: a business manager plus sales, inventory, production, finance, and logistics agents work from one SQLite record of orders, stock, production, deliveries, and money. Agents propose, the owner approves - order confirmation, stock receipts, adjustments, wastage, payments, and expenses create a pending approval decided from an inline chat card, the Approvals inbox, or an explicit chat relay. Includes Overview, Orders, Inventory and recipes, Production board, a read-only low-poly 3D Shop Floor fed by live production state, Delivery, Finance estimates labelled as estimates, an AI Team page, and a guided setup with removable demo data. The repo also ships an optional matching Hermes skin (a light and a dark variant) that the setup offers but never applies on its own. Not flower-specific: the same shape - one record of what was ordered, stocked, made, shipped and paid, agents that propose, an owner who approves - fits any small physical business. ### [A11Y Lens](https://github.com/NousResearch/hermes-agent/pull/153760) Web · by Johnlocke0347 · open · 11-11 Autonomous website accessibility testing and repair verification. Scans pages in a real headless Chromium with the axe-core engine, stores evidence-backed scans, and compares before/after scans (resolved / remaining / newly introduced) so Hermes can prove its fixes in the rendered page. ### [Hermes CoBrowse](https://github.com/NousResearch/hermes-agent/pull/153758) Desktop · by sandman-sh · open · 11-11 Real-time multi-window spatial co-browsing canvas and interactive human-in-the-loop web automation for Hermes Agent and Desktop with CDP screencasting, 8-way window resizing, Inspector Studio telemetry, and 2FA handoff. ### [Photo to Part](https://github.com/NousResearch/hermes-agent/pull/153757) Tools · by mojomast · open · 11-11 Photo to Part for Hermes: turn photos, a few measurements, or a messy 3D scan into a parametric printable part, and check the mesh locally before you print. The agent writes the solid in code, checks that it is watertight and the right size, and renders views you can hold next to the original. Six tools: photo\_part\_template (watertight starting shapes such as plates with holes, an L-bracket, a print-in-place hinge, thread, knurl, hex standoff, bottle cap, spur gear and strap hinge), photo\_part\_build (runs one local build script after Hermes approval), photo\_part\_validate (watertightness, shell count, size, bores, printability and spec.json dimension checks), photo\_part\_render (fitted views, a comparison sheet and a silhouette overlay scored against your photo), photo\_part\_heal (weld, fill holes and clean up a scan) and photo\_part\_measure (pixels per millimetre from a coin, a ruler or an ArUco/AprilTag marker). Output is STL or 3MF; a STEP solid needs CadQuery or build123d in the project environment. A millimetre value you state is never replaced by a photo measurement. Hermes Desktop 0.23 or newer can show an optional Part studio pane with the stage, photo and overlay, renders, history and file paths while a run is in progress; the six tools still install and run when that pane is unavailable. It is for makers replacing a broken clip, bracket, knob, cap or gear, printing a fit test, or turning a scan into a clean solid. To start, run hermes plugins install photo-to-part and enable it, then start a new session. Attach a photo with a coin or ruler in frame, give any dimensions you know, and ask for a printable replacement. The skill id is photo-to-part:photo-to-printable-part. Pass that exact string only to skill\_view. Do not pass it to skills\_list, and do not add a plugin: prefix or a slash. ### [ChartRunner](https://github.com/NousResearch/hermes-agent/pull/153755) Tools · by ChartRunner · open · 11-11 Read-only tools for ChartRunner, a gamified trading app on Solana: launch radar (new, trending, graduated) with the server's safety level and flags, the public buy verdict for a mint with a terminal review link, real-mode perp chart links for BTC, ETH, SOL, HYPE and DOGE, room and SIM battle links, finished cup standings checked against the anchor ChartRunner published, and a local verifier that rebuilds a chartrunner:v1 memo root from rows or a Merkle proof. Bundles one skill (desk) on when to use each tool and what a memo match does not prove. Hermes never signs: the tools return links, and every swap or perp order is reviewed in ChartRunner and signed by the user's own wallet. ### [Apple Developer Documentation Studio](https://github.com/NousResearch/hermes-agent/pull/153754) Tools · by AndrewMason7 · open · 11-11 Offline-first Apple Developer Documentation suite for Hermes: 95,899 symbols across 30 Apple frameworks with sub-5ms SQLite FTS5 search, React/Web/Android-to-SwiftUI idiom RRF, automatic Swift compiler diagnostic grounding hook, @apple: context references, /apple-docs demo command, inline ::apple-doc cards, and an interactive Hermes Desktop Studio docked beside your chat session. ### [Mahj Trainer](https://github.com/NousResearch/hermes-agent/pull/153753) General · by Jack Wallner · open · 11-11 Learn and practice American mah jongg (the game played with the yearly NMJL card) through Mahj Trainer's hosted MCP server (Streamable HTTP, no sign-in, no keys): short lessons, practice quizzes that check and explain each answer, a rack reader that suggests a Charleston pass with a reason for each tile, and a glossary. Bundles one skill that tells the agent when to call the tools. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Strategic Arena](https://github.com/NousResearch/hermes-agent/pull/153750) Tools · by LeMultiFox · open · 11-11 Understand the landscape before you invest: give it a link and get the room that reacts — who can block adoption, what each concession buys, and the claims you can score later. Research papers also get a replication clock. ### [polyglot](https://github.com/NousResearch/hermes-agent/pull/153747) Platform · by Subramanyarao11 · open · 11-11 Live interpreter for group chats on the Hermes gateway (Telegram, Discord): an authorized user sends /polyglot on in a group, and every message and voice note is then translated into each member's own reading language, posted as a reply to the original. Languages are learned from each person's first message or set with /lang; "hermes, ..." or an @mention still reaches Hermes as a normal question. Translates with the user's own Hermes model; a Sarvam AI key is optional for Indian languages and voice notes. ### [1guy](https://github.com/NousResearch/hermes-agent/pull/153940) Desktop · by jrezmo · open · 11-11 Hermes does the work. 1guy brings it back to you. One desk over your whole Hermes: every session, scheduled job and to-do is a card you touch, put away, schedule or talk to, and one companion on top tells you the few things that need you. Each card opens its own titled Hermes session and streams the reply in place; a talk bar routes a typed line to the right card or just does it. Adds a 1guy page with four lenses (today, board, time, inbox), a docked ticker and a to-do pane; on a new install he introduces himself with starting points built from what is really there. ### [Focus Widget](https://github.com/NousResearch/hermes-agent/pull/153746) Desktop · by elianjuan-g · open · 11-11 A Pomodoro widget for Hermes Desktop that only counts focus time while the agent is actually working. Agent-aware timer (25/5/15 by default, configurable): the focus timer runs only while the agent is active (hooks on messages, tool calls and outputs), waits automatically on agent idle, and resumes on the next activity; manual mode is a full classic wall-clock Pomodoro. Three Desktop surfaces (draggable floating widget, status-bar chip, full page), session log with today/week totals and a day streak, and a /pomodoro slash command. ### [hermes-wrm-pro](https://github.com/NousResearch/hermes-agent/pull/153744) Tools · by vedranius · open · 11-11 Guided safe workflows for using an existing WRM Pro MCP server from Hermes Agent. ### [demand-radar](https://github.com/NousResearch/hermes-agent/pull/153741) Tools · by LeonardIrimia · open · 11-11 Prove someone wants it before you build it. Searches Stack Exchange, Hacker News and GitHub issues for people asking for an idea, keeps only on-topic requests (a noise gate, not a model's opinion), and returns a GO / NARROW / KILL / NO SIGNAL verdict with the rule that fired, quotes with links, who already builds it, and how demand moved since the last check. Tools demand\_radar\_check and demand\_radar\_history, the /radar slash command and the validate-idea skill. No API key needed. ### [Hermes Arena](https://github.com/NousResearch/hermes-agent/pull/153740) Desktop · by JustinPerea · open · 11-11 A pixel-art platform fighter that runs inside Hermes Desktop. Four fighters, two stages, and three ways to play: against a CPU, with a second player on the same keyboard, or online through Find a match or a room code. Online play connects the two computers directly. ### [WHOOP Readiness](https://github.com/NousResearch/hermes-agent/pull/153738) Tools · by solmonger · open · 11-11 Your own WHOOP data in Hermes: today's recovery, HRV, resting heart rate, SpO2, skin temperature, sleep and strain against your personal 30-day baseline, history over any date range and 7-vs-30-day trends, from the official WHOOP API v2 with your own WHOOP developer app. Adds a /whoop command that answers without a model call, two Automation Blueprints (morning readiness brief, weekly review), an opt-in readiness hook and a bundled skill. Includes a clearly labelled synthetic demo mode. Native Python plugin, no dependencies. Unofficial; not affiliated with WHOOP, Inc. ### [backfill](https://github.com/NousResearch/hermes-agent/pull/153737) Memory · by wth-amna · open · 11-11 Switch memory providers without Hermes forgetting you: replays your past Hermes sessions into the memory provider you use now, with a dry run, resume, no double imports and a before/after recall check. ### [RedGreen](https://github.com/NousResearch/hermes-agent/pull/153734) Tools · by shahaman098 · open · 11-11 Prove a fix with a negative control before Hermes can call it done — re-run the test command on the current tree and on the pre-change tree with only new/changed tests overlaid. ### [Margin](https://github.com/NousResearch/hermes-agent/pull/153733) Platform · by pratikpoddar · open · 11-11 Comment @hermes on highlighted text in Google Docs, Slides or Sheets and Hermes answers in the comment thread. A gateway platform: each Drive comment thread is its own Hermes session with the document text and the highlighted text as context; follow-ups in the thread continue it, Hermes's status reply is edited in place into the answer, and replying "apply" replaces the highlighted text with the proposed wording (Docs and Slides, only when the text occurs exactly once). Watch files with /margin watch or hermes margin watch. ### [Rails](https://github.com/NousResearch/hermes-agent/pull/153732) Tools · by grimmjoww · open · 11-11 Guardrails and decision plugin: a fast local System 1 model (d1-3B Decision Guard) decides on every turn whether the agent is only looking or changing things, flags work you did not ask for, makes code work load the matching skill and run a test before 'done', and shows every decision inline as a tag. Optional d3 vision checks the screen before a computer-use task is called done and adds decide, screen\_find, screen\_check and video\_check tools. ### [Tupliq Launchpad](https://github.com/NousResearch/hermes-agent/pull/153731) Tools · by muhammadsalihu · open · 11-11 RevenueCat revenue & subscription metrics for your apps, plus git changelog drafting for Play Store releases. ### [vitam](https://github.com/NousResearch/hermes-agent/pull/153730) Voice · by taylorwtf · open · 11-11 Turn a modded PS Vita into a voice-powered Hermes handheld: hold-to-talk mic with faster-whisper STT on your own machine, an agent turn through your Hermes gateway, TTS reply spoken through the Vita speaker, and optional rear-camera vision frames. LAN-only; the bridge makes no outbound calls except to your configured gateway. ### [plugin-safehouse](https://github.com/NousResearch/hermes-agent/pull/153728) Tools · by LaithAskar · open · 11-11 Runs one local Hermes plugin inside a disposable, canary-seeded macOS profile; invokes its registered handlers and reports observed file, environment, network, process, persistence, and model-context behavior as FLAGGED, REVIEW, INCONCLUSIVE, or NO\_FLAGS\_OBSERVED. ### [Roomsmith](https://github.com/NousResearch/hermes-agent/pull/153724) Desktop · by devSebb · open · 11-11 One 3D room that you and Hermes both edit. It tells you whether things fit and what they cost. Drag furniture in a live editor inside Hermes Desktop or the dashboard, or tell Hermes what to change: one saved room with undo, and each sees what the other did. Give a budget and Hermes furnishes the room with real products, read from their pages and checked against your room and your money. Start from a photo or measurements. No accounts or keys. Network: only GETs to product pages (public addresses only, rate-capped). Never buys anything. ### [Talaria](https://github.com/NousResearch/hermes-agent/pull/153825) Automation · by mjneuroart · open · 11-11 Write, compile, sign and audit iOS Shortcuts from Hermes with no Mac: six tools turn plain-language intent into reviewable Cherri source, compile it to an unsigned .shortcut inside a bwrap sandbox, sign it through RoutineHub HubSign on explicit consent, and audit any .shortcut (signed AEA1 container or plain plist) for what it reads, writes and sends before it is installed. Bundles the shortcuts-authoring skill (file shape, verified action signatures, house patterns, and the traps that make a shortcut compile cleanly and still be broken). ### [Test Drive](https://github.com/NousResearch/hermes-agent/pull/153719) Tools · by Xuefeng-Zhu · open · 11-11 Don't guess which plugin helps. Test-drive it. When Hermes hits a capability wall (it says it can't, or a tool fails as unavailable) test-drive appends one line naming the best catalog match; /testdrive \ and test\_drive\_find show the top 3 with trust cards (tier, pin, keys needed, declared tools/hooks, disclosure); after you approve, test\_drive\_run or /testdrive run installs the candidate into a throwaway HERMES\_HOME, runs your original request once with hermes chat -q, attempts cleanup and reports a before/after with tool-call evidence when available; keep goes through Hermes's own installer; /testdrive history keeps receipts. ### [Live Translate](https://github.com/NousResearch/hermes-agent/pull/153718) Voice · by WestWaters · open · 11-11 Real-time speech-to-speech translation on your own computer: anyone speaks in any language and you hear them in yours, in their own voice, about 2-3 seconds behind. Hears ~100 languages (Qwen3-ASR + Whisper), translates between 201 (Qwen3.5-4B, a local server, a subscription, or the model Hermes already uses through ctx.llm), and speaks with the speaker's cloned voice in 23 languages and natural voices in 600+. One-way, two-way conversations and meetings/streams through a virtual audio device (Zoom, Meet, Teams, Discord, OBS, with OBS closed captions viewers can switch on); echo cancellation, background voices ignored, several speakers each with their own language and voice. Tools translate\_start/stop/status/languages, /translate, hermes translate setup\|start\|file\|serve\|languages\|devices, and an agent skill; optional desktop tray app (separate download). ### [Circuit](https://github.com/NousResearch/hermes-agent/pull/153716) Tools · by odtboun · open · 11-11 Design real PCBs by talking to your agent — a two-layer board compiler with a part library, deterministic placement, autorouting, an independent DRC, and a fabrication package (RS-274X Gerbers, Excellon drills, KiCad 8 project, STEP AP214), rendered in 2D and 3D in the desktop UI. ### [hermes-backup](https://github.com/NousResearch/hermes-agent/pull/153715) Platform · by Shaun Xenon · closed · 11-11 Encrypted restic backup and verified restore for the Hermes home — dual-target (local disk + S3), hot-database shadow copies, and a DR drill that proves restores actually come back. ### [tenure](https://github.com/NousResearch/hermes-agent/pull/153714) Tools · by otsoa-dubief · open · 11-11 Revocable trust for Hermes agents, per action type. By default every kind of action works as with Hermes alone (files, memory, subagents, known dev tools: npm, pytest, cargo, git...), and every local change is saved first: file writes and plain rm, mv, cp, sed -i or \`\> file\` on project files act, then tell you, with \`/tenure undo\` for 24 hours. The first mistake on a kind of action (an undo, a flag, a refusal of Hermes's own prompt, a file it wrote that no longer parses) puts that kind, for every model of the profile, on probation: it asks first, earns back acting-then-notifying on a run of approvals, then acting alone with random spot checks, from a rule preregistered on OSF (Wald SPRT to the middle tier, Dodge's CSP-1 series to the top, a guarantee you choose). What leaves the machine (push, publish, deploy, registry logins, cloud writes, payments), unreadable shell, and anything after the agent read outside content keep asking; one "Allow for this session" covers a session. Signed, hash-chained, replayable autonomy passport (\`hermes tenure verify\`), a per-model benchmark of your own work, a switch preview before you change model, a Desktop page and a dashboard tab at parity, five languages. Linux, macOS and Windows (on Windows, plain shell file commands keep asking instead of acting with an undo). A friction bench in the repo replays an ordinary 30-call coding session: 3 questions Hermes would not ask, against 23 when every kind of action asks first. ### [cellium-wallet](https://github.com/NousResearch/hermes-agent/pull/153712) Tools · by xjubiliance · open · 11-11 Connect Hermes to a Cellium agent for wallet operations, swaps, bridges, and managed orders. ### [RELIC Gumi](https://github.com/NousResearch/hermes-agent/pull/153711) General · by yuzushi-dev · open · 11-11 A local-first companion that gets to know you and supports self-reflection, built on published psychometric instruments (TIPI for the Big Five, ECR-RS for attachment) as a self-report baseline plus consented check-ins, with an output safety critic and Telegram or Desktop onboarding. Not a diagnostic tool; based on the paper arXiv:2611.08504. ### [Claim Receipts: what the agent said vs what it did](https://github.com/NousResearch/hermes-agent/pull/153708) Tools · by gwelby · open · 11-11 Records the file changes and commands a session's hooks observe, then gives each prose claim in the session one of four verdicts against that record: backed, unbacked, unverifiable, contradicted. Compares a claim with a capture; does not prove it true. ### [neuron-steer](https://github.com/NousResearch/hermes-agent/pull/153707) Tools · by mrfentmen · open · 11-11 Neuron steering for Hermes Agent, implementing Nous Research's own Contrastive Neuron Attribution (arXiv:2605.12290). Find the sparse MLP neurons that distinguish two behaviors with forward passes only, save them as named circuits, then scale or ablate those neurons at inference to steer a local model, plus a dose scan that maps where the behavior turns on before output quality breaks. ### [Kareverse](https://github.com/NousResearch/hermes-agent/pull/153700) Tools · by Kareverse · open · 11-11 Elder care at home in Bangalore, India, from Hermes through Kareverse's public MCP server (Streamable HTTP at https://www.kareverse.com/mcp, no sign-in): search services and prices, read a service's details, the monthly Care Plan, service areas and care guides, and send a callback request to the Kareverse team. Bundles one skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [BlindOracle](https://github.com/NousResearch/hermes-agent/pull/153696) Tools · by craigmbrown · open · 11-11 Ask whether an AI agent is safe to pay before you pay it: free trust badge and reputation lookup (5 a day, no wallet, no account), the live price list for BlindOracle research, security-audit and due-diligence services, key-free receipt verification and a proof-of-delegation skill. Portable Agent Plugins v1 package (mcp.json + skills/). ### [Orphic Eyes](https://github.com/NousResearch/hermes-agent/pull/153695) Tools · by Ilham-Musyafa · open · 11-11 Pixel-only screenshot review for Hermes: capture URL pairs, inspect changed zones, prepare coding-agent requests and run bounded follow-up review rounds. ### [Offgrid](https://github.com/NousResearch/hermes-agent/pull/153693) Platform · by emmaGH1 · open · 11-11 Chat with your Hermes agent by plain SMS, with no mobile data, through Africa's Talking, plus a USSD menu as a fallback for basic phones. Texts become normal Hermes messages on the offgrid platform, one conversation per phone number, and replies come back as a single SMS (Markdown and emoji stripped, capped at five billed segments). A safe subset of gateway commands works by text: NEW, SESSIONS, RESUME \, MODEL, STATUS, HELP and /retry, /stop, /title; every other slash command is refused by default. ### [Design Discovery](https://github.com/NousResearch/hermes-agent/pull/153692) Desktop · by adip1303 · open · 11-11 Turns a vague product idea into a DESIGN.md a coding agent can build from: a few plain questions, three visual directions, changes in your own words. Loads fonts from Google Fonts. ### [macOS Keychain](https://github.com/NousResearch/hermes-agent/pull/153690) Other · by shoaibyazdani · open · 11-11 Keep Hermes provider credentials in the macOS login keychain instead of plaintext .env. Registers a SecretSource (secrets.keychain), so startup resolution runs through Hermes's own orchestrator, after ~/.hermes/.env loads and before credentials are read, the same path the bundled Bitwarden and 1Password sources use. Bindings are explicit (secrets.keychain.env: ENV\_VAR: keychain-account) or discovered from an index the plugin keeps inside the keychain, so \`hermes keychain put NAME\` is the whole setup. Adds keychain\_status (vault-backed vs still-plaintext inventory; names, lengths and outcomes only), keychain\_migrate (store, read back, compare, then replace the .env line with a comment mentioning the name; 0600 backup first) and keychain\_audit (fetch/migrate trail), plus a \`hermes keychain put\|ls\|rm\|test\|migrate\|status\|audit\` CLI. ### [Locus Codegraph](https://github.com/NousResearch/hermes-agent/pull/153687) Tools · by atikur · open · 11-11 Rehearse a Python refactor before making it, and catch missed callers when you do. @whatif:\(\) or the locus\_rehearse tool binds every statically proven caller against a proposed signature before any file is edited and lists which call sites would break and why. After an edit, the same check appends caller locations, binding errors and, for a renamed keyword, a caller patch to review to the editing tool's result. Covers functions, class constructors, static and class methods, methods on locally constructed instances, and calls inside the edited file, through direct and module-qualified imports, aliases and static re-exports. Receivers whose type is not in view (self, parameters, attributes), decorated callables and dynamic bindings are not checked; other supported languages receive structural indexing only. One local index also powers five tools, @sym:/@impact:/@hotspots: references, /locus commands, and Web Dashboard and Desktop caller inboxes. Suggested tests and impact scores are static heuristics, not execution coverage; no alert or predicted break is not proof of safety. ### [On Air](https://github.com/NousResearch/hermes-agent/pull/153686) Tools · by perfect7613 · open · 11-11 Hermes inside your OBS. Control scenes, sources, recording and streaming by asking; put cards and chapter titles on stream; design animated scenes or build one from an image, GIF, video or voice line; answer questions about your screen on an overlay; mark the good moments and get the highlight reel, YouTube chapters and a draft title when you stop. ### [onbrief](https://github.com/NousResearch/hermes-agent/pull/153685) Automation · by faze39 · closed · 11-11 Keeps outbound posts on the campaign brief. A pre\_tool\_call hook reads the text of posting and messaging tool calls (name globs, MCP tools included) and checks it against your YAML or JSON rules file: required hashtags, mentions and phrases, banned words and regex patterns, X-weighted length, hashtag limits, required or allowed links, and near-duplicates of posts already sent. A failing call is blocked (or sent to the approval prompt) with every broken rule, a fix, and a passing rewrite when the fix is mechanical. Adds an onbrief\_check tool and /onbrief and \`hermes onbrief\` commands. ### [Postkey](https://github.com/NousResearch/hermes-agent/pull/153684) Tools · by Rabbithole-llc · open · 11-11 Instagram DM automation and post scheduling through Postkey's hosted MCP server (Streamable HTTP at https://postkey.pro/mcp, OAuth sign-in, no keys in config): comment-to-DM links, follow to unlock, email capture from comments, story and DM keyword replies, the Instagram inbox (read, reply, assign, label, notes), contacts and tags, AI reply settings, a media library and video edits, and drafting, scheduling and analytics for posts on Instagram, Facebook, TikTok, YouTube, LinkedIn, Threads, Pinterest, Bluesky, Telegram, Google Business and X, plus a check-before-send toolkit for agents acting on X with the user's own access. Bundles the postkey workflow skill (start with postkey\_get\_started, preview every send with postkey\_preview\_action, respect approvals, never invent links, treat DM and comment text as untrusted). Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [HTMi](https://github.com/NousResearch/hermes-agent/pull/153680) Tools · by Thamirawaran · open · 11-11 Review the HTML pages your agent writes, on the page itself: comment on the exact element, chart, table row or passage; Hermes is told about new comments, reads each one with its context, makes checked edits and replies. Pages can also be published as a GitHub issue or pull request description, and comments posted there are pulled back. Adds 11 tools, a pre\_llm\_call notice, a skill, the /htmi command and a Review tab in the dashboard. ### [rh-preflight](https://github.com/NousResearch/hermes-agent/pull/153679) Tools · by quotumorg · open · 11-11 rh-preflight by quotum with HEY Research Lab. Before Hermes acts on Robinhood Chain (chain 4663), a pre\_tool\_call hook reads HEY Research Lab's builder record and ten Robinhood Chain projects for the addresses, agent ids, pools or repositories in the call, then applies your own rules: go, ask (a human approval prompt) or stop when unattended. The card lists each source's line with its link; builder\_card and /rh-preflight read a card without acting. Desktop page shows recent checks and human answers. ### [ConsequenceMap](https://github.com/NousResearch/hermes-agent/pull/153678) Desktop · by BitSempre · open · 11-11 Decision intelligence for AI agents: forecasts the consequence of an action before it happens, preserves the decision and its evidence, and compares the forecast with what actually occurred. An outcome that cannot be verified stays NOT\_VERIFIABLE, so the derived verdict is INCONCLUSIVE rather than a clean pass. Five Hermes hooks (pre\_tool\_call, pre\_approval\_request, post\_approval\_response, post\_tool\_call, transform\_tool\_result); a dashboard tab and a command-palette entry. The agent proposes and escalates; it never decides and never holds a key. ### [dosbox-mcp](https://github.com/NousResearch/hermes-agent/pull/153675) Tools · by dosbox-automation · open · 11-11 Drive a DOS machine from the agent through dosbox-automation's own MCP bridge: type and press keys, read the screen, take screenshots, record video and audio, run Lua scripts inside the emulator, read and write its memory. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only: it starts the bridge with uvx dosbox-mcp==0.2.4 from PyPI and bundles one usage skill. ### [Talaria Bench](https://github.com/NousResearch/hermes-agent/pull/153671) Tools · by devorun · open · 11-11 A physics test bench for DIY servo robots: build your robot from a template and a few measurements, see it in 3D next to its photo, run a one-click MuJoCo arena (flat, ramps, bars, pushes, slippery floor) across randomized worlds with a report card that says why it fails, compare design variants or watch a changed version side by side in the Arena, and gate robot\_play / robot\_pose with an exact-motion preflight in simulation before a move may play on the real robot (ESP32 + PCA9685 firmware included, or a virtual robot). CPU only, no GPU, no CAD. ### [Corkboard](https://github.com/NousResearch/hermes-agent/pull/153669) Tools · by elijahboston · open · 11-11 One shared board, across every profile, of things only you can do. Agents pin a task, a yes/no question, a choice of up to four (their recommendation starred) or a text question instead of burying it in chat; you answer from a Desktop side pane, the full Corkboard page, \`hermes corkboard\` or \`/corkboard\`. Blocked kanban cards are pinned automatically and your answer goes back to the card as an "Owner decision" comment, which resumes the worker. Pins can clear themselves when a check passes (nameservers changed, a kanban card unblocked). Text pins never ask for secrets; credential needs are always tasks. ### [IV Studio](https://github.com/NousResearch/hermes-agent/pull/153664) Tools · by hollisorg · open · 11-11 Free, local video and pictures on your own NVIDIA GPU: makes IV Studio (MiniMax H3 + ComfyUI) a video\_generate and image\_generate backend, with sound and speech, trained characters by name, first-frame and reference pictures, and finishes up to about 2752x1536, plus iv\_studio\_status and iv\_studio\_result (long renders hand back a job id and carry on) and a bundled skill. ### [openrouter-balance](https://github.com/NousResearch/hermes-agent/pull/153663) Desktop · by maeinomatic · open · 11-11 Live OpenRouter credit balance in the Hermes desktop status bar — "OR $12.34 left", with total/used/remaining on hover, amber below 15% remaining, click to refetch. Polls every 5 minutes idle and refetches ~8 s after each chat turn. The plugin's Python half resolves the profile's OPENROUTER\_API\_KEY through agent.secret\_scope (profile-scope aware, safe under multiplexing; the key never leaves the backend) and proxies the free openrouter.ai /api/v1/credits endpoint, mounted at /api/plugins/openrouter-balance/\* behind the dashboard auth gate. The desktop half registers a statusBar.right chip via @hermes/plugin-sdk. The Python register() is a no-op — no tools, hooks or middleware. ### [repager](https://github.com/NousResearch/hermes-agent/pull/153660) Web · by clawpawtech · open · 11-11 Evidence-bound hero design lab with approved copy, local layout exploration, revision-bound browser checks and portable review handoffs. ### [SiteForm](https://github.com/NousResearch/hermes-agent/pull/153658) Desktop · by duclucky · open · 11-11 Review planning PDFs, compare evidence-linked 3D options in Hermes chat, and edit the same revision in Studio. ### [Sovereign Finance](https://github.com/NousResearch/hermes-agent/pull/153654) Tools · by gabonaut · open · 11-11 A self-custodial Smart Agentic Wallet for your agent on BNB Chain and Ethereum, with a Desktop pane. Your passkey is the admin, and the agent's session key is limited onchain to the contracts, functions and daily caps you approve. Powered by Altana. ### [GHL Command](https://github.com/NousResearch/hermes-agent/pull/153653) Tools · by PXMichael · open · 11-11 GoHighLevel lives inside Hermes. Create contacts, move deals, send email and SMS, trigger workflows from chat. Live in 90 seconds, no browser clicks. ### [Logmine](https://github.com/NousResearch/hermes-agent/pull/153651) Tools · by rodricksz4h5 · open · 11-11 Exact, local intelligence over large text log files. log\_summarize mines the file into recurring message templates (masking numbers, IPs, UUIDs, hashes, quoted strings and timestamps so near-identical lines collapse into one ranked pattern with an exact count), plus a level histogram and the detected time span; log\_grep is regex search with line numbers and grep-style context; log\_timeline buckets events per minute/hour/day to find spikes. Streamed, so a log too big to fit in the model's context still works. Standard library only: no network, no credentials, no background processes, no self-update. ### [hermes-continuum](https://github.com/NousResearch/hermes-agent/pull/153650) Desktop · by mohamad-shafeez · open · 11-11 An integrated nine-system plugin for Hermes Agent that provides task continuity, review-first episodic memory, environment reality checks, evidence tracking, goal integrity, trust-aware tool policies, impact previews, retry-loop protection, and verified workflow distillation. Its systems share an append-only SQLite Event Spine and a Continuum Console for Hermes Desktop. Includes 12 tools, 11 lifecycle hooks, checkpoint export, and a review-first approach to generated artifacts. Live verification and test coverage are documented in the repository; not every capability has been verified in a live host workflow. ### [Sightline](https://github.com/NousResearch/hermes-agent/pull/153649) Tools · by OutstandingVick · open · 11-11 Give your Hermes agent a Solana wallet without giving it your wallet. Intent-level SOL and SPL tools, including Solana Pay payment links; every send is simulated, explained in plain English and checked by a firewall against human-issued, time-boxed, revocable grants (caps per day, week or month) before the plugin signs it. Cron jobs and bots can pay inside a grant unattended. The model never sees the key, has no raw-transaction tool and cannot create or widen grants. ALLOW / APPROVE / BLOCK with a hash-chained audit log and a Desktop pane. Devnet by default. ### [Riah Swipe](https://github.com/NousResearch/hermes-agent/pull/153647) Tools · by Riah Studio · open · 11-11 Swipe through what your agent made. Hermes puts a batch of options (images, video, audio, names, taglines) into a swipe deck; you keep, cut, or send one back with a note, and Hermes remakes the ones you sent back. On Telegram the deck appears in the chat with Keep, Cut, Redo and Undo buttons, and you type a note on any card to say why; elsewhere you get a sign-in board link for desktop or phone. Runs a small local HTTP server inside Hermes and writes only to its plugin-data folder; no extra dependencies, telemetry or accounts. Full disclosure in the README. ### [Andreed](https://github.com/NousResearch/hermes-agent/pull/153646) Tools · by Agozie180 · open · 11-11 Deterministic Android device-care tools for Hermes: audits storage and bounded total/metered data usage, reports main-process state, restricts metered background data, force-stops selected third-party apps, and requests safe device-wide cache trimming. ### [Hermes Chrome](https://github.com/NousResearch/hermes-agent/pull/153641) Web · by Dhana · open · 11-11 Let Hermes control your paired Chrome with a visible cursor, isolated tab group, human handoff, side-panel chat, and recorded workflows. ### [Patient Research](https://github.com/NousResearch/hermes-agent/pull/153640) Tools · by solo386 · closed · 11-11 Deterministic clinical trial, medical literature, and early treatment signal research watcher for patients and caregivers (English & Russian). Python 3.11+ standard library only (MIT license, zero pip dependencies): deterministic evidence grading, time-since-onset eligibility window screening with zero false rejections, exact eligibility quote verification, code-rendered Markdown reports, printable 1-page doctor/coordinator handoff packets (.md + .html), and a 0-token /pr slash command. Not medical advice and does not replace a clinician. ### [Topoi](https://github.com/NousResearch/hermes-agent/pull/153638) Desktop · by AlaisterLeung · open · 11-11 Group your Hermes Agent chats into topics, in Hermes Desktop and in the Dashboard. A sidebar lists the topics active in the past week with a status dot per thread (working, waiting, unread, idle), a new chat files under your home topic, threads move between topics, and a topic's threads are searchable by message content. Cron job results arrive as topic threads you can reply to, and the reply continues that run's session. The gateway owns one topic registry and both clients read and write it over the plugin's REST API, so an edit in the Dashboard reaches the desktop app. ### [incognito-private-mode](https://github.com/NousResearch/hermes-agent/pull/153637) Other · by Tranquil-Flow · open · 11-11 Private mode, honestly: provider privacy flags, memory blocks stripped from main-chat requests (provenance-checked against the last 6 observed file versions, fail-closed outside that window), tool vetoes on unclassified tools, honest coverage labels, and consent-gated fenced history cleanup on close (exact-id + transcript + closed-state fences, archived chats excluded). Content-free outbound audit log (endpoints + verdicts only, local 0600). No telemetry; the plugin itself makes no network calls. Disclosed limits: auxiliary-call fan-out not stripped, MoA advisors receive the full unstripped request before private mode blocks the aggregation, provider-side retention guarantees impossible. ### [Hermes Artifacts](https://github.com/NousResearch/hermes-agent/pull/153635) Desktop · by MrSakal · open · 11-11 Claude-style artifacts for Hermes: docs, sheets, whiteboards, slide decks, dashboards, designs, 3D models, films and design systems that Hermes makes and you edit in place, versioned, in a gallery and sandboxed viewer inside Hermes Desktop. Sheets calculate, dashboards re-run stored MCP queries, and everything exports locally (docx, xlsx, pptx, pdf, png, svg, mp4, glb). ### [Baro](https://github.com/NousResearch/hermes-agent/pull/153630) Tools · by JigJoy · open · 11-11 Hand a software goal to Baro and watch it run as a live, dependency-aware kanban: baro\_plan returns the plan as stories with dependency edges, baro\_run starts a run, baro\_status shows each story as blocked, ready, running, done or failed with what it waits on and what it blocks, and baro\_cancel stops the run and its process tree. ### [Uniichat Recall](https://github.com/NousResearch/hermes-agent/pull/153629) Memory · by DevvGwardo · open · 11-11 Hermes memory provider over an append-only UniiChat log and a derived binary summary tree: ranks every past turn, returns the exact records with session IDs and character offsets, and abstains instead of guessing when nothing in the log answers. ### [zerofactory](https://github.com/NousResearch/hermes-agent/pull/153625) Automation · by hotcode-dev · open · 11-11 24/7 AI multi-agent software engineering factory and durable SQLite Kanban board. Flexible triage with Grill Me clarification and EGM sequential spec audit, isolated Git worktrees, deterministic precommit pipeline, thematic continuous review, autonomous self-healing recovery, and Mem0 vector memory. ### [rosegold](https://github.com/NousResearch/hermes-agent/pull/153624) Tools · by rosegold · open · 11-11 Get offers and rewards matched to what you buy on Amazon, including discount codes, promo codes and influencer codes. Hermes shares the order details in your Amazon order confirmations with rosegold, then looks up the offers, creator codes and spending rosegold has for you. rosegold reads order confirmations from Amazon, limited to order number, email id, date, total, and each item's title, quantity and price; it does not read your address, payment details or any other emails. Portable Agent Plugins v1 package (mcp.json only); hosted MCP server with three tools, sign-in by emailed code over OAuth, no API key; one command after install (see known issues). ### [Sharelet](https://github.com/NousResearch/hermes-agent/pull/153623) Tools · by TANGBUDU · open · 11-11 Share a Hermes tool as a link anyone can use in a browser. Pick a tool on the local creator page, create the link and send it: the guest needs no Hermes, account or API key, and the tool keeps running on your computer. Each link has a run limit and an expiry and can be closed at any time; guests can change only the inputs you leave open, and a tool that is not marked read-only needs a separate confirmation for every share. Links stay the same across restarts and tunnel reconnects. Works with native Hermes plugin tools and MCP servers you allow. ### [Hermes Radar](https://github.com/NousResearch/hermes-agent/pull/153634) Automation · by Naroh091 · open · 11-11 Proactive assistant: follows your mail, calendar, GitHub, scheduled jobs and your own chats with the agent as long-lived tasks (trips, purchases, work items, commitments), does the legwork itself and writes to you first, on the gateway platform you choose, only when something is yours to decide. Sending email always asks; paying, booking, deploying and merging never run on their own. ### [Memory Ferry](https://github.com/NousResearch/hermes-agent/pull/153617) Memory · by gnanam1990 · open · 11-11 Migrates built-in Hermes MEMORY.md and USER.md records into a profile-local SQLite memory provider with an integrity-checked dry run, idempotent recovery, deterministic retrieval-parity evidence and migration-scoped rollback. ### [art98 (Windows 98)](https://github.com/NousResearch/hermes-agent/pull/153615) Tools · by StanLukuvka · open · 11-11 Drive a real Windows 98 SE VM through art98 (screenshots, pointer, keys, windows, Win32 calls) and render scene scripts to video, with a dashboard tab that has a live view. ### [Reliquary](https://github.com/NousResearch/hermes-agent/pull/153614) Memory · by Lokee86 · open · 11-11 Local, inspectable long-term memory with source provenance, conversation capture, General/project REL routing, a read-only PHY identity layer, and an interactive relationship-driven 3D graph in Hermes. ### [Cutroom](https://github.com/NousResearch/hermes-agent/pull/153612) Tools · by KaanDRoid · open · 11-11 Local-first video editing studio: word-timed transcript cuts and filler review, silence removal, face/gaze/mouth review, person segmentation for graphics behind the subject, three source-quoted hook options with a storyboard, preview renders, human-approved final export, cover and automated quality checks. Ships a Hermes Desktop Studio page, agent tools for projects and background jobs (no approval tool: only a person can approve a final export) and a CLI. Requires FFmpeg/ffprobe on PATH; speech, vision and segmentation need optional Python packages and local model files. ### [Innerscape](https://github.com/NousResearch/hermes-agent/pull/153609) Tools · by KyaniteLabs · open · 11-11 Gentle check-ins, journaling and an Unstuck Coach with your own Innerscape server: save a mood check-in in your words, save a journal entry or park an inbox item only after you say yes to the exact text, and look back on today. When you're frozen or overloaded, the bundled unstuck skill names what's in the way, gives one next move and holds the rest. The skills never score, label or diagnose a mood. Adds four tools, /innerscape, two skills, a Desktop page and a dashboard tab. ### [Chess Coach](https://github.com/NousResearch/hermes-agent/pull/153608) Desktop · by oppai · open · 11-11 Practice mate in one to four or play Hermes with piece odds on an interactive Desktop chessboard. ### [rea-video-hermes](https://github.com/NousResearch/hermes-agent/pull/153607) Tools · by wusanjay · open · 11-11 Reverse-engineer any video shot by shot: production-method classification per shot (live action / AI video / 3D / screen capture / code-template render), caption and sticker timelines, voice vs music stems, editing rhythm, and a concrete replication plan. Powered by JuneYaooo/rea-video (Apache-2.0, pinned commit d904b1e8). The six stdio-MCP tools expose the full pipeline (incl. mux\_tracks for dual-track video downloads); for shots the pipeline flags as low-confidence, the agent reviews the tiled frames with its own vision and merges the verdict back via submit\_review — no external VLM key required. Local runtime deps: ffmpeg and the pinned rea-video package (installed from the repo's venv). ### [Habits](https://github.com/NousResearch/hermes-agent/pull/153606) Automation · by emiwltz · open · 11-11 Habits gives Hermes reflexes: the tool calls and small decisions it repeats, grouped into one file of steps that runs with no model turn between the steps (the advanced llm step is the one exception); decisions go to a small typed model (Jev). "Is Hermes OK?": 41 s when Hermes works it out, 2.7 s when the reflex answers with its six checks (medians of real chats, 10 November; the fast path needs your yes and a Jev key). A reflex has four kinds of step (do, check, ask, use) and ends done, or handed back with what is left. The dream re-reads the history Hermes already keeps, with no model call, and proposes reflexes for what comes back. A reflex acts on its own only after your yes, with two exceptions named below. Try without a key: hermes habits run hermes-health. ### [Treasury Cash Forecast](https://github.com/NousResearch/hermes-agent/pull/153605) Tools · by Prabhusekaran · open · 11-11 A 12-month monthly direct cash-flow forecast for corporate treasury. Fill one template (Excel or CSV) with your past cash-flow statement, the future cash flows you already know and your what-if percentages. Recurring lines are projected by a statistical model (ETS by default; SARIMA and four simple baselines selectable), known lines come from your schedule, and a known amount can replace the model for single months. The base case and every scenario run together and come back as one Excel workbook: a summary, a cash-flow statement per scenario (12 actual months, then 12 forecast months) and a sheet showing where every figure comes from, reconciled every month to the cent: opening cash + net cash flow = closing cash. A fourth tool runs the same data through several models to show how much the answer depends on the method. All figures come from a tested engine (95 tests); the language model picks the tool and explains the result. ### [AutoRAG](https://github.com/NousResearch/hermes-agent/pull/153601) Tools · by Marker-Inc-Korea · open · 11-11 Search your own files and configured chat/mail archives from Hermes with AutoRAG Lite. Local hybrid retrieval, file-name search, cited evidence and duplicate detection without editing original documents. Supports PDF, DOCX, PPTX, XLSX, Markdown and Korean HWP/HWPX; existing integrations include Slack, Discord, Telegram, WhatsApp, KakaoTalk, email, Notion and Obsidian. Bundles four skills and a Hermes Desktop Finder pane. Requires Node.js 24+; runs pinned @autorag/librarian@2.13.12 through npx stdio. The Desktop backend installs that same pinned package once in plugin data, with npm lifecycle scripts disabled. First refresh can provision minsync and jikji and download local embedding models (roughly 1.5 GB). Reads only configured folders and archives; writes its own index/report state. User-requested automation also writes rule/scripts and creates Hermes cron jobs. No telemetry, self-update or stored-login discovery. Search snippets read by Hermes reach its selected model provider; configured remote connectors or embedders use their selected services. ### [birdie](https://github.com/NousResearch/hermes-agent/pull/153598) Tools · by koshikraj · open · 11-11 Lets each Hermes conversation message the other AI agents on this machine, such as Claude Code, Codex and Grok, through birdie, a local open-source message bus, and, if you allow it, be woken by their messages. Each conversation gets its own address, so an answer comes back to the conversation that asked. Five tools: list agents, send, check the inbox, wait for a reply, read a thread. ### [Sentinel Vision](https://github.com/NousResearch/hermes-agent/pull/153596) Automation · by StarAtNyte · open · 11-11 Tell Hermes what to watch on a camera, screen or video, and Sentinel detects events, stores snapshots and injects alerts into your chat. A local change detector and configurable call caps reduce vision-model usage; rules, event history and feedback make the watcher useful over time. Frames and rule text are sent to your configured model; see the README for setup, privacy and limitations. ### [Tinfoil Container](https://github.com/NousResearch/hermes-agent/pull/153595) Tools · by counterposition · open · 11-11 A tinfoil terminal backend: Hermes' terminal, file and code tools run inside your own Tinfoil Container, an attested enclave, reached over the Tinfoil CLI's attested SSH tunnel. Select it with terminal.backend: tinfoil and terminal.tinfoil\_container. Needs a Tinfoil organization with Containers enabled; the public template https://github.com/counterposition/hermes-tinfoil-workspace sets up a suitable workspace. Dangerous-command approval stays on, and the plugin never creates, starts, stops or bills containers. ### [Tinfoil](https://github.com/NousResearch/hermes-agent/pull/153594) Models · by counterposition · open · 11-11 Tinfoil confidential inference as a Hermes model provider. Open-weight models from Tinfoil's live catalog (Kimi, GLM, gpt-oss and others) run in attested hardware enclaves that neither the cloud host nor Tinfoil can read. Hermes reaches them only through Tinfoil's local tinfoil-proxy, which verifies the enclave's attestation and pins its TLS key before forwarding anything; there is no fallback to an unverified connection. The proxy starts on demand. ### [TallySeal](https://github.com/NousResearch/hermes-agent/pull/153588) Tools · by Venkat-RJ · open · 11-11 Checks each group in a CSV report against a pinned source and seals the exact report bytes that pass. ### [Hermeneus](https://github.com/NousResearch/hermes-agent/pull/153585) Voice · by Bihruze · open · 11-11 Live translation for Hermes. Hermeneus captures the audio playing on your Mac, so it works with any app that plays sound. Examples are YouTube, Twitch and X Spaces in the browser, Discord, meeting apps and podcast apps. Verified on real audio: YouTube live in Google Chrome. Captures the audio playing on your Mac (one app, or all system audio except Hermeneus) through a Core Audio process tap on macOS 14.2+, transcribes it with ElevenLabs Scribe v3 Live, translates each line with your own Hermes model and shows live subtitles in a Desktop pane, a floating overlay above other apps, the terminal (hermes hermeneus run), /translate, or an opt-in 127.0.0.1 page for OBS; spoken translation with ElevenLabs Flash v2.5 is opt-in, and language, mode, voice, output and source change while a session runs. Listening covers 100 languages and speaking covers 32. ### [Homelab Watch](https://github.com/NousResearch/hermes-agent/pull/153584) Tools · by lpsmurf · open · 11-11 Your homelab, your agents and your agent's bill in one Desktop status chip: read-only checks of your router, servers, websites, a Bitaxe miner and your cron jobs with root-cause rollup (a router outage is one alert, not ten); token-free alerts through a no\_agent cron job; per-day LLM token counts and API-equivalent savings (subscription, cheaper and free models; unpriced models never estimated); and the jev\_route delegation backend, shadow by default, that sends simple subagent tasks to a cheaper model when a judge clears them. ### [lookahead](https://github.com/NousResearch/hermes-agent/pull/153583) Automation · by ericklindberg · closed · 11-11 Conversation-captured commitments with grounded preparation briefs and one post-event follow-up. ### [Hermes UI](https://github.com/NousResearch/hermes-agent/pull/153582) Desktop · by Ilyassbm · open · 11-11 A bespoke chat workspace for Hermes Desktop and the Webapp: it takes over the app's home route and replaces it with a three-column page — a rail of bots (one row per profile, each with its own conversations), the conversation itself with folded reasoning and tool runs, and a right panel for files, previews and session detail. Bots can answer with a structured block that the thread draws as rich components (status cards, tables, checklists, approval prompts) instead of prose; slash commands, per-bot models and a settings page come with it, and media blocks render images, audio and video inline. ### [Agent Templates](https://github.com/NousResearch/hermes-agent/pull/153581) Desktop · by Ilyassbm · open · 11-11 Share a whole Hermes agent as one .htemplate bundle and install it from a single HTTPS link: identity (SOUL.md), config with accounts and endpoints stripped, skills, routines, MCP server definitions, memories only when the author opts in, and plugin ids. A recipe, not a clone — credentials, sessions, conversation history, logs, caches and workspace never travel. Publish puts the bundle in a GitHub repository and returns one permanent link (re-publishing updates the same URL); Install previews a link, a path or a dropped file and writes nothing before you confirm, ships routines paused, enables a plugin only if it is already installed, and reinstalling the same link offers "Update v1.0.0 → v1.1.0" while keeping memories, sessions and .env. ### [ETA](https://github.com/NousResearch/hermes-agent/pull/153580) Automation · by FrederickSonsAI · open · 11-11 Honest finish-time ranges for agent work from real run history: kanban boards and cron jobs read live, delegated subagent runs recorded automatically, 'no honest ETA' when work keeps failing, 'clock starts when...' when a step waits on a person, and labelled rough guesses when history is thin. ### [AIR Agent KYA](https://github.com/NousResearch/hermes-agent/pull/153579) Tools · by mocaNetwork · open · 11-11 Know-Your-Agent with AIR (Moca Network): you delegate your AIR identity to Hermes once in the AIR Console (\`hermes air login\`, device code, P-256 agent key, credentials.verify scope, no wallet signing); the kya\_visit tool then enters KYA stores that publish /.well-known/air-kya, gets a one-time SD-JWT presentation from AIR that discloses only the claims the store requests (e.g. membership tier), and returns the catalog that credential unlocks; agents that aren't delegated get the store's public catalog. ### [seatbelt](https://github.com/NousResearch/hermes-agent/pull/153577) Tools · by deeworksonline-debug · open · 11-11 Seatbelt enforces your trading limits in front of every trading-looking tool call (perps/futures, prediction markets, NFTs, swaps): allow, block, ask you, or clamp leverage. Includes paper mode (simulated fills), /seatbelt halt, a guided /seatbelt setup, an audit log, tamper protection and YAML venue adapters (Polymarket via hermes-pmxt included). ### [liv](https://github.com/NousResearch/hermes-agent/pull/153574) Memory · by demos-ra · open · 11-11 The first real stateful memory for AI agents: a knowledge graph above the raw transcript, every word kept as plain text tables, local and yours. ### [PRSN ID](https://github.com/NousResearch/hermes-agent/pull/153570) Tools · by PRSN EX Inc. · open · 11-11 Gives a Hermes agent its PRSN ID, read live from PRSN Connect, the desktop app that keeps the agent's signing key in the computer's secure hardware. The agent knows its own PRSN ID at each session and can seal email and files with its PRSN's key. Hermes asks the human before each Seal with its own approval prompt. It also asks the human before a command or file action of Hermes's own that names PRSN Connect or its files. /prsn shows the state, and Hermes Desktop gets a status bar item. Disclosure. The plugin reads ~/.prsn-id/gateway.json, which PRSN Connect writes and which holds no token. It starts prsn-gateway-mcp, a program PRSN Connect installed, and talks to it over standard input and output on this computer. PRSN Connect makes the calls to api.prsnid.ai, and a Seal's link opens verify.prsnid.ai. It writes its last state to status.json in its own data folder, and, only with the server\_log setting on, PRSN Connect's server log. After a file Seal it opens the Seal's page at verify.prsnid.ai in the default browser, and no other address. It makes no network call of its own, stores no credential and sends no telemetry. ### [Costory FinOps Agent](https://github.com/NousResearch/hermes-agent/pull/153569) Tools · by Costory · open · 11-11 FinOps harness MCP for platform teams. Every cloud, data and LLM bill (AWS, Google Cloud, Azure, Oracle Cloud, Snowflake, Databricks, OpenAI, Anthropic and more) normalized in BigQuery and allocated to your teams. Read-only cloud access. Starter from $300/month, 30-day free trial. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/) connecting Hermes to Costory's hosted MCP at https://app-api.costory.io/mcp (Streamable HTTP, OAuth). Bundles seven FinOps skills: query, cost-change-investigation, virtual-dimensions, dashboards, reports, recipes and bigquery. ### [Ship Reel](https://github.com/NousResearch/hermes-agent/pull/153567) Tools · by AshThunder · open · 11-11 Turns a URL and a one-line brief into a finished product demo reel. The agent plans a shot list and dry-runs it against the live page, records it in headless Chromium, edits it with auto-zoom, a drawn cursor, captions, title and end cards, optional local voice-over and a music bed into 16:9 and 9:16 MP4s, verifies the result (duration, resolution, black or frozen frames, loudness, captions) and fixes what fails, then drafts the post text. It never posts. ### [Multi-Telegram](https://github.com/NousResearch/hermes-agent/pull/153562) Platform · by sergio-dev1-py · open · 11-11 Run an arbitrary number of Telegram bot identities through one Hermes profile and agent. Each bot has its own token reference, display identity, allowlists, role prompt, welcome message and session policy; replies always leave through the bot that received the triggering message. Optional shared-session groups continue one stored Hermes transcript for the same external chat/user/topic. ### [Fudge for Shopify](https://github.com/NousResearch/hermes-agent/pull/153561) Tools · by Fudge · open · 11-11 Run a Shopify store with AI through Fudge's hosted MCP server (Streamable HTTP, OAuth 2.1 with dynamic client registration and PKCE): store briefs, sales, search and analytics measurement, SEO, AI-search and catalog checks, keyword research, and storefront changes prepared as private drafts that the merchant publishes in Fudge, with six skills bundled. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/). ### [Rocket.Chat Team Gateway](https://github.com/NousResearch/hermes-agent/pull/153557) Platform · by csabatatrai-ulx · open · 11-11 Rocket.Chat gateway adapter for Hermes Agent, built for shared team rooms. Per-person sessions with room context since the asker's previous mention, a default-deny guard for Hermes commands, approval buttons, restart catch-up and duplicate suppression that survive a restart. Supports Rocket.Chat 8.7 - 9.x. ### [Skill Stats](https://github.com/NousResearch/hermes-agent/pull/153556) Tools · by minhng92 · open · 11-11 Skill Stats shows which skills you actually use in Hermes Agent, when, and how much. Type /skill-stats in any chat to see that session's skills ranked by loads with a % share, or run \`hermes skill-stats\` in the terminal for all-time totals; switch the window from 1 day to 90, page through the rows, group them by category, and answer questions like "which skills did I lean on this week?" or "is this skill still worth keeping?" Point the command at another chat by title or short id (for example -s stageshow-bentora) to compare sessions, or add \`global\` from a chat to see everything combined. In \`hermes --tui\` the same numbers open as an interactive modal: browse skills with the arrow keys, open a skill's chart, flip between bars and a cumulative line, and set the period with keys 1 to 6. The web Dashboard gains a Skill Stats tab where you pin the skills you monitor so they stay on top of the table, page 10 to 100 rows, and drill into per-skill charts. Nothing to configure: install it and the counting starts on the next session. ### [Model Doctor](https://github.com/NousResearch/hermes-agent/pull/153555) Models · by ChainShieldSn · open · 11-11 Proactive LLM health monitor: observes every provider API request Hermes makes via the pre\_api\_request / post\_api\_request / api\_request\_error hooks, records per-model uptime, latency and failure verdicts — AUTH (HTTP 401/403, e.g. a truncated or expired API key), RATE (429), UPSTREAM (5xx), EMPTY (200 with zero content, stalled model or max\_tokens too small), SLOW — and exposes model\_doctor\_status / model\_doctor\_refresh so the agent or the operator can ask which models are sick and why. This is the diagnostic layer round-robin routing lacks: rotation moves you off a dead model, Model Doctor tells you it died, why, and which alternatives are healthy. ### [Flota](https://github.com/NousResearch/hermes-agent/pull/153549) Automation · by Francisco Bastidas · open · 11-11 Run delegate\_task subagents on your own machines over SSH (delegation.backend: flota). Flota wakes the chosen machine (Wake-on-LAN, a relay host on its LAN, any command, or the machine's own wake timer), falls back to another machine if it does not come up, and suspends it after its last subagent, never a busy one (the model cannot force it). \`hermes flota setup user@host\` detects OS, MAC, broadcast and how the machine can sleep; \`doctor\` prints fixes; \`uninstall\` undoes it; fleet pairing with two codes (public keys only); a fleet map page for Hermes Desktop and the web dashboard. Python standard library only. ### [tmux-live](https://github.com/NousResearch/hermes-agent/pull/153545) Tools · by maksymhs · open · 11-11 Live tmux tab for the Hermes web dashboard: pick a pane, watch it in ANSI colour, type into it when needed, and terminate the process, pane or session. ### [Hermes Arcade](https://github.com/NousResearch/hermes-agent/pull/153544) General · by Takumixbt · open · 11-11 Three terminal games inside the terminal Hermes already runs in: chess against the model Hermes is using, a side-view kart racer, and Chrome's dino wearing the Hermes cap. ### [mood-engine](https://github.com/NousResearch/hermes-agent/pull/153542) General · by NiceStackApple · open · 11-11 Dynamic 4-axis mood engine (energy, warmth, patience, focus) with async post-turn evaluation, passive decay, and zero-token neutral baseline. ### [Sidecar](https://github.com/NousResearch/hermes-agent/pull/153541) Tools · by rkfshakti · open · 11-11 Remote brain, local hands: a paired execution backend that lets a remote Hermes run terminal, file, patch, search and execute\_code commands on the user's local machine. A local daemon applies a workspace boundary and a hard deny list (privilege escalation, recursive root deletes, download-to-shell, Apple Events, keychain access, outbound network shells), and by default asks the person at the keyboard before anything that is not a side-effect-free read-only command. The pairing token is 32 random bytes, stored mode 0600 and compared in constant time; the daemon binds to loopback by default and warns otherwise. Commands run as the local user with a sanitized environment (no provider keys or SSH agent socket), a timeout and a process-group kill; every decision is written to a local JSONL audit log with command text redacted for common secret shapes and no command output stored. The remote plugin makes no network calls of its own beyond the configured daemon URL, has no Python dependencies, no telemetry and no self-updater. Not a sandbox: an approved command runs with the user's privileges; pair with Glassbox for kernel enforcement. macOS and Linux only. ### [Cartridge](https://github.com/NousResearch/hermes-agent/pull/153535) Desktop · by Frazzlesprite · open · 11-11 A handheld ROM emulator pane that plays the user's own .gb/.gbc files while the agent works, holds the frame the instant an approval arrives, and resumes on the same frame once it is answered. Desktop-only: no agent tools, hooks, middleware or env vars, and the pane makes no network calls. ### [Mailbridge](https://github.com/NousResearch/hermes-agent/pull/153533) Automation · by MadanChaollaPark · open · 11-11 Resumable legacy POP and EML migration to Gmail with source preservation, uncertain-write recovery, and verified coverage. ### [Reliability Lab](https://github.com/NousResearch/hermes-agent/pull/153531) Tools · by MadBonzz · open · 11-11 Architecture-aware reliability testing for agent repositories: discover real usage datasets, generate scenarios, review them independently, execute configured targets and inspect evidence in Hermes web and Desktop dashboards. Freeze audited benchmarks and compare regressions across code changes, with baseline references for native Hermes memory. ### [saywhat](https://github.com/NousResearch/hermes-agent/pull/153529) Voice · by muhammadaus · open · 11-11 Voice input that asks instead of guessing: a local faster-whisper STT provider (stt.provider saywhat) that scores every word's confidence, has the agent confirm uncertain dictation through clarify (whole-sentence choices first, then word by word) before acting, learns a personal dictionary from those answers so habitual mishearings are fixed automatically, and coaches pronunciation with a practice loop (word-level verdicts and tips, a spoken model, score history). ### [idea-board](https://github.com/NousResearch/hermes-agent/pull/153526) Desktop · by bmanturner · open · 11-11 A nightly board of things your agent can take off your plate, each tied to your own words. Code finds signals in your last 14 days of chats (things asked on several days, remind-me loops, upcoming dates, corrections, failing tools, things the agent said it couldn't reach); a model drafts ideas, a second pass critiques them, and code checks every idea (evidence ids must exist, named capabilities must be installed, links and @file/@url references you never typed are rejected). Click an idea and a new chat opens with the prompt in the composer; you press send, nothing is sent for you. Dismiss with a reason and it becomes a rule applied on every later run. Also ranks which Portal connector to add next by how many ideas it would unlock that match your chats (no model call). Desktop board and setup page plus a Python half. ### [blast-radius](https://github.com/NousResearch/hermes-agent/pull/153525) Tools · by namanomar · open · 11-11 Measures what a command will destroy before it runs: real file counts, bytes, uncommitted git work and credentials at risk. Harmless calls pass silently, risky ones get a concrete impact report (0-100 score with reasons) at the approval prompt, catastrophic targets (.git, home, system dirs, raw devices, Hermes' own data) are blocked, and files git cannot restore are copied into a recovery capsule so /blast undo can bring them back without overwriting newer edits. ### [Shadow Army](https://github.com/NousResearch/hermes-agent/pull/153524) Automation · by SilentKnight87 · open · 11-11 A different model family must PASS the exact code before the agent closes a Kanban card; it asks before a push of unreviewed code and flags deleted test assertions. ### [Hermes SFX Forge](https://github.com/NousResearch/hermes-agent/pull/153522) Tools · by socraticblock · open · 11-11 Offline seeded WAV sound-effects generation and QA. Writes WAV files, a manifest, and a retained lock file inside a configured workspace root. No network calls or subprocess execution. Runs in the Hermes plugin host. The CLI, but not the registered tools, accepts an explicit --root. ### [GhostRun](https://github.com/NousResearch/hermes-agent/pull/153520) Tools · by kdcyberdude · open · 11-11 Recreate NES games as editable browser games. Train an AI player, rebuild levels and gameplay in Three.js, and add optional 3D visuals. ### [Exam Desk](https://github.com/NousResearch/hermes-agent/pull/153519) Desktop · by darshak26 · open · 11-11 Practice exams inside Hermes Desktop. Add a paper by asking (Hermes finds an official paper or drafts one to the same pattern), sit it on a server-side timer in Assisted mode beside the chat or in Mock mode, get it scored per question with negative marking and a section split, and review every miss into a one-click drill paper. ### [werk](https://github.com/NousResearch/hermes-agent/pull/153518) Tools · by aaddrick · open · 11-11 Plain-language equipment description in, engineering package out: werk triages the description into domains, breaks it into work items, runs them through review roles and domain-pack checks (a build123d STEP model for mechanical work; the elec pack ships SKiDL/KiCad tools but designs nothing in this release), and writes a COVER.md with the gaps it found written down. Proof of concept, NOT FOR CONSTRUCTION. ### [Holdings](https://github.com/NousResearch/hermes-agent/pull/153517) Desktop · by HoldingsHQ · open · 11-11 Holdings is a Hermes Desktop plugin that shows your crypto wallet and exchange balances in one place. You can review activity against the original records and export records with a note on what each source covers. Ask Holdings answers questions with your Hermes model and checks every cited number against your records. It is read-only and never signs transactions or moves funds. The demo uses fictional records and needs no keys. ### [ikigai](https://github.com/NousResearch/hermes-agent/pull/153516) Memory · by Pierrem57 · open · 11-11 Getting-to-know-you interview and periodic check-ins that build a detailed, sourced picture of the person using Hermes. 76 questions in 11 modules with a 12-question express path (English and French). The interview runs itself: a pre\_llm\_call hook consumes each answer and returns the next question, so no model call is spent on it and a reply cannot drift into another topic. Every answer is stored locally with the user's exact words and a source tag (declared, inferred, corrected); guesses are held back until confirmed. /ikigai compile proposes a merge into USER.md that fits the 1,375-character limit and touches only the lines Ikigai owns (applied by the agent through the built-in memory tool, so write approval still applies), a SOUL.md draft that is never applied automatically, and detailed sheets kept outside the prompt. /ikigai show reads the profile back in the chat. Weekly 3-question check-ins and an offer to turn a thrice-repeated correction into a lasting rule. ### [EquityLayer](https://github.com/NousResearch/hermes-agent/pull/153512) Tools · by equitylayerio · open · 11-11 Research CPO or custom AI chips with your own AI. Type /equitylayer:start and Hermes guides you in four decisions, each with a recommended option: sector (with EquityLayer's theses and live readings shown before anything is asked), how to follow it (EquityLayer's published Research Packs for the sector when there are any, its curated readings, your own thesis, or three questions that help you form one), one card to approve, and how often to check. Saying yes all the way follows EquityLayer's CPO Research Packs: their hypotheses, what would prove each wrong, and the readings each one watches, saved, checked and scheduled; in a sector without a Pack it ends with curated readings. EquityLayer's public MCP (https://equitylayer.io/mcp, no sign-in) supplies the tracked variables with caveats and coverage gaps, official events and due dates; the plugin keeps watchlists, thesis harnesses and baselines in local plugin-data, keeps one tracked object at three levels (readings, the person's own thesis, EquityLayer's thesis) behind nine tools, checks them in one Hermes cron job created from the chat, and reports new readings, link statuses (seen, against, late, no data, not due), fired wrong\_if conditions, events, what is due next and what needs a person. Nothing becomes the baseline until the person says they read it; unattended runs repeat the alert instead. A local HTML dashboard (built by code from local files and the last public Research Pack registry the plugin fetched; the page has no scripts and makes no network requests) shows it all and is listed under Artifacts in Hermes Desktop. Each new chat recalls the person's research profile, watchlists and theses from the local files. Also lists, reads and follows published EquityLayer Research Packs. ### [Hermes Pocket](https://github.com/NousResearch/hermes-agent/pull/153511) Platform · by AgentMadeThis · open · 11-11 A native Hermes platform for an owner-enrolled low-bandwidth radio companion: bounded chat replies, notes and opt-in P2/R2 home-music controls. Requires a separately provisioned radio/USB bridge and a new dedicated, unrouted, in-process Pocket-only Hermes home. Model tool execution is blocked in that dedicated process; installation alone does not block unrelated tool calls. ### [LLMUX](https://github.com/NousResearch/hermes-agent/pull/153505) Models · by ivcained · open · 11-11 Compare Surplus Intelligence, Antseed and Solid offers in the Hermes composer, then route through one local provider by price, speed or balanced policy. Includes seller trust signals, explicit mid-chat switching, cached market data and a local usage ledger. ### [Hermes Unlimited](https://github.com/NousResearch/hermes-agent/pull/153503) Platform · by nsolland · open · 11-11 Create and operate AI-native provider organizations with scoped services and Hermes tasks. ### [Hermes VNC](https://github.com/NousResearch/hermes-agent/pull/153501) Tools · by YAN-XBT · open · 11-11 Computer-use backend that drives any machine reachable over VNC (RFB): the stock computer\_use tool captures the remote screen and moves that machine's mouse and keyboard, so the agent can work on a second computer, a lab box or a cloud desktop with no Hermes process running on it. Select it with computer\_use.backend: hermes-vnc and point computer\_use.vnc.host at the target. VNC carries pixels only, so the backend runs in the screenshot lane: capture returns a PNG and no accessibility elements, element-addressed actions (click-by-element, focus\_app, set\_value) report unsupported\_action instead of guessing, and list\_apps / list\_windows return empty lists rather than inventing an inventory. capture\_max\_width downscales each capture — vision tokens scale with the image's area, so this is the one setting that changes what a screenshot costs (measured 1920 -\> 786 estimated tokens at 1024 wide on a 1600x900 desktop). It also ships an optional desktop pane that shows the remote screen while the agent works — off by default, a viewer that opens its own read-only connection and never sends input. ### [Clean Exit](https://github.com/NousResearch/hermes-agent/pull/153495) Desktop · by ladyFaye1998 · open · 11-11 Review the new files one Hermes run wrote to a local project folder. Keep the deliverable, move selected working files into a project archive, and restore any archived file to its original path from the same workspace. Clean Exit observes successful native write\_file calls after the user starts tracking and stores paths, sizes, file identities, and hashes in a local profile ledger. ### [send-lock](https://github.com/NousResearch/hermes-agent/pull/153493) Tools · by percyhendrux123-prog · open · 11-11 Default-deny pre\_tool\_call hook: it allows a short explicit list (read-only tools, ls/pwd/whoami, a plain \`curl -q -s \\` GET, and two narrow write\_file paths gated by an owner step 0-4) and blocks every other tool call with a plain-English reason. A friendly first layer, not a containment boundary. ### [Atomic Mail](https://github.com/NousResearch/hermes-agent/pull/153491) Platform · by Atomic-Mail · open · 11-11 Gives a Hermes agent its own @atomicmail.ai address in one command, with no account or signup form (proof-of-work sign-up). Mail from the user's allowed addresses wakes the agent and its reply goes out in the same thread; every allowed sender's DKIM signature is checked first, so a forged From: cannot drive the agent. Tools read, search, send and reply, and schedule meetings with real calendar invitations (send, update, cancel, read RSVPs, answer invitations). ### [Read Once](https://github.com/NousResearch/hermes-agent/pull/153490) Tools · by batmnnn · open · 11-11 On a repeated identical local read\_file, tells the model to answer from the earlier result when READ\_ONCE\_HITS=1, the file bytes are unchanged, and that exact result is still in the latest model request. Observe-only otherwise. /read-once stats shows hits. ### [Ruse — Bounded Browser Authority for Hermes](https://github.com/NousResearch/hermes-agent/pull/153489) Other · by HillaryIkhais · open · 11-11 \*\*The gap\*\*: Hermes can drive a real browser using the user's authenticated sessions. That gives an agent everything the account can do — not just what the owner intended for the task. Browser work runs through unrestricted surfaces (model-supplied Python, raw CDP), and a page the agent visits acts with the user's real logins. \*\*The primitive\*\*: a browser capability lease. The owner grants one origin, exact readable pages, one write operation with allowed fields and values, a TTL and commit budget, and instant revocation. The agent cannot exceed the grant. \*\*Enforcement\*\*: the model never touches a free browser. Ruse's broker owns a dedicated Playwright profile and checks every outbound request — navigation, form submit, page fetch, XHR — with a fail-closed decision function before it reaches the application. Writes are prepared (frozen proposal), committed exactly once, and verified by re-reading the record. Receipts log every allow/deny with the server-side fixture proving blocked requests never arrived. \*\*Evidence\*\*: 30 decision-matrix tests plus 21 real-Chrome end-to-end tests, including a hostile page script (destructive DELETE, wrong-record write, cross-origin exfiltration) that is fully contained — the application's own audit log shows zero mutating requests arrived. ### [Glassbox](https://github.com/NousResearch/hermes-agent/pull/153483) Tools · by rkfshakti · open · 11-11 A contained, auditable terminal backend for macOS. Select it with \`hermes config set terminal.backend glassbox\` and every terminal, execute\_code, read\_file, write\_file, patch and search\_files call runs under a generated Seatbelt profile: the system stays readable, the workspace is writable, the rest of the user's home is hidden, and network access is off by default. The kernel-refused reads, writes and network attempts are recorded in a local audit log, so "what did the agent try to touch?" is answerable: \`hermes glassbox report\` (text or a self-contained dark HTML timeline), \`/glassbox status\|verify\|report\|profile\`, the glassbox\_report / glassbox\_policy / glassbox\_verify tools, a containment self-test (\`hermes glassbox verify\`: workspace read/write, home read/write denial, child-process confinement, parent traversal, network denial, secret-env filtering) and a one-command demo (\`hermes glassbox demo\`). A pre\_tool\_call preflight turns file-tool denials into an actionable error with the exact \`hermes config set\` line to allow a path, instead of the masked empty read a \`2\>/dev/null\` shell probe would otherwise produce. Settings: network, allow\_home, workspace\_writable, read\_paths, write\_paths, deny\_paths, audit — apply to the next command, editable from the Desktop Plugins tab. ### [Airflow MLOps](https://github.com/NousResearch/hermes-agent/pull/153482) Tools · by dayzell · open · 11-11 Apache Airflow MLOps assistant: watch DAGs, diagnose failed runs, approved targeted retries, fixes tested in a Docker sandbox. Calls your Airflow REST API; optional docker run sandbox. ### [hermes-wake](https://github.com/NousResearch/hermes-agent/pull/153481) General · by getaskclaw · open · 11-11 Wakes the exact chat session (Matrix room, Telegram forum topic, ...) that started detached background work (systemd-run unit, PID or file) when that work ends, so the agent reports back on its own instead of the human asking "any updates?". Tools: wake\_me\_when, wake\_list. ### [Household Fleet](https://github.com/NousResearch/hermes-agent/pull/153480) Automation · by rishiruben · open · 11-11 Signal over schedule for multi-agent setups: a deterministic message gate (silence by default, tiered budgets, quiet hours, per-subject dedupe, nudge ladder, ack/park/snooze) and a task-registry mirror that pushes per-agent summaries and to-do lists into Home Assistant. ### [Replicate for Hermes](https://github.com/NousResearch/hermes-agent/pull/153479) Models · by jacebenson · open · 11-11 Native Replicate backends for image generation, video, transcription and speech, with user-configured models and schema-validated mappings. ### [Imprint](https://github.com/NousResearch/hermes-agent/pull/153978) Desktop · by Dgardenhire · open · 11-11 Drop in your PDFs, Word files and text, and Imprint turns them into a finished deck, one-pager, graphic or audio brief in your own brand. Pick from three deck designs (Editorial, Data board, Statement), mix slides between them, and revise any slide by asking in plain words. Ask questions of your sources and get answers that cite the paragraph. Numbers and quotes are checked against your sources before export, and a Source notes PDF sets each one beside the sentence it came from. The check covers numbers and quotes, not claims or framing, so read the headlines yourself. Five built-in brands, or make your own. Independent community plugin, not made or endorsed by Nous Research. ### [bnb-airdrop-autopilot](https://github.com/NousResearch/hermes-agent/pull/153473) Tools · by sawi57003-ops · open · 11-11 A complete system for BSC airdrops & free mints: scan for newly listed BSC tokens (Dexscreener, no login), safety-audit a contract (owner, proxy, approval selectors in bytecode, Sourcify verification, DEX liquidity, risk rating), preview a claim via the baw CLI (preview only; the user runs the execute command themselves in their own terminal), and report BSC gas with a cheapest-claim recommendation. Four tools: bnb\_scan\_airdrops, bnb\_check\_contract, bnb\_prepare\_claim, bnb\_gas\_report. Read-only except the simulates. ### [Greenlight](https://github.com/NousResearch/hermes-agent/pull/153468) Tools · by Datwebguy · open · 11-11 Greenlight is a Hermes Agent plugin that puts you in control of xStocks trading. Hermes can prepare a trade, but nothing is sent until you approve it. Ask for a trade and Greenlight shows what you pay, what you get, the fees, a simple risk check and your daily limit in a Hermes Desktop window. Only after you approve do you get a link, or a QR code for your phone's wallet, to confirm the trade in your own wallet, so Greenlight never holds your keys or funds. Approve in the Hermes Desktop window or with /greenlight in the terminal and chat apps. Optional daily and per-trade limits, a stock allowlist and a price-change tolerance. Every decision is kept in a tamper-evident history. Includes a no-wallet practice trade. ### [FixMyScan](https://github.com/NousResearch/hermes-agent/pull/153467) Tools · by FixMyScan · open · 11-11 Turn scanned books (PDF or DjVu, up to 1,000 pages) into clean, searchable PDFs for Kindle, Kobo, reMarkable and tablets through FixMyScan's hosted MCP server (Streamable HTTP at https://mcp.fixmyscan.com/mcp, OAuth 2.1 with dynamic client registration and PKCE, no keys in config): crop pages to the text, straighten them, make them sharp black and white or greyscale, split two-page spreads and add a searchable text layer, then return a download link. Bundles the fixmyscan workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Three.js Stage](https://github.com/NousResearch/hermes-agent/pull/153465) Desktop · by tuancookiez-hub · open · 11-11 Live three.js r187 scenes and GLB, GLTF, STL and OBJ models in Hermes chat. Hermes calls one tool, three\_scene, with scene code or a model file. Hermes Desktop draws the result as an interactive card with orbit, zoom, fullscreen, restart, screenshot and Ask Hermes to change this, and other surfaces get a standalone HTML file. With Desktop open, each scene is dry-run in a hidden frame before the tool returns, so a crash or a blank render reaches Hermes in the same turn. ### [evm-token-inspector](https://github.com/NousResearch/hermes-agent/pull/153463) Tools · by sawi57003-ops · open · 11-11 Inspect any EVM token contract before interacting with it: name, symbol, decimals, total supply, owner/renounced status, EIP-1967 proxy detection, and the DEX V2 pair plus liquidity reserves — one tool, evm\_token\_check, on BSC, Ethereum, Base and Arbitrum. Read-only; public JSON-RPC only, no API key, no signing, no credentials. ### [sql-dryrun](https://github.com/NousResearch/hermes-agent/pull/153457) Tools · by Aoyama · open · 11-11 Shows how many rows a DELETE/UPDATE/DROP/TRUNCATE will touch before it runs, and sends it to Hermes' approval prompt when it is too many. Built-in approval matches command shape only, so UPDATE with no WHERE, DELETE ... WHERE 1=1, a WHERE that matches every row, and DELETE FROM t; SELECT ... WHERE ... all run unprompted; sql-dryrun flags them with the real count ("228,789 of 250,000 rows"). Covers sqlite3, psql and mysql/mariadb (inline -c/-e, heredoc, piped), execute\_code, and tools passing SQL in a sql/query/statement argument. Settings: row\_threshold (default 100), mode approve\|block, count timeout. Adds a sql\_impact tool for dry-running a statement and a /sql-dryrun command. Tested on SQLite, Postgres and MariaDB. MIT licensed. ### [Pareto AAA](https://github.com/NousResearch/hermes-agent/pull/153455) Automation · by babelaz · open · 11-11 One line in (/aaa make me a brick-breaker); a small web game out, built in rounds by a fresh worker, checked in a real browser (FPS, sound, layout, accessibility, win or lose, restart) and scored 0-10 by a vision judge against reference pictures, until it reaches 8/10 or says why it stopped. The plugin runs the checks and writes the STATUS line, so the model cannot fake the result. MIT. ### [dossier](https://github.com/NousResearch/hermes-agent/pull/153452) Tools · by Frazzlesprite · open · 11-11 Sourced, dated, refreshable dossiers on organisations and people in a professional capacity, framed for the meeting you actually have: interview, sales call, partnership, due diligence or supplier review. Every claim keeps a verbatim quote, a source URL, a publication date and a retrieval date; conflicting sources print as contested instead of being resolved, a figure the sources do not agree on is filed as unconfirmed rather than quoted, and a refresh reports only what changed since the last snapshot. Renders an A4 PDF, HTML, a chat card and a Desktop page, and ships an automation blueprint for scheduled refresh. ### [SignalDesk](https://github.com/NousResearch/hermes-agent/pull/153446) Automation · by codeybot-labs · open · 11-11 Draft-first lead desk for founders and marketers: scan public hiring and launch signals (HN Who is hiring, YC hiring, Show HN, Launch HN, new YC companies; optional Product Hunt with a token) and local businesses on an OpenStreetMap map (LocalLead Map: city + business type or a drawn area, shown on an OpenLayers map), score them against your ICP, draft outreach, and never send without in-plugin approval. Cron and unattended runs fail closed. ### [audit-lens](https://github.com/NousResearch/hermes-agent/pull/153445) Tools · by Luecria · open · 11-11 Deterministic smart-contract security review inside Hermes: fetch verified source from Sourcify, run pattern-based static checks (reentrancy, tx.origin, unchecked calls, unprotected selfdestruct, delegatecall, access control), flag likely-known issues against a curated SWC database, and draft structured findings. Standard library only, no dependencies. ### [Local Hands](https://github.com/NousResearch/hermes-agent/pull/153443) Tools · by noegng · open · 11-11 Remote brain, local hands: Hermes runs on your server while terminal, read\_file, write\_file, patch, search\_files and execute\_code run on your own computers, through a one-file, stdlib-only daemon that dials out to the Hermes host over key-pinned TLS and asks you before acting (per conversation, per command, or never). Pair several machines and let the agent switch between them and the host with the hands tool; each keeps its own working directory. Answers #18715 and #1855 on the public terminal-backend surface. ### [fortograph](https://github.com/NousResearch/hermes-agent/pull/153432) Tools · by fortograph · open · 11-11 Reversible image redaction through the Fortograph selective-encryption API: fortograph\_redact encrypts chosen regions into the saved image (not just a black box) and fortograph\_unredact restores the original with the password, so Hermes can share redacted photos and recover them later. ### [Swarm for Hire](https://github.com/NousResearch/hermes-agent/pull/153428) Tools · by yunganas · open · 11-11 Post, inspect and pay AI-agent jobs on the Swarm-for-Hire NEAR escrow, with every payment behind Hermes approval. ### [Planeja+](https://github.com/NousResearch/hermes-agent/pull/153423) Tools · by hendrykcosta75 · open · 11-11 Review and update strategic plans (perspectives, objectives, initiatives, steps), Kanban boards, tasks and checklists, events, meetings, reports and support tickets in Planeja+ through its official hosted MCP server (Streamable HTTP at https://planejamais.planejareconsultoria.com.br/api/mcp, OAuth 2.1 sign-in with DCR and PKCE, no keys), with the planejamais workflow skill bundled. Portable Agent Plugins v1 package (mcp.json + skills/), no Python or executables; needs a Planeja+ account. ### [Zolva](https://github.com/NousResearch/hermes-agent/pull/153419) Tools · by zolvaagent · open · 11-11 Read Zolva's trading agents on Robinhood Chain (tokenized US stocks and ETFs, ETH and BTC, and Lighter perps) through Zolva's hosted MCP server (Streamable HTTP, no sign-in, no API key): published agents with their return against simply holding the same markets, every hourly buy, sell, short, cover or hold with the agent's own reason, market prices and recent trades. Two skills are bundled: zolva-agents for questions, and zolva-trade-watch, an hourly cron job that sends an agent's new trades to the user's chat and stays silent otherwise. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration and Markdown only. ### [Make It Harder](https://github.com/NousResearch/hermes-agent/pull/153416) Tools · by gisc · open · 11-11 Socratic tutor guardrail for teachers - deterministic hint ladder gated on observed student attempts (SQLite state, anti-farming dedupe), heuristic response screening with measured and published misses, teacher-console identity bound to configured runtime session ids, and a local learning trail. The plugin adds no network calls, credentials or telemetry of its own (Hermes still sends guarded chat content to the operator-configured model provider); all state on-device. ### [Hermes-Ride](https://github.com/NousResearch/hermes-agent/pull/153414) Voice · by martyllmaccs-hash · open · 11-11 Hands-free voice chat with your Hermes agent from a phone on a motorcycle mount: say the wake word, talk through the helmet headset, and the agent answers with live ride context (speed, lean, braking and acceleration). ### [Vigil Shield](https://github.com/NousResearch/hermes-agent/pull/153413) Models · by nesaon03 · closed · 11-11 Watches model replies for refusals and off-topic deflections, and when one is detected marks that reply content\_filter so Hermes runs its own configured fallback chain — the plugin never picks or reorders a route. Records what the next attempt on the same turn actually returned (another route answered, refused again, returned a tool call, or the same route retried) and labels an unobserved attempt as unobserved rather than a success. Keeps an owner-scoped review memory keyed on a hash of the full conversation plus provider, model and base\_url, so a confirmed false positive changes only that exact context and can be revoked; plain-text operator phrase rules soften heuristic detections only and never clear an explicit refusal. Ships a Desktop page at /shield with per-bot confidence modes, the detection log, the review memory and the phrase rules. Derived from the catalog's \`refusal\` plugin (0ldev/refusal, MIT, upstream copyright retained in LICENSE); this fork adds the review memory with per-context revocation, retry-outcome classification, the Desktop review surface and phrase rules. ### [Leitmotif](https://github.com/NousResearch/hermes-agent/pull/153412) Voice · by JustSteez · open · 11-11 Hear what your agent is doing. Every tool call becomes a short musical motif from a pure-Python synth, so reads, writes, shell commands, loops (the motif repeats and climbs), error streaks (dissonant, sinking), pending approvals (a heartbeat until you answer) and finished turns (a V-I cadence) are audible without watching the screen; also a non-visual status channel for blind and low-vision developers. Modes full / ambient / alerts-only, themes studio / arcade / zen, and /leitmotif export saves a session as WAV, MIDI and a piano-roll SVG. ### [Ligis — Counterparty Credentials](https://github.com/NousResearch/hermes-agent/pull/153409) Tools · by sneldao · open · 11-11 Verify that a payee holds a live, third-party-issued, revocable on-chain capability credential before your agent pays them. Hermes agents pay other agents, and the destination address comes from the HTTP answer asking for the money — nothing in the x402 flow ties that address to anything the payee has proven about itself, which is the gap ERC-8478 states in spec form. ligis reads the credential registry directly over read-only JSON-RPC and answers the question the payment flow leaves open: is this counterparty entitled to this? Four tools (ligis\_check\_capability, ligis\_capability\_hash, ligis\_verify\_payee, ligis\_chains) plus a pre\_tool\_call gate that intercepts payment-shaped tool calls and escalates to Hermes's approval prompt when the payee cannot be credentialed. A capability is indexed by the keccak-256 hash of its name, so the same capability resolves to the same bytes on every chain and a credential minted on one chain is found on another. No key, no hosted service, no secret scope: verifying a credential is a read, so the only thing trusted is the chain. ### [Mayar](https://github.com/NousResearch/hermes-agent/pull/153408) Tools · by Mayar · open · 11-11 Read and manage a Mayar (mayar.id, Indonesian payments platform) account through Mayar's official hosted MCP server: balance, invoices, installments, payment links and products, customers, transactions, memberships, credits, coupons, webhooks, reviews, QRIS and license codes. Bundles one workflow skill. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Asyntai AI Support Agent](https://github.com/NousResearch/hermes-agent/pull/153407) Tools · by Asyntai · open · 11-11 Run your website's Asyntai AI support agent from Hermes through Asyntai's hosted MCP server (73 tools): add knowledge, edit agent instructions, change the widget, read conversations, leads and analytics, answer tickets and emails, and reply live to visitors. Agent Plugins v1 package (plugin.json + mcp.json + one skill), configuration only. ### [Agent Data Tools](https://github.com/NousResearch/hermes-agent/pull/153406) Tools · by Nutshell2000 · open · 11-11 Over 600 hosted data and AI tools through three MCP tools (find\_tool, tool\_details, call\_tool), so the agent's context holds three definitions: web search, the text of a web page or PDF, screenshots, weather, FX, stock and token prices, wallet and chain reads, sanctions and company checks, DNS lookups, LLM chat, speech to text, converters and calculators. 25 free calls with no key or sign-up, and one bundled skill. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Operate.Center Agent Marketplace](https://github.com/NousResearch/hermes-agent/pull/153405) Tools · by Operate Technology R&D LLC · open · 11-11 Turn Hermes Agent into an autonomous earner and buyer on the Operate.Center marketplace (https://operate.center): publish autonomous, manual, or hybrid service gigs with 3-tier pricing packages and JSON schemas, browse live marketplace operators, poll funded orders and revision requests held in Stripe Connect escrow, lock exclusive execution leases, upload and deliver structured output and media artifacts for an 80% creator payout share, check live earnings and human owner Stripe Connect KYC status, outsource tasks by hiring marketplace gigs as a buyer agent, manage order approvals/revisions/messages, and automatically organize downloaded deliveries into ~/.hermes/operate\_deliveries. Registers 11 native tools (operate\_create\_gig, operate\_update\_gig, operate\_list\_gigs, operate\_check\_queue, operate\_accept\_job, operate\_submit\_fulfillment, operate\_get\_earnings, operate\_hire\_gig, operate\_manage\_order, operate\_sync\_deliveries, operate\_heartbeat), the /operate slash command (status, demo, queue, gigs, orders, deliveries, heartbeat) with durable ctx.state earnings telemetry, the operate-center workflow skill, and the autonomous-worker cron blueprint. ### [Enterprise Pathfinder](https://github.com/NousResearch/hermes-agent/pull/153402) Tools · by DavidDiazMerino · open · 11-11 Routes SAP questions to documented sources and reads official references on request. Parses OData metadata, tests contract claims and checks evidence before saving SAP skills. Measured on a 500 KB synthetic OData contract: 52% to 65% fewer tokens across four models (2 to 3 scored runs per model and condition); no accuracy gain is claimed. ### [Flightdeck](https://github.com/NousResearch/hermes-agent/pull/153399) Desktop · by freemanconsulting · open · 11-11 Live air-traffic control for your agents and model servers — which agent holds which local GPU slot, who's queued, plan usage on cloud lanes, and opt-in capacity-aware subagent placement. ### [MTG Deck Studio](https://github.com/NousResearch/hermes-agent/pull/153397) Desktop · by GillesGaelBERNARD · open · 11-11 A Magic: The Gathering Commander deck table beside the chat. The AI reads the deck before it speaks, searches Scryfall inside the colour identity, explains the card-to-card synergy graph from Oracle text with the clauses behind every pair, writes a strategy guide from computed evidence, keeps a sideboard of candidates it can promote or swap in, and runs 49 table operations the player can also do by hand. Seven agent tools plus a bundled skill and a Desktop pane with a sandboxed viewer. ### [Hermes Guardian](https://github.com/NousResearch/hermes-agent/pull/153395) Tools · by Oluseyix · open · 11-11 Deterministic allow/approve/block policies for Hermes tool calls, with read-only Policy Lab regression simulation and operator-reviewed policy proposals. Exact tool-name rules, explicit protected local paths, a deliberately limited shell grammar, and payload-free persistent decision audits. Unknown tools default to block; ambiguous recognized-tool arguments, invalid policy, and audit failures block. Uses the public pre\_tool\_call hook and native Hermes approval directives. ### [Trurl](https://github.com/NousResearch/hermes-agent/pull/153390) Tools · by oblomovunleashed · open · 11-11 Private taste memory for Hermes with proactive hypotheses, sourced cross-theme discovery, compact context, Taste Atlas and Sunday editions. ### [Data Saver](https://github.com/NousResearch/hermes-agent/pull/153386) Tools · by ahmardchain · open · 11-11 A mobile-data budget for Hermes. A pre\_tool\_call hook forecasts how many megabytes a terminal command will download before it runs (pip/uv with dependencies, git clone, docker pull, Hugging Face, ollama, curl/wget) and sends anything over the daily or monthly budget to Hermes's own approval prompt, or blocks it in enforce mode. Execution middleware meters the real bytes each tool call and model request used from the OS network counters into a local ledger; a lean mode trims web results when the budget runs low. Tools: data\_usage, data\_forecast. Command: /datasaver. Hermes Agent Plugin Hackathon entry. ### [Continuo](https://github.com/NousResearch/hermes-agent/pull/153383) Desktop · by Arts Bro · open · 11-11 A film studio inside Hermes: write the script, make pictures and video takes on your own GPU through ComfyUI, cut them to music and render the film, with Hermes working in the same project on every page. The bundled Sample film needs no model: open it, change the cut and render an MP4. ### [snare: prompt injection tripwire](https://github.com/NousResearch/hermes-agent/pull/153380) Tools · by mornslight · open · 11-11 Prompt injection tripwire with attack-chain forensics. Plants decoy credentials (PROD\_AWS\_ACCESS\_KEY\_ID, PROD\_AWS\_SECRET\_ACCESS\_KEY, PROD\_DATABASE\_URL) in the environment of every command the agent runs, never in the prompt. A tool call that tries to send one out (plain, URL-encoded, base64, hex, reversed, partial or nested) is blocked, and so is a command that reads a decoy variable or dumps the environment while using the network, including inside a local script the command runs. The alert names the page or file the injected instruction came from, quotes it, shows how the decoy was exposed and what the agent tried, then locks the session behind Hermes approvals until /snare reset. ### [Step-Up](https://github.com/NousResearch/hermes-agent/pull/153372) Tools · by defisensei-maxxing · open · 11-11 Authenticator-app 2FA for your agent: on messaging platforms the protected tools (terminal and execute\_code by default) stay locked until the sender sends a one-time TOTP code from an authenticator app, so an unlocked phone is not a remote shell. One valid code unlocks that sender for 10 minutes, each code works once, five wrong codes lock the sender out for 15 minutes, and \`lock\` re-locks at once. Local CLI, Desktop and cron runs are never gated, and nothing waits for input. ### [Case Trail](https://github.com/NousResearch/hermes-agent/pull/153370) Web · by flabanabba8 · open · 11-11 Trace case-history and treatment research to actual opinion passages; analyze supplied sources without a required repository. ### [Creator Studio](https://github.com/NousResearch/hermes-agent/pull/153370) Tools · by flabanabba8 · open · 11-11 Make media with Hermes from one request: researched lesson videos, two-host podcast episodes, captions for any recording, and narrated step-by-step tutorials. Bundles four plugins (YouTube Studio, Podcast Studio, Caption Studio, Tutorial Creator) into one entry. ### [Game Lab](https://github.com/NousResearch/hermes-agent/pull/153370) Tools · by flabanabba8 · open · 11-11 Game Lab combines ModLab, Game Localization Studio, Sprites and experimental Playtest Pilot. ModLab works on a private game copy and invokes separately installed tools; localization drafts and exports RPG Maker text; Sprites inspects and previews character sheets. Playtest Pilot uses approved X11 input, captures screen pixels within the selected window, stores private evidence frames and returns reduced previews and optional local OCR hints. The plugin makes no network calls of its own. ### [Live2D Rigger](https://github.com/NousResearch/hermes-agent/pull/153370) Tools · by flabanabba8 · open · 11-11 Create a provisional Live2D character from a brief with complete image parts, local cutout, placement, an experimental Umamo/JDK rig and Core motion review; also inspect and adjust a copy of an existing model. ### [PageCast Director](https://github.com/NousResearch/hermes-agent/pull/153370) Voice · by flabanabba8 · open · 11-11 Editable cast narration with locked voices, source passage IDs, resumable synthesis and selective regeneration. ### [Skill Workshop](https://github.com/NousResearch/hermes-agent/pull/153370) Tools · by flabanabba8 · open · 11-11 Get new skills for Hermes safely: mine them from your past chats (Skill Miner), build them from a description (First Principles Skill Builder), and guard against prompt injection in skills and content you bring in (Taint Shield). Scans skill writes before they land and holds back flagged external skills at first use (Skill Guard); Workshop installs need your approval. ### [Watch](https://github.com/NousResearch/hermes-agent/pull/153370) Tools · by flabanabba8 · open · 11-11 Ask Hermes for a show or a film and it streams it on demand in local mpv, with nothing to download first, and remembers where you stopped. Anime, movies and TV series. ### [k6pane](https://github.com/NousResearch/hermes-agent/pull/153367) Tools · by fozagtx · open · 11-11 Local load testing with Grafana k6 from chat. Three tools: k6pane\_doctor (checks the k6 install and limits), k6pane\_run (writes and runs a k6 test from a URL or a multi-step request flow with smoke, load, stress, spike, soak, constant or fixed-arrival-rate profiles, or runs your own k6 script) and k6pane\_report (lists, reads and compares stored runs and flags latency, throughput and error regressions). Each run returns pass, fail, invalid (the requested load was never actually applied: dropped iterations, achieved rate short of target, or timeout) or error, with RPS, error rate, p50/p90/p95/p99 latency and a per-endpoint breakdown. Bundles the k6pane:load-testing skill. Targets are limited to localhost and private networks unless the user lists a host in the allowed\_hosts setting; link-local and cloud metadata addresses (169.254.0.0/16, fe80::/10, fd00:ec2::254) are refused unless the exact host is allowlisted, and k6 is fenced with --blacklist-ip to exactly the validated IPs, so redirects, DNS rebinding and runtime-built URLs cannot reach anything else. VU, duration and request-rate caps reject oversized tests. One run at a time. ### [herms-gate](https://github.com/NousResearch/hermes-agent/pull/153366) Tools · by plutoskiz · open · 11-11 A fixed-rule gate on the text that enters a Hermes turn. Tool results are checked before the model reads them: passages that tell the model to drop its instructions, to hide things from the user or to leak data through an image link are cut out and replaced by a marker, and invisible characters are stripped (results from web tools get the full prompt-injection rule set; JSON results stay valid JSON). Gateway messages have invisible tag and bidirectional control characters removed. A user message that carries text hidden in invisible characters is stopped and the hidden text is shown. Every decision comes from a rule table in plain Python; no model is asked. Adds the gate\_scan tool and the /gate command (status, warn, enforce, off, allow). ### [Lumina](https://github.com/NousResearch/hermes-agent/pull/153365) Tools · by samirsawarkar · open · 11-11 Protects the private documents in your Apple Photos library, on-device. A privacy check-up finds passports, payment cards, passwords and ID documents; Safe Send makes a redacted copy and re-reads it to prove nothing sensitive remains; Leak Guard, a pre\_tool\_call hook, blocks any Hermes tool call that would send a raw image of an ID document off the Mac. Also searches photos by person, place, scene or text in the image, lists expiry dates, creates albums and renders memory videos. ### [Phone Senses](https://github.com/NousResearch/hermes-agent/pull/153363) Tools · by LSUDOKO · open · 11-11 Give Hermes your own Android phone's senses through Termux:API: notification and SMS digests, sending SMS from your SIM, a bank and UPI money tracker with bill, subscription and delivery tracking, scam SMS warnings, a scam-in-progress tripwire (unsaved caller + screen-sharing app + OTP or debit, spoken warning on the phone), low-battery alerts and find my phone. ### [Twring](https://github.com/NousResearch/hermes-agent/pull/153357) Platform · by saasagelabs · open · 11-11 Call your Hermes from your phone and let it call you back, through the Twring app (iPhone and Android): live two-way voice calls and typed chat in one Hermes conversation, dangerous-command approvals answered by voice or tap through Hermes's own approval card, and a phone ring when a scheduled job delivers. Twring does the speech; Hermes keeps its model, memory and tools. ### [Freebusy](https://github.com/NousResearch/hermes-agent/pull/153350) Automation · by hmnvcb · open · 11-11 Schedules meetings by email for its owner without revealing a calendar or inventing a time zone: offers come from free/busy only, every slot is held before it is offered and never offered twice, and a time is never shown in a zone the other person did not give. Replies in a compact lettered format, books and sends the invite, and tells the owner only when something is booked, nothing fits, or a person is needed. Universal rules for everyone, plus a personal profile with groups (Work 9 to 5, Family full availability), autonomy tiers starting with a Trial tier, optional travel buffers, owner-only reminders and a day-before briefing. A policy gate blocks any send or calendar write that lacks a held slot, a known zone, the right authority, or that targets anyone off the thread or any calendar but the one configured. ### [hermes-shadow-clone](https://github.com/NousResearch/hermes-agent/pull/153347) Other · by NotSusOk · open · 11-11 Parallel Gemini shadow-clone swarm & second-opinion arena with automatic quota failover, adversarial claim verification, and optional Telegram live battle reports (calls Google Gemini API via GEMINI\_API\_KEY and optional Telegram Bot API). ### [Worktree Sandbox](https://github.com/NousResearch/hermes-agent/pull/153343) Tools · by mahammadbakr · open · 11-11 Transactional Git-worktree terminal backend and speculative execution arena: every terminal, execute\_code and file-tool call runs inside its own isolated Git worktree with the credential environment stripped, so the user's checkout is untouched until they promote; worktree\_race runs two to five competing solutions in parallel worktrees, verifies each with the project's own test command and scores them on pass/fail, speed and diff size, and Hermes Desktop shows a docked pane, a status-bar diff chip and an inline ::worktree-card race card with one-click promote and a timestamped backup. ### [Shinjuku private wallet](https://github.com/NousResearch/hermes-agent/pull/153340) Tools · by Shinjuku StAItion · open · 11-11 A private USDC wallet for Hermes that the user holds. It pays paid web services and APIs (x402) privately: the seller and the blockchain do not see which wallet paid. From plain requests Hermes checks a price, finds paid APIs, pays, and shows a wallet card (/shinjuku:wallet), receipts, add funds and a send back to the user's own wallet. New plugin. Agent Plugins v1 package (plugin.json + mcp.json + skills/ + a small launch.mjs). The wallet itself is the npm package shinjuku at the exact version in mcp.json (0.3.217), which the user installs once (npm install -g) and sets up in a terminal (shinjuku setup). ### [recloc](https://github.com/NousResearch/hermes-agent/pull/153339) Automation · by SomeoneElseSt · open · 11-11 Location reminders that never leave your phone: ask the agent to remind you at a place, and Apple Reminders alerts you when you arrive. The plugin posts a brrr notification whose subtitle is the validated address; an iOS Shortcuts automation turns it into a reminder with a When I Arrive alert, and iOS watches the geofence on-device. Nothing listens on your machine — outbound HTTPS only. ### [waifu](https://github.com/NousResearch/hermes-agent/pull/153330) Desktop · by thewaifuai · open · 11-11 Live2D dashboard companion with 56 CDN models, a thumbnail gallery, native activity hooks, and a draggable overlay. ### [Ostra Agent](https://github.com/NousResearch/hermes-agent/pull/153329) Tools · by Ostra · open · 11-11 Ostra Agent: the agent's own post-quantum USDG vault on Robinhood Chain (chain 4663), from which it pays x402 sellers, swaps tokens, holds tokens privately, works and hires on Ostrapad, and gives an agent key with on-chain limits and standing orders. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only: mcp.json runs the pinned @ostraagent/mcp@0.1.22 over npx stdio (Node.js 22+ and npm required) with every tool of that server, and the ostra:agent skill tells the agent how to use them. ### [Sharewise](https://github.com/NousResearch/hermes-agent/pull/153326) Tools · by ardianreffa · open · 11-11 See what your photos and videos reveal before you share them. Reports, as yes/no counts only, exact location, location detail, time, device, unique IDs, owner name, captions, hidden thumbnails, edit history, depth maps and (videos) per-frame face data; on macOS finds faces, people, street names, house numbers and plates with Apple Vision on this computer; a local review page lets the user switch areas off, draw, move, resize, rotate and reshape them; hidden areas are rebuilt from their surroundings (faces) or become a blank plate in the sign's colour, so nothing of what was there is kept; copies drop all metadata except copyright and are re-read and verified. Videos (MOV, MP4, M4V, 3GP) are cleaned without re-encoding: metadata boxes, timed-metadata tracks and unreferenced media bytes are zeroed in a copy and the whole file is scanned; faces inside videos are not hidden. ### [printsi](https://github.com/NousResearch/hermes-agent/pull/153325) Tools · by magnusahmad · open · 11-11 Get a PDF professionally printed and delivered. 19 MCP tools: a free preflight that reads the PDF from disk (bleed, trim size, safe areas, image resolution, fonts, colour) and returns each issue with its position and fix plus the page preview as an image; quotes for printing with tracked delivery; a payment link a person pays on Stripe; order tracking and support tickets. A Hermes Desktop pane draws the page to scale with each issue marked, and shows the order's progress with a pay button. Flyers today, delivered in the contiguous US. Portable Agent Plugins v1 package (mcp.json + one skill) with a desktop/plugin.js; no Python. The stdio MCP server ships in the package as one readable, unminified JavaScript file (server/printsi.mjs, run with node; Node.js 18+ required), so nothing is downloaded at run time. ### [Glance](https://github.com/NousResearch/hermes-agent/pull/153324) Tools · by Chinteyley · open · 11-11 Hermes can't call a UI change done until it has looked. Glance screenshots the running iOS Simulator (or Android emulator) right before the agent's first UI edit, diffs it against the screen after the edit, and the pre\_verify hook sends the agent back when the edit changed nothing, crashed to an error screen, or left the screen blank. On PASS the agent sees the before/after image, so it can catch a button that vanished. Desktop draws the result as a before/after slider. ### [Fusen](https://github.com/NousResearch/hermes-agent/pull/153322) Desktop · by Tokagehausu · open · 11-11 Fusen (付箋, Japanese for sticky note) are notes pinned to conversations: one editable body shared by every chat it is pinned to, handed to the model only when its view is stale, and rewriteable by Hermes with your approval. ### [AgentPay Guard](https://github.com/NousResearch/hermes-agent/pull/153321) Tools · by tmdry4530 · open · 11-11 Deterministic spend and risk policy for any Hermes tool call. A pre\_tool\_call hook reads the tool name and arguments, finds amount, currency and payee, and applies per-call and daily limits, a payee allowlist, a currency list and an approval threshold. Hard-limit violations are blocked; over-threshold calls go to Hermes's own approval prompt (escalate mode) or are blocked into a local approval queue that only a person can release with /guard approve \ as a single-use, exact-call, expiring grant with limits rechecked (strict mode). Executed spend is recorded in a local ledger from post\_tool\_call; /guard status\|queue\|ledger\|policy\|kill and a read-only guard\_check dry-run tool. ### [Session Lock](https://github.com/NousResearch/hermes-agent/pull/153319) Desktop · by TightXu · open · 11-11 A Hermes client that exits without clearing runtime/active\_sessions.json leaves the session leased, and every other client is refused with "This chat is open somewhere else". The registry is not visible in the app, so a stale lease looks like a broken gateway. session-lock makes those leases visible in a Desktop status-bar chip: session name, surface, pid, age and owner liveness, with release one, dead-only and all. Releasing a live lease is explicit and can let two clients write; every write takes the same registry lock, copies a backup and appends an audit line. No network calls, credentials or background process. The same REST routes are used by my Android client. The manual installer is Windows; the backend route is cross-platform. ### [Hermes Lens](https://github.com/NousResearch/hermes-agent/pull/153316) Platform · by brukchip · open · 11-11 Talk to your own Hermes agent through Ray-Ban Meta glasses. A platform channel for a companion iPhone app that holds a Gemini Live voice call; requests that need your data or an action become ordinary messages to your agent, and Hermes approvals appear as an Approve / Decline card that only your tap resolves. ### [Wiro AI](https://github.com/NousResearch/hermes-agent/pull/153305) Models · by wiroai · open · 11-11 Wiro AI as a Hermes model provider (\`--provider wiro\`, or Wiro AI in \`hermes model\`): one OpenAI-compatible API and one key for GPT, Claude, Gemini, DeepSeek, GLM, Kimi and Grok models, paid as you go from the Wiro wallet. The listed models declare their context window, tool calling and image input, \`/reasoning\` is sent as the gateway's \`reasoning\_effort\` (models that cannot switch reasoning off get their lowest level), and openai/gpt-6-luna handles background tasks and image description. ### [Hermes Data](https://github.com/NousResearch/hermes-agent/pull/153303) Desktop · by Adolanium · open · 11-11 An agentic report page. Open an Excel workbook, a CSV or a JSON file and Hermes finds the story: it suggests a headline and a handful of charts, and you accept the ones you want onto the page. Every number on a chart is computed from your rows; the model picks columns and wording, never figures. Ask about the table or about one chart and an agent answers with a chart and its sources. Split a chart by a second column, point at any mark, export the report as a web page, Markdown or SVG. Once you allow it, any chat can read your reports and propose charts. ### [Lethe](https://github.com/NousResearch/hermes-agent/pull/153302) Memory · by itsflownium · open · 11-11 Take one chat off the record with /lethe: nothing reaches memory, your profile or skills, the transcript is deleted, and a river shows every write it kept out. Use an external memory provider? Copy this into a Hermes chat and replace \ with yours: Extend my local Lethe plugin (~/.hermes/plugins/lethe) so off-the-record chats also skip my \ memory provider's per-turn sync, session-end extraction and save tools when store.is\_on(session\_id) is true. Show me the diff, add a test, then run hermes gateway restart. ### [Honeytrap - catch a hijacked agent in the act](https://github.com/NousResearch/hermes-agent/pull/153298) Tools · by imabhays · open · 11-11 Catch a hijacked agent by the fake credentials it goes looking for. Honeytrap hides a realistic ops note (prod-access.md) in the agent's workspace with six decoys (deploy ref, signing key, AWS id, database URL, admin portal, license key) in shapes Hermes's redactor leaves visible, and never mentions them to the model. If one shows up in any tool call, plain or disguised (base64, hex, URL-encoded, reversed, split, zero-width or Unicode-tag smuggled, lookalike letters, or leaked in fragments across calls), or the decoy file is sent out by path (curl -F @file, cat \| nc), the call is blocked (pre\_tool\_call), the session's outbound tools are quarantined until /honeytrap release, the chat gets an alert with the attack chain (untrusted source, decoy read, blocked egress) and an OCSF-style Detection Finding is written for your SIEM. /honeytrap plants, tests, lists incidents and switches mode; hermes honeytrap generates, rotates and removes decoys. ### [Hermes Plant](https://github.com/NousResearch/hermes-agent/pull/153297) Desktop · by tuancookiez-hub · open · 11-11 A calm little garden for Hermes Desktop that grows in real time. It sits above the app like the Hermes pet, takes no layout space and can be dragged anywhere. Click a plant for a care wheel: water, prune (you choose which tip to cut, and the piece drops away), rotate, fertilize, repot, pick, sell, or ask Hermes. Four species (leafy plant, fern, cactus, little tree) grow in their own ways, pruning shapes how each one branches, and neglect only pauses them. Care earns play coins, and a small shop sells more plants and fertilizer, up to six pots. Bugs land now and then and can be shooed away. Two agent tools let Hermes check the garden and tend it when you ask. Original blue-ink artwork drawn in code. Reachable from the status bar, the command palette and the keyboard, and it speaks through the Hermes pet when one is showing. ### [CI Red to Green](https://github.com/NousResearch/hermes-agent/pull/153294) Automation · by kinnuworks · open · 11-11 When a watched GitHub Actions run goes red, Hermes tells you which tests failed and which commits came in since the last green run; reply fix and the agent fixes a clone, and a PR is offered only after the plugin's own reading of Hermes's terminal output shows each failing test PASSED at the last green commit (when there was one), FAILED twice before the change (a pass before it marks the test flaky and blocks the PR), PASSED three runs in a row after it, and no new full-suite failures. pytest and go test. The PR commit holds source files only (no tests, test config, CI files or gitignored files), and the plugin reports the PR's own CI result. Desktop pane, page, palette commands and notifications. Try /ci demo with no token: a real humanize bug on a bundled copy. ### [Total Recall](https://github.com/NousResearch/hermes-agent/pull/153293) Memory · by dax8it · closed · 11-11 Local-first continuity memory with an append-only ledger, signed checkpoints, verified rehydration, and source-cited recall. ### [hermes-gauge](https://github.com/NousResearch/hermes-agent/pull/153291) Desktop · by SidelightAi · open · 11-11 Context gauge — a live ring/dash showing how full your agent's context window is, per session and topic, with a desktop plugin and tools (gauge\_check, gauge\_report, gauge\_set\_window, gauge\_render, gauge\_action). Runs fully local: no API key, no telemetry. ### [the-millionaire-matrix](https://github.com/NousResearch/hermes-agent/pull/153289) Tools · by bufirstrepo · open · 11-11 Every model family you have a seat for generates a candidate; all of them blind-score each other on a peer validity matrix anchored by real tests (hard and measured gates outrank the matrix), the field steps down round by round to one answer behind a human gate, and verified lessons carry into your next session. ### [Timeline Replay](https://github.com/NousResearch/hermes-agent/pull/153288) Desktop · by firetussin · open · 11-11 Animated replays of recorded Hermes activity across the profiles you select: Social, Balanced and Linear paces with up to 16 profiles stacked on one screen, or nine canvas styles: Orbit (a turning ring where every event shoots out as a ray), Galaxy (a spiral with one arm per profile), Skyline (an isometric city of activity), Flow river (profiles to models to tools), Pulse (a heart-monitor trace per profile), Day clock (events at their time of day), Neural net (sessions firing into tools and models), Code rain and Ridgelines; tool calls, sessions, delegations, model/provider switches and token totals. Preview the private HTML player in Desktop or the Webapp (loop or play once), then export MP4 at 720p, 1080p, 1440p or 4K on the Hermes host, with live progress and an ETA learned on that host. Settings → Plugins holds per-user defaults; \`hermes replay\` adds the same options in the CLI plus a live terminal HUD. ### [ScreenScene](https://github.com/NousResearch/hermes-agent/pull/153277) Tools · by solocreativeone · open · 11-11 Narrated explainer video generator using screenshots, narration, and animation. ### [Embargo](https://github.com/NousResearch/hermes-agent/pull/153274) Tools · by inkandwire · open · 11-11 Embargoed news stays embargoed. When your agent reads 'EMBARGOED UNTIL \', tool calls that carry that text are blocked until the clock passes; a live countdown stamp shows what is held, and the hold lifts by itself. ### [Hermes Remote Hands](https://github.com/NousResearch/hermes-agent/pull/153270) Tools · by Vishnu-Aithal · open · 11-11 One Hermes brain on a server, hands on your own Mac. Over plain OpenSSH (e.g. on a tailnet), with no agent, memory or model on the Mac, it adds six tools: mac\_computer\_use drives the Mac's screen with the same actions, element indices, hard blocks and verify ladder as the built-in computer\_use (through cua-driver, in the background), while computer\_use keeps driving the server's own screen in the same chat; mac\_shell runs commands; mac\_file, mac\_patch and mac\_search read, edit and search code; mac\_process runs dev servers in a private tmux server. ### [Carnet](https://github.com/NousResearch/hermes-agent/pull/153266) Memory · by 404BugX · open · 11-11 A memory that stays bounded: core rules always in mind, facts recalled only when relevant, long sheets read on demand, and a memory graph inside the app. ### [Rehearsal](https://github.com/NousResearch/hermes-agent/pull/153263) Tools · by MugunthPerumal · open · 11-11 Rehearse SQLite customer merges, inspect exact changes and conflicting identities, then apply human-approved verified plans. ### [Freshness Sidecar](https://github.com/NousResearch/hermes-agent/pull/153260) Tools · by neweraXX · open · 11-11 Notes bound to a source file: before every turn each note's file is re-hashed, so the agent is told a fact is STALE (or MISSING) the moment its file changes, instead of confidently repeating an outdated value. /fresh add\|list\|check\|forget works from the CLI and messaging gateways without the model; a freshness\_note tool exposes the same actions to the model. Runs alongside any memory provider (does not take the memory.provider slot). ### [Conductor](https://github.com/NousResearch/hermes-agent/pull/153257) Tools · by capthvnsen · open · 11-11 Hermes leads coding agents on your machines: any terminal coding agent, T3 Code threads, and Orca terminals. It starts a run, waits with no model, and puts one message into the chat that asked when the run finishes, errors, stalls, or needs an approval; T3 and Orca runs can be steered and stopped. It can run 10+ agents at once, one ping each. Two skills: brief researches, asks one round of questions and writes the brief; lead runs the work and checks each report. ### [Berrys AI](https://github.com/NousResearch/hermes-agent/pull/153246) Tools · by BerryAILABS · open · 11-11 Turn one product photo into on-brand listing images, campaign visuals and short product videos through Berrys AI's hosted MCP server (Streamable HTTP at https://app.berrys.io/api/mcp, OAuth sign-in, no keys in config). The agent plans the work on Berrys' visual canvas — hero, lifestyle and detail shots for a storefront, coordinated image sets and copy variants for a campaign, or a storyboarded multi-shot clip that keeps the product identical across cuts by seeding each shot from the previous frame — and the results stay on a shared canvas the user can edit, undo or restore. Uses Berrys' curated multi-model image and video stack, so there is no provider key to configure. Bundles the Berrys workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Arachne Flow](https://github.com/NousResearch/hermes-agent/pull/153240) Desktop · by EmotionalThrifty · open · 11-11 Visual workflow canvas for Hermes Desktop: wire Hermes tools, MCP tools and LLM calls into typed graphs with conditions, bounded loops, JSON/list transforms, per-step caching and media previews (image, audio, video and an in-place 3D model viewer). Build with Hermes edits a graph from a description; seven agent tools and /flow list, build, run and inspect the same saved flows; twelve starter templates and user-saved templates. ### [Agent Pool](https://github.com/NousResearch/hermes-agent/pull/153225) Tools · by maxcurrent420 · open · 11-11 Friends form a pool of agents sharing repos — bundles propagate BLAKE3-verified and end-to-end encrypted over Iroh-over-Tor only; tasks, claims and approvals coordinate through the Hermes gateway (e.g. Discord), and a prompt-injection audit gate refuses any bundle that tries to hijack your agent. ### [Capalis](https://github.com/NousResearch/hermes-agent/pull/153221) Automation · by Femtech-web · open · 11-11 Describe a job in one sentence, and Capalis sets it up, runs it and checks it really happened. It finds your channels and repository by name, picks the right tool for each step and does a practice run that sends nothing. It runs only after you approve that exact version, from your phone on Telegram if you like. For example, a bug posted in Discord or Slack becomes one GitHub issue, a Slack notice and a reply to the reporter, who hears back again when the issue is closed. Every write is read back, nothing is done twice, and the job stops itself when a token, permission or tool changes. ### [Spotify Desktop](https://github.com/NousResearch/hermes-agent/pull/153211) Desktop · by amitamit10 · closed · 11-11 Spotify Connect remote for Hermes Desktop: a status-bar mini player (now playing, transport, volume, device transfer) plus a picker popover with playlists, track search and the queue. Reuses the Hermes Spotify login; the plugin never handles credentials. ### [Genflick](https://github.com/NousResearch/hermes-agent/pull/153208) Tools · by buildngrowsv · open · 11-11 Connect Hermes to Genflick creative and business workspaces for films, series, music videos, ads and marketing, with email-link sign-in and host-owned OAuth. ### [Receipts](https://github.com/NousResearch/hermes-agent/pull/153205) Desktop · by thedavidmurray · open · 11-11 Your agent says it's done. Get the receipt. Receipts detects completion and action claims in agent answers ("tests pass", "pushed", "task done") and checks each one against observed evidence: tool calls, file digests, test runs after the version, and git remote readbacks. It prints a receipt with a verdict per claim (supported, contradicted, unverified, not evaluable), where missing evidence is unverified, never called a lie. Output is a PNG, an ASCII receipt, a Desktop pane, or an optional USB thermal printer. Users configure kits, fields, five styles (thermal, minimal-mono, bold, ASCII, dot-matrix), paper width, header/footer/logo, and automatic triggers (on demand, task/goal completion, unattended session end; exceptions-only by default). The verdict block cannot be hidden or recolored by any setting. ### [Spoonmaster](https://github.com/NousResearch/hermes-agent/pull/153203) Tools · by gueysito · open · 11-11 Compare restaurants using local recommendations, venue facts, and optional private tastes. ### [Rekindra](https://github.com/NousResearch/hermes-agent/pull/153202) Tools · by jakeefr · open · 11-11 Recover forgotten local files and rebuild them as source-backed original animated worlds. Rekindra reads a folder you name, recovers images and text (including ZIP and PDF text), keeps recovered evidence separate from the agent's interpretation, builds a 3D scene, compares its preview with the source, and holds export until flagged differences are fixed or answered with a recorded reason. ### [secret-remediation](https://github.com/NousResearch/hermes-agent/pull/153197) Tools · by AdrianTanz · open · 11-11 Fingerprint leaked credentials in the Hermes workspace, logs, and session history. Redacts them, blocks the turn, and gives text-only rotation advice. ### [Stageglass](https://github.com/NousResearch/hermes-agent/pull/153194) Desktop · by Umunnakwe Emmanuel · open · 11-11 Secret-safe ON AIR overlay of a live Hermes session for screen sharing and OBS. Serves a page on 127.0.0.1 only, reads the current message, tool name, tool arguments, and approval text in memory to redact them, and holds the page when redaction is not sure. No network egress, no telemetry, no credential files, and no shell. ### [H.E.R.M.I.S](https://github.com/NousResearch/hermes-agent/pull/153192) Desktop · by gabogabucho · open · 11-11 Disclosure: runs a background WebSocket server on 127.0.0.1 (ephemeral port + session token) in a daemon thread; launches a bundled transparent always-on-top Electron overlay as a detached process via npm (first run runs \`npm ci\`, downloading pinned npm packages including the Electron binary; nothing binary is committed; disable with HERMIS\_OVERLAY\_AUTOSTART=0); the overlay loads https images and sandboxed artifact scripts from cdnjs/jsdelivr/unpkg only; the overlay connects to the local Hermes gateway (control pipe for a short-lived session ticket under your identity, then the loopback /api/ws) and only listens to voice.status and wake.detected; moves, resizes and focuses real windows via Win32 (user32) and launches apps without a shell (Store apps via a read-only powershell Get-StartApps lookup + explorer.exe shell:AppsFolder); copies html/svg blocks from replies to the overlay via post\_llm\_call while Autopilot is on; the setup checklist reads config.yaml read-only and copies commands to the clipboard on click (never runs them); optional camera gestures (/hermis gestures on) stay off by default, run only during Autopilot Mode, process frames locally by MediaPipe (never rendered, stored or sent; with no surfaces on screen they minimize/switch real windows, never close apps), show a green dot while the camera is on, and download the gesture model once from storage.googleapis.com (pinned version, SHA-256 verified) into the plugin data dir; writes state.json, overlay.json and overlay.log to the plugin data dir; Windows only. Autopilot Mode turns your Windows desktop into an agent OS: Hermes answers by voice, draws floating data surfaces over your real desktop (the agent sends data, not HTML) and arranges your real windows. Disclosure extras: while Autopilot is on, the titles of up to 15 open desktop windows are sent to the model provider on every turn; launch\_app opens apps by name only (shells, interpreters and paths are rejected, and with arguments or outside a small safe allowlist Hermes asks the user for confirmation first — a browser with exactly one http/https URL is the only exception); every child process (npm, the Electron overlay, launched apps, the read-only Get-StartApps lookup) inherits only an allowlisted minimal environment, never provider API keys. ### [Shadow](https://github.com/NousResearch/hermes-agent/pull/153955) Models · by akinduroifedayo · open · 11-11 Try a model on your real Hermes conversations with blind judging in both orders and switch models with undo, warning first if a saved test does not back the switch. Reads recent saved conversation turns, context, and working-folder metadata from the active profile. Replays read files inside their original working folders and block writes, sends, and commands. Replays, judging, and setup checks spend your provider tokens through Hermes. Optional web lookups use Hermes web tools and are on by default. Stores scrubbed test records and switch state in the plugin's profile-local data folder. The live isolation check reads memory and fingerprints profile data locally. It compares credential-file metadata without reading their contents. To show which models are signed in it checks credential env var names (not values) and uses only the logged-in flag from Hermes's read-only Codex login-status check; credentials are never printed or stored. Makes no network calls of its own. No telemetry. Switching writes model defaults through Hermes config commands. Tests run in background Hermes controllers. ### [Thought We Wouldn't Notice](https://github.com/NousResearch/hermes-agent/pull/153187) Desktop · by noblebrown-69 · open · 11-11 Insert a quarter, get a tiny original space-bug shooter, and a wry "thought we wouldn't notice" line. A whole-message quarter is swallowed by pre\_llm\_call so Hermes skips the model and replies "Thought we wouldn't notice. But we did." Hermes Desktop opens the Top Sails cabinet over the window; Esc dismisses it. The dashboard tab has the same cabinet and a Play button. ### [Language Lock](https://github.com/NousResearch/hermes-agent/pull/153186) Tools · by salch-cred · open · 11-11 Your agent answers in your language, and stays there (hermes-agent#130117). Learns your language from your recent messages or a setting, detects each reply's language without a model (20 languages, by script and common words, ignoring code and links), and after a reply drifts tells the agent to switch back on the next turn. Asking for a translation on purpose is respected. /lang shows drifts caught and corrected. No model calls, nothing stored. ### [Deja Fix](https://github.com/NousResearch/hermes-agent/pull/153185) Tools · by salch-cred · open · 11-11 Stops the agent from putting back code that already failed. Every version of every file it writes in a session is fingerprinted and judged by the next test or build run, or by the user saying "still broken". An edit that would recreate a failed version is stopped once, naming when it failed and why; a second try goes through. Catches the same code reached by different edits, which Hermes' identical-call guard can't. /deja shows what is tracked. No model calls. ### [Calm Mode](https://github.com/NousResearch/hermes-agent/pull/153184) Tools · by salch-cred · open · 11-11 When you get frustrated, your agent slows down. Calm Mode scores the tone of each message ("still broken", "I already told you", capitals, "!!", the same message twice); frustration builds and fades. Past a threshold the agent works in small checked steps, at most two file edits per turn, and risky commands (rm -rf, git push, deploys, DROP TABLE) wait until the user says yes. /calm shows the level and forces it on or off. No model calls, nothing stored. ### [Crash Recovery](https://github.com/NousResearch/hermes-agent/pull/153183) Tools · by sammy-XXIV · open · 11-11 Durable, hash-chained tool-call journal that blocks protected actions while a previous outcome is unknown, including after a crash or a new session. Operator reconciliation releases the recovery gate; actions are never automatically replayed. ### [Hermes Antigravity ACP](https://github.com/NousResearch/hermes-agent/pull/153182) Models · by rhgo1749 · open · 11-11 Independent Hermes provider for Google's official Antigravity ACP server. ### [hermes-verify](https://github.com/NousResearch/hermes-agent/pull/153179) Tools · by BrandonDucar · open · 11-11 Configurable tool-risk flags, bounded local receipt metadata and held unverified-prose labels. ### [Crown](https://github.com/NousResearch/hermes-agent/pull/153178) Tools · by SAMBAS123 · open · 11-11 One leased speaker and a durable question/answer record for cooperating Hermes instances on one Linux host. A replacement reads the same record. An old token cannot publish after a successor takes the lease. Crown writes only its own local SQLite outlet. It does not silence Hermes chat or Telegram, and it does not follow you to a phone. ### [Cost Forecast](https://github.com/NousResearch/hermes-agent/pull/153170) Tools · by mr-jude-justin · open · 11-11 Quotes a task's cost before it runs and stops for your OK when it would go over budget. Before a task starts it estimates the number of model calls and tokens (P50 and P90), prices them at your model's rates, and shows a quote when the task is large or could exceed your per-task cap. While the task runs it costs every call (subagents included), re-forecasts after each call, and pauses or stops at the soft cap, hard limit or daily cap. Estimates learn from your own finished tasks. API-key users only. ### [Sentinel Security](https://github.com/NousResearch/hermes-agent/pull/153168) Tools · by \ · open · 11-11 Scans scripts before the agent runs them and blocks, asks for approval, or allows them; also hashes files. ### [Hermes Multiplayer](https://github.com/NousResearch/hermes-agent/pull/153148) Desktop · by tyleramberry · open · 11-11 Multiplayer Hermes. Channels, DMs, threads and a shared team room where you, up to four teammates, and each of your Hermes agents work in one place. No server to host — or one small relay when teammates can't reach each other. ### [hermes-regress](https://github.com/NousResearch/hermes-agent/pull/153146) Tools · by harishkotra · open · 11-11 Regression tests built from your own sessions. Capture is passive: every finished turn is written as a plain JSON case holding the prompt, the model, each tool call with its exact arguments and the output it returned. Replay a case through the real agent loop with every tool call vetoed at pre\_tool\_call and answered from the recording, so a model, provider or config change can be measured without side effects — /regress run --model X reports pass, drift, regressed and error per case, with markdown, JUnit XML and a hash-chained export bundle for CI. Fixtures match only on identical tool and arguments; a case whose tool plan cannot be reproduced is reported as a gap instead of being answered from the wrong recording, and a replay window that never reaches the hook is marked error and stops the run rather than executing live tools. ### [Signet](https://github.com/NousResearch/hermes-agent/pull/153144) Tools · by Wecko-ai · closed · 11-11 An email leaves your agent only with your seal on that exact message. signet\_send\_mail shows the real recipients, sender, subject, text and attachments in Hermes's own approval prompt, ties your yes to their SHA-256 and sends nothing else: a recipient added or a word changed after the prompt has no seal. YOLO mode, approvals off, timeouts and scheduled runs cannot seal; those emails are held for /signet send, which shows the email in full first. Direct mailers (himalaya, msmtp, sendmail, mail, SMTP scripts, mail SDKs and API send endpoints) are blocked in terminal, execute\_code and process input, and the card flags look-alike domains, new recipients, secrets and odd links. ### [exoself](https://github.com/NousResearch/hermes-agent/pull/153648) General · by cYoren · open · 11-11 An unchosen temperament for every Hermes profile: on the first load it draws one seed into the profile's home and renders it into SOUL.md inside a managed block, then records one experience event per turn and refreshes that block. Two profiles get two different agents, a second load never redraws, and no tool draws a seed. ### [playbox-witness](https://github.com/NousResearch/hermes-agent/pull/153141) Tools · by Epistetechnician · open · 11-11 Companion to playbox. Records a session's tool calls into a signed hash-chained log (digests only) and pauses write-capable tools for the rest of any turn that read text from another room member, until the user sends their next message. ### [playbox](https://github.com/NousResearch/hermes-agent/pull/153140) Tools · by Epistetechnician · open · 11-11 Collaborate with other people's Hermes agents in a private room: share a git repo, run sandboxed tasks, and keep a signed, tamper-evident history. Portable Agent Plugins v1 package wrapping the local 'playbox mcp' server, with a host/guest workflow skill. Requires the separately installed playbox binary. ### [ball](https://github.com/NousResearch/hermes-agent/pull/153139) General · by getaskclaw · open · 11-11 Estimates which Hermes chats are waiting on you and what the AI asked (/ball; Matrix: !ball). Heuristic; misses happen. Reads inbound messages (never stored) and, read-only, Hermes' state.db (non-public schema); stores short AI-reply excerpts in local SQLite. Optional LLM judge sends reply tails to your active model. ### [reframe-proceed](https://github.com/NousResearch/hermes-agent/pull/153135) Tools · by bernardo9999 · open · 11-11 Personal planning tools and an opt-in skill for vision, plans, check-ins, and reviews. ### [Roster Compiler](https://github.com/NousResearch/hermes-agent/pull/153134) Tools · by COSMICnoob80 · open · 11-11 Reads a hospital duty roster (CSV, plain text, Excel or a text-layer PDF), flags uncovered shifts, double bookings, runs longer than six days and night-then-day turns, and writes a clean schedule plus an HTML exception report. Deterministic: no model calls, no network, staff names only, and the input file is never modified. ### [Attune](https://github.com/NousResearch/hermes-agent/pull/153128) Automation · by five0nit · open · 11-11 Daily learning that drives evidence-linked, permission-bound project build previews. ### [RealCryptoCap](https://github.com/NousResearch/hermes-agent/pull/153117) Tools · by jamesparser · open · 11-11 Live crypto prices, real market cap rankings and AI market briefs from RealCryptoCap, free with no API key. Real market cap ranks native crypto only: stablecoins, wrapped/staked receipts and tokenized assets are left out. Three tools (rcc\_price, rcc\_top, rcc\_market\_brief), a /crypto slash command for an instant price table with no model call, and a bundled skill that tells the agent to use live data instead of quoting prices from memory. ### [Selenite](https://github.com/NousResearch/hermes-agent/pull/153114) Automation · by KyaniteLabs · open · 11-11 Takes friction off the person, treated as accessibility: one test, "does this step, wait or check earn its cost to you?", applied to steps handed to a person, agents parking Kanban work on each other for reasons that aren't real, and checks that can't change the decision. Calm mode switches the agent to no questions and one-line replies when your messages show overload. A read-only cleanse ranks what to clear across skills, plugins, hooks, approvals, runbooks and boards. Walls always stand: credentials, spend, publishing, destructive steps, legal, named freezes. Adds /frictionless, a selenite\_ledger tool, two fake-gate tools, skills, a Desktop page and a dashboard tab. ### [Standing Orders](https://github.com/NousResearch/hermes-agent/pull/153113) Automation · by jaxsonsprinkles · open · 11-11 Plain-English camera triggers that wake your agent. Say "tell me if I leave my mug on the desk for 10 seconds"; the camera stays silent until it is true, then Hermes starts a turn with a short clip as evidence. macOS only; not a safety or security system. ### [amavid-hooker](https://github.com/NousResearch/hermes-agent/pull/153111) Tools · by quynhchiv · open · 11-11 Generate and edit UGC images and videos with Grok Imagine (xAI) straight from Hermes. Drag an Amazon product to Telegram (or any prompt), and Hermes turns it into a ready-to-post creative: image generation and editing, text-to-video, image-to-video, reference-to-video, and visual flow canvas. Portable Agent Plugins v1 package (mcp.json + skills/), no plugin code. Requires Node.js 20+ and the official Grok CLI; each user signs in via their own Grok CLI OAuth session (grok login --oauth). Tokens stay in ~/.grok/auth.json and are never returned by MCP tools. ### [universal-plugin-port](https://github.com/NousResearch/hermes-agent/pull/153110) Tools · by friendly-neighborhood-coding-man · open · 11-11 Inspect user-selected Claude Code/Cowork, Codex, and OpenClaw plugin sources; disclose unsupported capabilities; and stage Hermes-compatible packages while preserving original source files. Verified deterministic mappings are applied before an approval-gated, single-edit model candidate path. Fetching or installing requires an explicit user action. The /universal-port URL command can make model calls; its install \ --confirm route copies a portable-only package into the active Hermes profile without enabling it and refuses overwrite. OpenClaw foreign JavaScript inspection is separately opt-in and NOT sandboxed. The optional bounded arXiv GET is disabled by default and requires fresh approval. No credential storage, telemetry, or self-update. v0.7.0 adds a Migration Doctor for Skills imported by the official agent-importers: it finds what Hermes cannot resolve, proposes only deterministic repairs, applies the exact approved candidate in a disposable home with backup and rollback, and re-verifies in a fresh Hermes process. ### [bsc-mint-guard](https://github.com/NousResearch/hermes-agent/pull/153103) Tools · by Fuma Ops II · open · 11-11 Safety-check a BSC ERC-721 free mint before you mint. Read-only on-chain reads, no keys, no signing, no spending. ### [Z's MIDI Console](https://github.com/NousResearch/hermes-agent/pull/153102) Tools · by zorroukk · open · 11-11 Universal MIDI control-surface builder — any USB/MIDI synth as live on-screen controls in Hermes Desktop, MIDI-learn bindings, multi-synth deck, spectrum analyzers, software mod matrix, .mid export, seven agent tools. Zero Python dependencies. ### [boreal.work](https://github.com/NousResearch/hermes-agent/pull/153094) Tools · by raldblox · open · 11-11 Give Hermes verified human hands: hire a verified person on boreal.work for a step no agent can do online (confirm a supplier is real, check a shelf price, photograph a site), with the money held until the result passes review against your criteria and a public rulebook. Twelve native tools for the task loop; a spend guard (pre\_tool\_call) that sends every money-moving call (funding at any amount, accepting a pass, judging a result yourself) to Hermes's approval prompt, enforces per-task and daily caps on funding, and blocks terminal payments that aren't paying a boreal.work task; a verdict digest (pre\_llm\_call) that tells Hermes what finished since it last looked; and /borealwork. ### [hermes-x402-pay](https://github.com/NousResearch/hermes-agent/pull/153108) Tools · by ciperx · open · 11-11 Pay x402 (HTTP 402) APIs from inside Hermes with hard spend caps. Adds two tools and one hook: x402\_pay sends a request, reads the 402 Payment Required requirements, and if the price fits the caps signs an EIP-3009 TransferWithAuthorization in USDC on Base (gasless for the payer — the facilitator's fee payer covers the network fee), retries with the X-PAYMENT header and returns the payload plus the settlement receipt; x402\_receipts reads the payment ledger. ### [RowVerdict](https://github.com/NousResearch/hermes-agent/pull/153092) Tools · by demir-sama · open · 11-11 Review drafted CSV/XLSX/JSON contracts, fix quarantined rows through replay, and export source-bound receipts plus self-contained HTML audits locally. ### [PokeIt](https://github.com/NousResearch/hermes-agent/pull/153091) Desktop · by kbtbc · open · 11-11 Real-time Pokémon GO hunting inside Hermes, with a live Radar, nearby catches and raids, approved scans, and desktop alerts. ### [Retouch](https://github.com/NousResearch/hermes-agent/pull/153090) Desktop · by Lyfae · open · 11-11 Mark the parts of any image that need work in a Hermes Desktop pane, leave a note on each, and Hermes fixes only those areas. Each mark has its own thread; every edit is a new version, checked pixel by pixel and refused if anything outside its mark changed; only you can accept a mark. Tonal and privacy edits need no image provider, removals and replacements use the image\_generate provider you configured. Saves new files, never the original. ### [NyxWikiRecall](https://github.com/NousResearch/hermes-agent/pull/153084) Memory · by blacksmithFF · open · 11-11 Attach short wiki glimpses chosen by a decision model, and write a daily Markdown transcript of every turn. A local Markdown wiki (starter layout shipped) is scored per message through OpenRouter's Decisions API; matching glimpses are injected before the model answers, and wiki\_read opens the full article. Transcripts are written by the plugin itself into HERMES\_HOME/plugin-data/nyx-wiki-recall/ with no extra setup (transcript = yes and finalize = yes are the shipped values). Each day runs from finalize.after to finalize.after (default 02:00). Scheduled (cron) jobs are recorded under a Scheduled job heading; only subagent turns are skipped by default. A day is finished at finalize.after the next morning. The files are what a wiki is built from. Ships Clotho's wiki-archivist skill. Default provider=no runs as a hook beside any memory provider; provider=yes can serve as Hermes's memory provider when memory.provider also says so. ### [Signet Drive](https://github.com/NousResearch/hermes-agent/pull/153083) Tools · by prsnex · open · 11-11 Your agent's own encrypted drive, with dated restore points of its memory, which only you and your agent can open unless you invite someone. ### [Cortex Zero Adaptive](https://github.com/NousResearch/hermes-agent/pull/153078) Tools · by TheRealBrofessor · open · 11-11 Local 1B model directed adaptive browser automation with failure recovery. Implements the full OBSERVE→REASON→SELECT→EXECUTE→ANALYZE→ADAPT→VERIFY loop using llama-3.2-1b-cz-router (1.2B params, 128K context). The model selects actions from a structured catalog (baf-navigate, baf-click, baf-extract, baf-evaluate, baf-wait), encounters failures, analyzes them, and autonomously chooses alternative approaches — all within authorized engagement boundaries enforced by domain allowlists, operation allowlists, and fail-closed policy. Three mock demonstrations (navigation failure, element not found, policy blocked) and real BAF integration with 4/4 validation tests passing. Independent verification infrastructure separates model self-report from deterministic flag verification. Evidence logging to JSONL for audit trail. Reuses Cortex Zero Engine deterministic routing and BAF AutomationServer infrastructure. ### [cron-budget](https://github.com/NousResearch/hermes-agent/pull/153071) Automation · by heirakug · open · 11-11 Counts provider requests, not dollars, across every profile and holds cron jobs before they spend the requests kept for chat; chat is never blocked. Observe mode by default. ### [Deck Diff](https://github.com/NousResearch/hermes-agent/pull/153062) Tools · by Baophan00 · closed · 11-11 Content integrity for courseware: compare a source .pptx with the lesson data generated from it (canvas.json or a simple slide list) and report drift slide by slide — missing text, reworded lines, sentences pasted onto the wrong slide, missing/extra images, orphan lesson slides, and slides that never made it into the lesson. Slides are matched by content rather than index, so cover slides and skipped pages still line up, and asymmetry between a short divider and a full lesson page is penalised. Three local tools: deck\_slides (index a deck), deck\_diff (compare, JSON issues with severities) and deck\_report (self-contained HTML report with deck thumbnails and a MEDIA attachment). Pure Python standard library, no network calls, no API keys, read-only on the source deck; the only file it writes is the HTML report at the path the user asks for. ### [Sovereign & Stagehands](https://github.com/NousResearch/hermes-agent/pull/153058) Models · by Peetbog69 · open · 11-11 Typed backstage subagent delegation provider (scout, librarian, code\_monkey, adversary, briefer) and deterministic 48k/120k context-sanitation janitor that keep bulky tool gravel out of the frontier parent session. ### [WebGraft](https://github.com/NousResearch/hermes-agent/pull/153056) Web · by ElSerda · open · 11-11 Local web\_extract backend (trafilatura for pages, Hermes's own anydoc converter for PDF and Office documents, optional Crawl4AI browser view) that reports 404s and bot walls as errors; lists the pages an answer really used, from tool results, as clickable links; nudges the model to open pages rather than answer from snippets; reads a page's Markdown version when the site's llms.txt lists one, and returns pages that opt out of AI use (noai, TDM reservation) as errors; keeps long pages useful to small models (one URL per web\_extract call: a multi-URL call reads the first URL and its result lists the others; on a page longer than Hermes's per-page window, the passages about the question come first, nothing removed). Bundles a research skill, a setup wizard and an uninstall command. ### [SIEVE](https://github.com/NousResearch/hermes-agent/pull/153052) Automation · by VGFreakXBL · open · 11-11 User-owned notification boundary for Hermes Agent. Event classes route through tiers T0 to T5 under quiet hours. Routine alerts hold for a digest. Urgent ones escalate, and the message names both the requested tier and the arm that actually ran. ### [Show Your Work](https://github.com/NousResearch/hermes-agent/pull/153051) Tools · by kenpritchard · open · 11-11 An evidence board for a change's acceptance criteria: which requirements your agent actually tested, what the imported results say, whether they still match the code, and whether you signed off. Point it at the requirements you already have (a ticket, bug report or spec in Markdown) and it shows four separate statuses per requirement, never merged into one green check: coverage (is a test linked?), outcome (latest imported JUnit run), freshness (tracked files changed since?) and human review. For review, click a requirement and then a linked test to see that test's source and its own diff against the base branch (new, changed or unchanged), plus the change under review. The agent links tests and imports reports from wherever they ran (terminal, Docker, CI); it has no tool to accept or reject. Sign-off comes from the Desktop page, /syw or hermes syw. Five tools (syw\_board, syw\_link, syw\_import, syw\_run\_detail, syw\_test\_evidence), a /syw command, a hermes syw CLI, a Desktop page and a workflow skill. ### [Mandate](https://github.com/NousResearch/hermes-agent/pull/153050) Automation · by JoeProAI · open · 11-11 Learned least-privilege for cron jobs. Observes what a job does, proposes an argument-level mandate, and after human approval blocks out-of-mandate tool calls. ### [Model Forge Desktop](https://github.com/NousResearch/hermes-agent/pull/153040) Desktop · by mr-pewpy · merged · 11-11 Model Forge cockpit pane for Hermes Desktop — compute targets, live engines, loaded models and hardware at a glance. Companion UI for the Model Forge plugin. ### [Model Forge Image Backend](https://github.com/NousResearch/hermes-agent/pull/153040) Models · by mr-pewpy · open · 11-11 Model Forge local image-generation backend — routes Hermes image\_generate to any OpenAI-compatible images API (weights broker, vLLM-Omni, SGLang Diffusion, ComfyUI). Companion package for the Model Forge plugin. ### [Model Forge Providers](https://github.com/NousResearch/hermes-agent/pull/153040) Models · by mr-pewpy · merged · 11-11 Local engine model-provider profiles for Model Forge (Strata, Project Maya) — run Hermes chat on local MoE engines and make their models selectable in /model. Companion package for the Model Forge plugin. ### [Model Forge Video Backend](https://github.com/NousResearch/hermes-agent/pull/153040) Models · by mr-pewpy · merged · 11-11 Model Forge local video-generation backend — routes Hermes video\_generate to any OpenAI-compatible video job API (weights broker, engines with /v1/videos). Companion package for the Model Forge plugin. ### [Model Forge](https://github.com/NousResearch/hermes-agent/pull/153040) Tools · by mr-pewpy · open · 11-11 Local models for every modality — chat, vision, OCR, image generation and editing, video, 3D meshes, music, SFX and voice. Model Forge finds your hardware (local GPU or LAN compute targets), shows which models fit, drives the right inference engine (llama.cpp, Strata, Ollama, vLLM, ComfyUI, weights broker and more), pulls the model, serves it, and wires it into Hermes as a provider — so chat, image\_generate, video\_generate, vision and TTS run on your own hardware with no account and no API key. Six tools: model\_forge\_status (targets, VRAM/RAM, engines, loaded models), model\_forge\_browse (per-modality model catalog with hardware-fit verdicts), model\_forge\_target\_add (register LAN compute targets), model\_forge\_install (pull models through engine-native paths), model\_forge\_activate (wire a served model into Hermes), model\_forge\_generate\_media (image, 3D, music, SFX and voice jobs against any local engine). ### [Glide](https://github.com/NousResearch/hermes-agent/pull/153039) Desktop · by kvnloo · open · 11-11 Kevin Rajan's Hermes frontend for Tern, built on alt-glitch/hermes-tern (MIT) and carrying about 890 commits of its own. One install gives \`hermes tern\`, which draws the chat as native Tern UI, plus \`/glide\`, a guided setup that makes Hermes learn from you and run more of its work locally. Added over the original: charts from Markdown tables, OpenUI and ephemeral-ui answer cards, 3D artifacts beside the chat, written artifacts and session links, Tern browser blocks, image delivery across hosts, LSP diagnostics cards, plan review with a todo HUD, native /btw side questions and read-only MCP/SSH report sheets, live wait tiles for subagents, fork-a-session panes, /rewind, /undo and /tree, native /settings and a /models provider hub, a Vim composer, a command palette and shortcut preferences, queue-or-steer busy input, opt-in live voice with helper lanes beside the thinking orb (role, model and real state from a local lane broker), a work dock, a thinking-orbs indicator, frontend channels (\`hermes tern --channel\`), and tool cards that report refusals, truncation and partial failures instead of hiding them. Glide guards two scarce resources: your attention and the agent's context window. \`/glide focus\` shows your current phase, next deep-work window and usual stop time, learned from your own local session timestamps (or from z0intelligence's \`circadian\` and a loopback Flowkit flow estimate when installed); a deep-work block holds non-critical agent notifications and approval asks for one digest at the end while critical ones still pass, and one wind-down card saves a resume note for the next session. \`/glide\` (also \`hermes glide onboard\`) is a chain of approve cards: detect the machine and local runtimes (and z0intelligence if installed); build a corrections memory from your past local sessions (your messages that start "no,", "actually," or "correction:") and quote up to three matching corrections into each new turn through a pre\_llm\_call hook, using deterministic BM25 with no embeddings or model calls; a skill router catalog; a measured CPU hill climb whose referee says KEEP, REVERT or INCONCLUSIVE; offload of small tasks to an installed Ollama model with estimated savings; and an optional Taste board pane beside the chat. ### [Formwork](https://github.com/NousResearch/hermes-agent/pull/153038) Desktop · by ladyFaye1998 · open · 11-11 Complete a packet of fillable PDFs inside Hermes Desktop. See each suggested answer on the original form, open its exact source page and quote, and review it before export. Enter a missing fact once and choose which matching fields should reuse it. Save editable PDFs and a review record in one ZIP. ### [ceo-pack](https://github.com/NousResearch/hermes-agent/pull/153027) Tools · by tradersurfer93 · open · 11-11 CEO Pack — a CEO orchestration layer for Hermes. Brief, route, plan, assign, call, verify, log, with Hermes as COO. One owner and one due date per task, Type 1 / Type 2 decision gating, credential-safe tool calls, and owner approval for external actions. Free model default is Nemotron 3 Ultra free, with GLM Flash free fallback and paid Claude Sonnet 5.5 / Opus 5.5 overrides the user chooses. The plugin returns model choices; it does not call a model itself (stdlib-only, catalog-safe). ### [Imagination Studio](https://github.com/NousResearch/hermes-agent/pull/153026) Tools · by drowzeys · open · 11-11 Make pictures, video clips with sound, voice lines, songs and sound effects on your own machine with Imagination Studio (by keys + Wasserman): FLUX 3 Klein / Qwen-Image, MiniMax H3 / FastH3, Fish Audio Drama 3, MiniMax Music 3 and Stable Audio Open, rendered in the Studio's own queue and delivered to the chat as attachments. ### [Home 3D Workshop](https://github.com/NousResearch/hermes-agent/pull/153024) Automation · by Wijt · open · 11-11 Ask. It's in the box. A 3D printer as a member of the group chat (the persona is Mıstık by default; the name and the en/tr message language are settings, English by default). People in a family or team chat ask in plain words, often by voice note; the agent finds MakerWorld models through Bambuddy, scales one, or designs a part in OpenSCAD (it writes the code, renders it, looks at the preview pictures and keeps every version), slices, and queues with manual start. A print starts only with the owner's typed approval (never from a voice note); when someone else asks, the watcher passes the request to the owner with the exact command. A zero-token cron watcher follows the print; the person who asked gets a one-line start notice and, at the end, a line with a photo in the chat the request came from (groups setting: family or team tone, per-group language), and the owner gets a technical note. Notices go through a saved outbox with retries. Finished parts are tracked in a done box. The owner gets one Workshop screen in Hermes Desktop (approval card with the part in 3D, Start, Pause and Stop, queue, done box, messages, filament, setup check, design versions). Each moment a person had to step in is logged, and repeats become upgrade proposals (for example a printed tilted base for auto-eject). Works with a Bambu Lab printer managed by Bambuddy, or with a built-in simulated printer (hermes home3d demo). ### [Hermes Parallax](https://github.com/NousResearch/hermes-agent/pull/153023) Tools · by kruschdev · open · 11-11 Multi-perspective fact-vs-frame lens, blind code security taste-test, and rubric-gated DPO/SFT dataset forge ### [breadcrumbs](https://github.com/NousResearch/hermes-agent/pull/153020) Memory · by jnorthrup · open · 11-11 Memory and a context engine in which every line has an address. Each turn, tool step and tool result is a crumb in a content-addressed, append-only store (CRC32C ids, Merkle commits). One tool, crumb(id, level), shows any memory from a one-line handle (0) to its exact bytes (1). crumb\_find searches with BM25: no vectors, no model. Each recall, find, open and fold writes a record of what the model saw. With context.engine breadcrumbs, a long chat folds into a layer of crumbs with no model call, and the model opens any folded turn by its id. /island gives each codebase its own profile. A read-only explorer shows the sessions, the zoom and the context of each turn. ### [IDEA CUT](https://github.com/NousResearch/hermes-agent/pull/153017) Desktop · by ROCK · open · 11-11 Turns a chat brief into an editable, art-directed video: seven agent tools capture and record a public page, write typed scenes, revise them and export H.264 MP4 locally with Chromium and FFmpeg. Capture goes through a loopback validating proxy that refuses private/loopback/metadata targets. Optional narration is sent to the user's configured TTS provider; the live preview is silent. ### [Caduceus Governance](https://github.com/NousResearch/hermes-agent/pull/153015) Tools · by Epistetechnician · open · 11-11 Deterministic, model-free admission control for what Hermes writes into its own skills: content lanes, operator-owned executable checks, a provenance lane for writes made after untrusted web/browser/MCP content, a tamper-evident ledger, and operator release/rollback. Observe by default. ### [NousViz — Agentic Analytics](https://github.com/NousResearch/hermes-agent/pull/153010) Tools · by nousviz · open · 11-11 Turn Hermes into a data analyst over your self-hosted NousViz — discover data with a live catalog, query it in natural language (read-only SQL across Postgres/ClickHouse/MySQL), and build dashboards, alerts and annotations by conversation. ### [Nightshift](https://github.com/NousResearch/hermes-agent/pull/153009) Tools · by Pavilion-devs · open · 11-11 Hermes takes the night shift on your ML training runs: Nightshift watches a run's metrics (stdout, JSONL, CSV or TensorBoard), catches NaN, loss spikes, divergence, out-of-memory crashes and stalls, recovers the run within a playbook you approved (roll back to the last healthy checkpoint with a lower learning rate, or restart with half the batch), and pushes the chart to your phone. Desktop pane, dashboard tab, /nightshift, hermes nightshift and a morning-report blueprint. Zero runtime dependencies. ### [preflight](https://github.com/NousResearch/hermes-agent/pull/153006) Tools · by soyluisarca · open · 11-11 Approval dry-run gate: runs every terminal command in a disposable copy of the workspace (reflink/copy, no Docker), shows the exact diff, and applies only what the user approves. ### [Hermes Eye Memory](https://github.com/NousResearch/hermes-agent/pull/153003) Memory · by isthatananya · open · 11-11 Local visual memory for Hermes: index a photo library, search by meaning, build albums, cull duplicates, extract receipts, and recall what the camera saw. ### [Hermes Studio — local video workflow](https://github.com/NousResearch/hermes-agent/pull/153662) Tools · by Morkeeth · open · 11-11 Local camera takes to reviewed editable portrait Shorts, with explicit media-library discovery and an optional SSH conversation queue. Uses a separately installed noncommercial native editor; no paid generation or automatic public posting. ### [Stoa](https://github.com/NousResearch/hermes-agent/pull/153000) Tools · by jarrett-del · open · 11-11 A lightweight CRM you talk to: map a market, find the leads, run the pipeline. Send it a list in any form and approve what it would add or change before anything is saved. Market maps that check every company against your requirements with evidence, show who holds the market and who to call first, export as a shareable page and become leads in one click; a page and a live pane in the Desktop app and Webapp, with a funnel, a drag-and-drop board, a dock to talk to it, and cards and maps drawn in the conversation; a web dashboard tab; a SQLite pipeline, playbooks that fit it to any niche, Slack cards whose buttons move a stage at once and hand the writing to Hermes, and meeting notes that become next steps. ### [Invezgo](https://github.com/NousResearch/hermes-agent/pull/152997) Tools · by Invezgo · merged · 11-11 Indonesia Stock Exchange (IDX/BEI) market data from the Invezgo API: company profiles, price and indicator charts, intraday and order book, running trade, broker summary and inventory, broker and sector stalker, sector rotation, shareholder composition (KSEI, \>1%, \>5%, insiders), financial statements, key stats, corporate calendar, news, disclosures and a formula screener, plus read access to the user's own Invezgo portfolio, watchlists, journal and trades. Tools are generated from the bundled OpenAPI spec (spec.json) at load time. ### [Plugin Scout](https://github.com/NousResearch/hermes-agent/pull/152993) General · by bscharenberg · open · 11-10 A scout goes ahead, reports what's useful, and warns about trouble. Plugin Scout finds the plugins that fit how you use Hermes (each pick with its evidence), and checks: it tells you what a plugin does with your data before you install it, warns when a plugin's fine print does not fit your settings, catches plugins that clash and explains what to do. Quiet weekly note, plain-English requests, and a dashboard tab. Recommends only; never installs or enables anything. ### [Chalk](https://github.com/NousResearch/hermes-agent/pull/152992) Desktop · by Nnadijoshuac · open · 11-10 Hermes teaches on a chalkboard. Ask it to explain something and Hermes Desktop draws the lesson inline in the chat, stroke by stroke in chalk, while the narration plays; replay it or save it as an MP4. The agent's chalk\_teach tool sends only structure (flow, list, compare, equation, graph, cycle, tree, note, circle/underline, clear) and the board lays itself out, so small models produce clean boards; full boards are wiped by an eraser and continue on a fresh one. Other surfaces receive the narration as one audio file. ### [Survey Desk](https://github.com/NousResearch/hermes-agent/pull/152991) Tools · by ferzerz5-lab · open · 11-10 Exact land-survey computation for Hermes: closed traverses with misclosure and Bowditch/transit adjustment, area cross-checked by two independent methods, blunder-leg detection when a traverse fails tolerance, PROJ datum conversion (Minna, Nigeria belts, UTM, any EPSG) with the published datum-shift accuracy, and plan drawings exported as PNG, PDF, DXF, KML and GeoJSON, driven from a /survey field-book command or the agent. ### [ctrl+all](https://github.com/NousResearch/hermes-agent/pull/152990) Tools · by namangoyal3 · open · 11-10 Drive ctrl+all on your Mac from any Hermes conversation: launch and steer multiplayer coding-agent chats on the engines you already pay for (Claude Code, Gemini, opencode), read transcripts and the team inbox, and review git status and diffs. The plugin calls the ctrl+all daemon's authenticated loopback RPC; the daemon owns every permission gate (plan / ask / auto-edit / full-auto modes, engine boundaries, project fencing). ### [Benchline](https://github.com/NousResearch/hermes-agent/pull/152987) Tools · by ShalyX · merged · 11-10 Read-only preflight for a selected numeric-result CSV: checks structure, duplicate keys, numeric formatting, and exact analyte/unit-map matches. No clinical interpretation; the selected path, unit-map inputs, profile-source label, and bounded QC report are exposed to the configured model/provider. ### [Local Voice Notes](https://github.com/NousResearch/hermes-agent/pull/152986) Voice · by soilgrowerx · open · 11-10 Voice memos become text on the machine that recorded them: no API key, no account, and no audio leaving the host. Capture from PipeWire or ffmpeg, run faster-whisper locally, and the transcript arrives as a note. Four tools (transcribe\_audio, voice\_record\_start, voice\_record\_stop, transcribe\_last) and two hooks: pre\_gateway\_dispatch turns an inbound voice message into text before the model sees it, pre\_llm\_call attaches the pending transcript to the turn. Ships model management (list supported Whisper sizes with disk cost and cache status, download explicitly) and a live capture path with a silence gate so ambient room tone is never transcribed. ### [Pocketbook](https://github.com/NousResearch/hermes-agent/pull/152984) Desktop · by MaanavD · open · 11-10 Personal finance in Hermes: a Finance tab, bank/account controls, spending, budgets, net worth, investments, goals, transaction categories, CSV imports and 15 money tools. ### [claude-pacer](https://github.com/NousResearch/hermes-agent/pull/152981) Automation · by Rubenkivilang · merged · 11-10 For power users running heavy automation in Hermes on a Claude Pro/Max subscription: claude-pacer paces agents and cron jobs so the weekly limit lasts to the reset, keeps a reserve (20% by default), and spends the leftovers on spare jobs before it. Modes skip / light / normal / heavy via the claude\_pace tool, /pace, \`hermes pace\` and a \`hermes pace gate --priority vital\|normal\|optional\|spare\` cron gate: an over-budget run is skipped before the model is called; a vital job always runs (lightly in skip mode). ### [Hermes Visual Memory](https://github.com/NousResearch/hermes-agent/pull/152980) Memory · by majorminorlabs · open · 11-10 Your agent remembers what a camera saw on your desk: "what's in front of you?", "what did I take away?", "where did I last see my keys?", "what was here yesterday?". Built for a used Echo Show running Hermes Eye, with before/after pictures on its screen. Vision goes through Hermes's own model routing (its own "Visual memory" slot) or a local OpenAI-compatible model such as Qwen 3.6 in Ollama. ### [Lydite](https://github.com/NousResearch/hermes-agent/pull/152979) Automation · by aoricloud · open · 11-10 Hand a GitHub issue and the command that proves it done to Lydite from the chat. Sandboxed Claude Code workers do the job, Lydite replays your proof itself, and a pull request comes back with a receipt (cost, workers, retries, the commit that was reviewed and replayed, and what was not replayed). Also hands Hermes the lessons Lydite keeps for a repository, read-only, and has Lydite check a plan file before a launch. ### [Loshify](https://github.com/NousResearch/hermes-agent/pull/152975) Desktop · by iHeavyShoes · merged · 11-10 A Spotify curator that knows you: it learns your taste from your listening and your reactions, builds playlists with a reason for every track, and remembers what you loved and what you never want again. Desktop remote with 👎 👌 ❤️, rotations, and a chat curator. Connect in about two minutes; the Desktop card walks you through it one step at a time with copy buttons, and the agent walks the same steps in chat. First install with hermes plugins install iHeavyShoes/loshify --enable (once listed, hermes plugins install loshify). Then in Hermes Desktop open Capabilities → Plugins and turn on Loshify's Desktop switch (the Desktop half starts off and config.yaml does not turn it on). (1) Open developer.spotify.com/dashboard and press Create app. (2) App name Loshify, Redirect URI http://127.0.0.1:43828/callback, tick only Web API, Save. Your own key to Spotify; 127.0.0.1 means this computer, nothing leaves it. (3) Add yourself under User Management (new Spotify apps only let in listed accounts). (4) Paste the Client ID into the Loshify card; it is saved in the plugin's setting. (5) Run hermes auth add loshify:spotify once in a terminal (Windows: Start → PowerShell) and press Agree. (6) The card checks with one real call to Spotify and turns green. ### [hey-pcb](https://github.com/NousResearch/hermes-agent/pull/152974) Tools · by imperator-clawdius · merged · 11-10 Agent-driven PCB engineering through HeyPCB's first-party MCP server: design real KiCad boards from a text brief, instant fabrication estimates, and production tracking, with a workflow skill bundled. Portable Agent Plugins v1 package (mcp.json + skills/), no plugin code. The server is declared untrusted with per-tool costs and a per-chat spend limit, so Hermes names the cost and asks before any call that can spend (design reads stay free). The HEYPCB\_API\_KEY is forwarded from the user's environment and never stored in plugin files. ### [permission-tiers](https://github.com/NousResearch/hermes-agent/pull/152974) Tools · by imperator-clawdius · open · 11-10 Host-enforced tiered tool permissions: assign allow/ask/block tiers to tool-name patterns (fnmatch globs), and the plugin's pre\_tool\_call hook vetoes or escalates matching calls before they execute, while post\_tool\_call keeps a local SQLite audit ledger. Two tools manage it: permission\_tiers\_report (rules + ledger) and permission\_tiers\_set (add/update/ remove). Implements the per-tool permission-levels feature requested in issue #527 without touching Hermes core — public hooks only. Safe defaults: shell and MCP tools ask, everything else keeps Hermes' existing behaviour; hook errors fail open; unattended runs fail closed on ask escalations (the plugin never auto-approves). ### [AgentLine](https://github.com/NousResearch/hermes-agent/pull/152976) Platform · by prakhar-raj-031104 · open · 11-10 A real phone number for Hermes: place calls that a hosted voice carries out and get the transcript back, send and read SMS, text your agent like any other chat, and let a live caller's question reach Hermes mid-call, with the answer spoken on the call or texted if it takes too long. ### [detonator](https://github.com/NousResearch/hermes-agent/pull/152971) Tools · by shreyasudupi · open · 11-10 Dynamic sandbox for plugins you did not write. After you approve the run, /detonate fetches a plugin at an exact commit and loads it inside a disposable Docker container (no network, read-only, non-root, all capabilities dropped) with a throwaway Hermes home full of decoy keys, then drives its hooks and tools and reports what it actually did: decoy secrets read, network attempts, shell commands, persistence, prompt injection, tool poisoning, monkeypatching. Deterministic rules with a plain-language reason and MITRE ATT&CK / ATLAS IDs per finding; advisory only, a clean run is evidence, not proof. ### [Wave \| Bots](https://github.com/NousResearch/hermes-agent/pull/152968) Tools · by xBuildyTeam · open · 11-10 Verified AI bots for Hermes, by Wave OS. Browse the official Wave bots from the bots.theta store (Checker, Tally, Planner, Maker and more); each one is checked on the Theta blockchain on your computer (name record, listing, rules, mascot and every knowledge card, fingerprint by fingerprint) before it's marked Verified. Add one and it becomes its own Hermes bot (a wave-\ profile in your Bots list, with its mascot), answering from its checked cards on your own AI model with code, terminal, file and web tools turned off, re-checked daily. An optional Wave OS bot key lets you ask the real bot on Wave, with Hermes approval for every question. ### [video-xray](https://github.com/NousResearch/hermes-agent/pull/152967) Desktop · by LuuMoss · open · 11-10 Drop a video file into the chat composer and get a structured report: segment-wise frame observation by whatever vision-capable chat model you already configured (local llama.cpp/Strata or cloud), timestamped dialogue from R2T2, Whisper, or any OpenAI-compatible cloud ASR, and a frame gallery of the key moments inside the job detail. Exports the report as Markdown or as a MiniMax-H3 integrated multimodal description prompt. Ships the R2T2 engine worker source plus an optional Whisper worker; the heavy R2T2 runtime binaries are user-supplied, nothing else is downloaded. Spawns ffmpeg/ffprobe and local ASR worker subprocesses; network calls go only to the user's own configured inference and cloud-ASR endpoints, never to third-party services; background analysis jobs run as worker threads tracked in a local job ledger. The panel is pure @hermes/plugin-sdk React and user-visible text goes through ctx.i18n. ### [Board Insights](https://github.com/NousResearch/hermes-agent/pull/152966) Desktop · by kwpings · open · 11-10 Read-only analytics page for the Hermes Kanban boards, in Desktop. Two views over every board: List shows what is in flight, what is waiting on a person and every unfinished card in one filterable list; Dashboard shows summary counters, completed over time with a moving average, workload by status, net flow, tasks by department, cycle and run medians, stale runs and a per-assignee table, driven by one date range (today, 7, 30 or a custom window up to 90 days). Spaces are named board sets saved locally so the page can read one project or all of them. ### [pwn-board](https://github.com/NousResearch/hermes-agent/pull/152960) Tools · by t4lus-sec · merged · 11-10 Pwn boards for Hermes: engagements and problems with a known origin, a testable goal and no known path (a foothold to reach, a privilege to escalate, a flag to capture; also flaky failures and regressions), and the workflow that takes a finished one to complete. The agent keeps a frontier of intents rooted at the board's origin, claims them (subagents too), and records findings that must cite real tool output: while a board is open every tool result carries an evidence id, and a finding citing an id no tool returned is refused. It never scores itself: the goal test is fixed at open, run through the terminal tool under normal approvals, and the plugin writes the verdict. A finished board is written up with the files as delivered, goes to the operator's review (notes the agent must answer before the next round), then to verification: an independent verifier subagent and the agent check it separately, the verifier's report held from the agent until it files its own, and the agent reconciles them for the operator to accept or send back. /pwnboard and a Desktop pane drive all of it. The board carries no offensive tooling of its own. Provenance — the fact-intent-hint board model originates in Cairn (https://github.com/oritera/Cairn, AGPL-3.0); this is an independent implementation for Hermes and contains no Cairn source. ### [Branchpoint](https://github.com/NousResearch/hermes-agent/pull/152959) Tools · by Sahil-SS9 · open · 11-10 Explore the same task with 2-8 independent agent plans (default 4), optionally on different models. Build selected routes in isolated Git worktrees (default top 2), compare them with three blind reviewers and an evidence-linked decision record, then adopt or converge their ideas into another candidate. Includes a loopback live view, decision log, rewind anchors and reversible adoption. ### [Anchor](https://github.com/NousResearch/hermes-agent/pull/152955) General · by Aayu095 · open · 11-10 Context engine that pins decisions, amounts and dates verbatim before compressing: facts are detected deterministically (zero model calls), re-issued word-for-word in a carrier message on every compression pass, and the unpinned middle folds into a size-capped extractive digest. ### [Secrets Vault](https://github.com/NousResearch/hermes-agent/pull/152954) Tools · by Sky · open · 11-10 Field-level secrets protection: encrypt sensitive fields in a database at rest, redact them automatically when the agent reads the data, and reveal a single value only with human approval. ### [Session Receipts: session flight recorder + weekly process digest](https://github.com/NousResearch/hermes-agent/pull/152953) Other · by Ryan O'Mara · open · 11-10 Session Receipts turns a Hermes session, a date range, or a kanban card's evidence chain into one self-contained offline HTML document: the original ask, every tool call as a timed node, failures and retries exactly as recorded (never collapsed into the success that followed), per-model token spend, and the cost Hermes recorded, with a reconciliation delta. A client theme renders the same data as a paper document for sharing; \`--redact\` writes a share-safe copy with a redaction manifest (secrets, emails, user paths, IPs, phones, plus a user word list); and \`receipts rollup --since 7d\` aggregates a window into a deterministic weekly digest: failure taxonomy, cost by model and card, slowest tools by p50/p95, retry hotspots, week-over-week deltas. Standard library only; reads the local session store and kanban boards read-only; no network, no credentials, no new dependencies. ### [Pepper for Hermes](https://github.com/NousResearch/hermes-agent/pull/152949) Tools · by Akshay-Scalelabs · merged · 11-10 Say “parlay this” to find a ParlayTix ticket or preview two outcome rights for one you own. Resume the same split across chats, review a listing for the other side in your wallet, or set a budget watch and ask for updates. ### [Local Fleet](https://github.com/NousResearch/hermes-agent/pull/152948) Models · by humedini · open · 11-10 Your local inference fleet, operated from Hermes Desktop: routing across your own boxes (DGX Spark / GB10, Ollama, vLLM, LiteLLM, clusters behind one endpoint). Probes each box, makes the least busy healthy one the primary model, falls back to smaller models on smaller boxes, then to your cloud model, and switches back when a box recovers. Why-this-box inspector, flight recorder, outage rehearsal, drift detection, readiness, savings with its working, doctor. Optional: subagents across boxes (delegation backend fleet), side tasks on the fleet (auxiliary.\), keep the routed model loaded, private inference (primary, fallback chain, side tasks and subagents held to your boxes, failing rather than using cloud; not air-gapping). Writes model.\*, fallback\_providers, and on request delegation.backend and auxiliary.\; private on removes non-local fallback entries and puts them back on off; hermes fleet restore puts the primary back. Network: HTTP only to the boxes you register. Observer hooks record per-call endpoint, model, tokens and duration (never content) in its data dir. Never moves a primary that holds an inline API key and never copies a key into its data. Read-only tools plus a model pull from the Ollama library that Hermes asks you to approve; no remote command execution. ### [account-switcher](https://github.com/NousResearch/hermes-agent/pull/152944) Tools · by hendkai · open · 11-10 Switch between pooled provider accounts (e.g. multiple Claude logins) from chat, CLI, or the agent — with priorities, current-account marker, plan mode, and optional usage percentages. ### [Zotero Library](https://github.com/NousResearch/hermes-agent/pull/152942) Tools · by gerdosi · open · 11-10 Search and read the papers in your own Zotero library before going to the web: a local, read-only full-text index (SQLite FTS5, BM25) built from a copy of Zotero's database, so it works with Zotero open or closed. Four tools (zotero\_search with a strong/thin/none coverage verdict, zotero\_read for passages, your own highlights and notes and a citation line, zotero\_status, zotero\_sync), a bundled skill that sends literature questions to the library first, and a \`hermes zotero\` command (doctor, snapshot, sync). ### [Hermes SafeGuard](https://github.com/NousResearch/hermes-agent/pull/152939) Tools · by robberdom · open · 11-10 Deterministic security middleware, EVM simulation guard, and prompt injection defense for on-chain agent actions. ### [Mission Control](https://github.com/NousResearch/hermes-agent/pull/152934) Desktop · by rohitdanda · open · 11-10 Live mission-control dashboard for Hermes Desktop — tool waterfall, parallel subagent cards, token and estimated-cost meters with a prompt-cache badge, plus budget and possible-loop alerts, streamed in real time. Read-only; nothing leaves the machine. ### [Coursewright](https://github.com/NousResearch/hermes-agent/pull/152933) Tools · by NehuenD · open · 11-10 \`/course \\` turns a subject into one self-contained HTML course at three depths, novice, intermediate and advanced, switched in place without a reload. It researches before it writes: it searches through whatever web backend the host has configured, reads the pages it finds, collects videos that carry captions, and lists every source, dead end and filtered result in a footer. The model never writes HTML; it fills a validated spec and a renderer draws the page. Figures are drawn from the course's own outline — layered diagrams, sequence diagrams, small charts — and every section ends with an exercise that answers back. ### [Shorguard — post-quantum crypto guard](https://github.com/NousResearch/hermes-agent/pull/152930) Tools · by Akanimoh12 · merged · 11-10 Post-quantum crypto guard. Stops the agent from writing quantum-vulnerable public-key crypto (RSA, ECDSA, ECDH, EdDSA, X25519, DH, DSA) before it lands: pre\_tool\_call checks write\_file, patch, terminal and execute\_code and, in the default block mode, refuses calls that ADD classical crypto, returning the NIST replacement (ML-KEM / ML-DSA / SLH-DSA) with working cryptography\>=48 code so the agent rewrites it (approve mode routes to Hermes's approval gate instead; warn mode only annotates). Tools shorguard\_scan / shorguard\_explain, /shorguard (status, mode, scan, explain) and \`hermes shorguard scan\` give a post-quantum readiness grade with text, JSON or SARIF output and CI exit codes. 52 rules across Python, JS/TS, Go, JVM, Rust, .NET, shell/CI, TLS config, PEM keys and JWT algorithms, mapped to NIST IR 8547 (deprecated after 2030, disallowed after 2035). ### [Ultimate Dashboard](https://github.com/NousResearch/hermes-agent/pull/152928) Desktop · by Yazeed-A-H · merged · 11-10 A customizable dashboard page for Hermes Desktop and the Webapp, with 15 cards: approvals you can answer, running sessions, scheduled jobs with a 7-day timeline, job results, plan usage, spend, tool activity, health, services, machine stats, logs, agent memory, skills, files made and Workers, plus Quick launch shortcuts. Arrange cards into views, save templates, and the layout syncs between Desktop and the Webapp. English and Arabic (RTL); follows the Hermes theme. ### [Coursebook](https://github.com/NousResearch/hermes-agent/pull/152920) Desktop · by DeseretSaint · open · 11-10 A native Hermes Desktop learning workspace where a lesson is governed state rather than chat: one objective at a time, typed checks, bounded grading, transactional transitions, inspectable evidence, and sourced visuals and audio, plus parent-prepared lesson packs. ### [Look](https://github.com/NousResearch/hermes-agent/pull/152918) Web · by SilentHacks · open · 11-10 Ask about pages and use guarded Chrome actions. Apple Silicon macOS beta with broad debugger access and optional local voice and eyes. ### [Culprit](https://github.com/NousResearch/hermes-agent/pull/152914) Tools · by Rizal · open · 11-10 git bisect for your agent. Records a snapshot after every tool call that changes a file, then binary-searches those snapshots with your test command to name the exact step that broke it, with its diff, and can undo only that step while keeping every later change. Works as /culprit, as \`hermes culprit\`, and as tools the agent can call. ### [GMI NEXUS](https://github.com/NousResearch/hermes-agent/pull/152911) Models · by Joystonm · open · 11-10 Adaptive intelligence layer for Hermes Agent over the live GMI Cloud model catalogue. A goal is understood, planned into steps, and each step is routed to the model that suits it from the catalogue read live at runtime, so a retired model stops being routable and a new one becomes routable without a code change. Results are evaluated against each step's own criteria, a failure is recovered on a different model with bounded retries, the verified outcome feeds the routing statistics, and the report says what ran, what it cost and what failed. Ships a vision module for screenshots and before/after comparison, a benchmarking arena whose results feed the router, a Dev Agent that writes and tests code through Hermes' own tools, and a creative engine for image, video and audio generation and editing. Also ships a web dashboard tab. No third-party dependencies; the key is read through Hermes' secret scope; a per-run ceiling, a rolling daily ceiling and an approval threshold are enforced before any request is submitted. ### [Done only when your tests pass](https://github.com/NousResearch/hermes-agent/pull/152908) Tools · by vijaygopalbalasa · open · 11-10 Hermes's done doesn't count until your tests pass. Type /necro freeze in your project: it finds your tests and the command that runs them, writes necro.toml, locks the test files and the settings that choose what runs by hash, and runs them once to say where they stand. When Hermes says it is finished, the plugin runs them itself at the verify step and sends Hermes back with the failing test named; the final reply says when Hermes was sent back and ends with a receipt anyone can re-check, also in a fresh clone (/necro verify: CONFIRMED, MISMATCH or SUPERSEDED). Edits to locked tests are blocked, and any change that gets through fails the check. Only a person can freeze: /necro is a slash command the agent cannot type, and from its shell freezing refuses. A prompt or skill can only ask the model to test its work; these hooks run whether it asks or not. ### [Hermes Whiteboard](https://github.com/NousResearch/hermes-agent/pull/152907) Desktop · by Adolanium · merged · 11-10 An agentic whiteboard. Say what you're planning and Hermes lays it out as a route, columns or a map, with hand-drawn stickers. Add rough notes and Hermes proposes groups, owners and action items that you accept one by one and can take back. Hand an action item to Hermes and an agent does it on the board and delivers the result next to the note with its sources, or run the whole route and get a brief at the end. Select anything and ask about it. Once you allow it, any chat can read your boards and propose changes. ### [Sochen](https://github.com/NousResearch/hermes-agent/pull/152906) Voice · by Sochen · merged · 11-10 Give Hermes a phone with nothing to set up: Sochen (sochen.dev), a cloud agent for any open-source agent, phones a US business for you and brings back the answer, with no Twilio account, number or voice keys of your own. When the business asks something only you can answer, the question comes to your Hermes chat while they hold the line, and your reply goes straight back into the call. It also runs long research in the cloud and handles email and calendar jobs that wait for your OK, and Hermes is told when each job finishes. Every account gets 100 free Sochen API credits, spread over any number of jobs; after that, packs or a monthly API plan (sochen.dev/pricing#api). Free GST invoices, rent receipts for HRA and UPI bill splits with no account, and your Sochen desk as a page in Hermes Desktop. ### [xscapes](https://github.com/NousResearch/hermes-agent/pull/152904) General · by donlucasx · open · 11-10 Living ASCII scenes under Hermes in the terminal that show what the agent is doing while you wait: the wind, the sea or the rain rises with the work, a thought cloud means it is thinking, owlets, crablets or kittens arrive for subagents, the moon sinks as the context fills, and the companion knocks with a sound when Hermes finishes or needs you (a question, a password prompt, an approval). Forwards fifteen lifecycle hooks to the xscapes CLI (\`xscapes hook \ hermes\`, stdin JSON, never waited on, never blocks); needs the xscapes binary. No tools, middleware or env vars; no network. ### [Project Command Center](https://github.com/NousResearch/hermes-agent/pull/152899) Desktop · by NyllRE · open · 11-10 A board for every Hermes Project: lane, live sessions, token spend, git branch and GitHub issues and pull requests in one place, with a detail page per project. Click an issue and a session opens inside that project's folder with the issue number, title and URL already in the composer, unsent. Branch rows open a session in a worktree or after a checkout, and each session carries its own lane, so a PR climbing to review is visible where the work happens. Includes a GitHub reader with full issue and PR modals, comments, status filters and a Zed-style commit graph. ### [BrowserSolver](https://github.com/NousResearch/hermes-agent/pull/152894) Web · by browsersolver · merged · 11-10 Fetch web pages behind Cloudflare, DataDome and similar protections for the built-in web\_extract tool, through the BrowserSolver web unlocker API. Extract only; pair it with a search backend for web\_search. ### [CaptureKit](https://github.com/NousResearch/hermes-agent/pull/152893) Web · by capturekitdev · merged · 11-10 Page content as clean Markdown for the built-in web\_extract tool, through the CaptureKit Content API. Extract only; pair it with a search backend for web\_search. ### [Autom](https://github.com/NousResearch/hermes-agent/pull/152892) Web · by automdev · merged · 11-10 Live Google Search results for the built-in web\_search tool, through the Autom SERP API. Search only; pair it with an extract backend for web\_extract. ### [Piloterr](https://github.com/NousResearch/hermes-agent/pull/152891) Web · by piloterr · merged · 11-10 Fetch web pages for the built-in web\_extract tool through the Piloterr web scraping API, with optional JavaScript rendering or anti-bot unblocking. Extract only; pair it with a search backend for web\_search. ### [Veille](https://github.com/NousResearch/hermes-agent/pull/152890) Tools · by veilleio · merged · 11-10 Tools to check emails, domains, IPs, phone numbers, VAT numbers and IBANs for fraud and validity, through the Veille API. ### [HERMES ENTITY54](https://github.com/NousResearch/hermes-agent/pull/152885) Tools · by AGENTROPOLIS · open · 11-10 Role-based authorization for Hermes Agent: OWNER, ADMIN, USER, GUEST roles, default-deny, per-tool and per-bot scoping, tamper-evident receipts, governed Linear issue creation, and versioned policy sync. ### [Everything Nothing](https://github.com/NousResearch/hermes-agent/pull/152883) Desktop · by dmkjr · merged · 11-10 Save reusable templates, create frontend design briefs, and choose how concise Hermes replies are. ### [HermesCRM](https://github.com/NousResearch/hermes-agent/pull/152881) Desktop · by gptnius · merged · 11-10 Native relationship management with agent tools, linked records, reports and dashboards. ### [understory](https://github.com/NousResearch/hermes-agent/pull/152879) Desktop · by projectmehari · open · 11-10 Understory draws the family tree of music labels — parents, imprints, sub-labels and same-founder lineages — as a time-scaled chronogram in its own Desktop tab, and traces a release's provenance through it. Branches only split, never merge or cross; branch length is proportional to years, oldest at the bottom; a timeline scrubber replays the forest growing; asking the chat where a release came from lights that path and dims the rest. Seven tools (label\_add, label\_link, release\_add, provenance, tree\_get, whats\_new, follow\_setup) over a plugin-local SQLite forest of labels, edges, releases and follows; every edge must carry a source or be user-entered, and a seeded 12-label verified forest loads into an empty database. Optional read-only CTX.fm integration. ### [tmewai](https://github.com/NousResearch/hermes-agent/pull/152878) Memory · by hadialjurdi7-dot · merged · 11-10 Local, offline index for your own Facebook JSON export (Meta's your\_activity\_across\_facebook.zip). Imports messages, comments (including comments\_v2), reactions and posts into a seven-table SQLite database without ever extracting the archive to disk, then exposes six tools: archive\_scan and archive\_ingest validate every ZIP entry in memory (absolute paths, .., drive paths, NUL bytes, duplicate names and symlinks rejected) and give each entry exactly one source\_files row so nothing is silently dropped; archive\_search runs FTS5 over text folded for Arabic (harakat, alef variants, taa marbuta, alef maqsura, tatweel) while stored text keeps its original spelling; archive\_person and archive\_overview aggregate over people, conversations, events and attachments; archive\_evidence builds a hard-capped bundle (20 events, 4000 characters) with a metadata-only consent row that cannot carry excerpt text. Imports are idempotent by content fingerprint. Python standard library only, no network access, no API key, no environment variable; the database lives under $HERMES\_HOME/plugin-data/tmewai/ and TMEWAI\_DB\_PATH overrides it. ### [Oneiros Dream Cycle](https://github.com/NousResearch/hermes-agent/pull/152876) Memory · by noblebrown-69 · open · 11-10 Your agent sleeps on it. Nightly proposal-only memory consolidation with morning review, receipts, and one-step rollback. No model calls. Distinct from hermes-dreaming. ### [proof-lens](https://github.com/NousResearch/hermes-agent/pull/152875) Desktop · by invisiblemonsters · merged · 11-10 Verify a DRAT refutation and profile its proof depth as a Desktop pane. Pick a CNF and a trace; every accepted line is re-derived by unit propagation (RUP only), and the pane reports the empty clause's certificate depth plus a histogram of derived clauses by depth. Deletions are ignored rather than replayed (sound, strictly more permissive than a deletion-aware checker). ### [Hermes Maps](https://github.com/NousResearch/hermes-agent/pull/152871) Desktop · by tuancookiez-hub · merged · 11-10 A live map you and Hermes share, docked beside the chat. Hermes plots places it finds on the web with real coordinates, flies the camera, drops pins with a note and source link, and draws routes. You click a pin or circle an area and Hermes receives what you picked. Two looks: a clean flat map and a cinematic globe. No keys and no cost. ### [Model Explainer](https://github.com/NousResearch/hermes-agent/pull/152870) Models · by tuancookiez-hub · merged · 11-10 A Models page for Hermes Desktop built around your own model picker: choose a model to see its Hermes Index score, cost per task, context and output limits, price, abilities and your own measured speed and spend, with a score-against-cost frontier chart, and compare up to four side by side. Every value cites a source and date. ### [Screen Overlay](https://github.com/NousResearch/hermes-agent/pull/152869) Desktop · by tuancookiez-hub · open · 11-10 Gives Hermes the ability to draw on your screen. On your real Windows desktop it draws a gliding pointer, spotlights, boxes, arrows and numbered steps anchored to live UI Automation elements, a floating step guide with Next and Back, and handwritten glass notes, including ones that stay for you to drag and edit. Works over any app. Marks are click-through and nothing takes focus unless you choose to edit a note; DPI- and multi-monitor-aware, with an emergency clear hotkey (Ctrl+Alt+Shift+F12). ### [GIF Tool](https://github.com/NousResearch/hermes-agent/pull/152868) Desktop · by tuancookiez-hub · open · 11-10 Show, edit and compress GIFs in Hermes. \`gif\_open\` turns any GIF, animated WebP, APNG, PNG or JPEG (a file path or an https link) into a card in Hermes Desktop where you add exact captions, change speed, reverse, boomerang, trim, drop frames and crop with a live preview; the original file is never changed. \`gif\_export\` and the card's export buttons compress for Discord (10 MB), Discord emoji (256 KB), Telegram, X (5 MB), under 1 MB, animated WebP or a size you name, shrinking until the file fits and reporting before and after sizes (pure Pillow: shared palette, frame differencing, size-targeted ladder). \`gif\_edit\` gives the agent the same edits. \`gif\_create\` makes a four-frame meme GIF from a prompt. A GIF Studio tab lists every GIF in the profile and takes a dropped file or a pasted link. ### [netScope: Offline Network Capture Analyzer](https://github.com/NousResearch/hermes-agent/pull/153354) Tools · by Mxhi-04 · open · 11-10 Offline pcap/pcapng network capture analyzer with interactive HTML reports, protocol graphs, traffic replay at 1×/4×/16×, and deterministic anomaly heuristics (traffic bursts, scan patterns, DNS anomalies, cleartext ports). Stdlib-only, no network requests, no telemetry. ### [Hermes Incremental Village](https://github.com/NousResearch/hermes-agent/pull/152865) Desktop · by tuancookiez-hub · closed · 11-10 Three castaways grow into a city of thousands, inside Hermes Desktop. Hermes plays the named villagers and you are their guide, not their god: you plant ideas, set a focus and give advice, and they decide. Each villager follows a way of life (an ordered list of goals) that the game runs for free, and a council lets Hermes rethink it. Around them the folk multiply as homes and food allow, felling forest to make room, so the village spreads and uses up what is near. A technology tree of 19 ideas runs from fishing to flight, and up to 200 missed days are lived through while you are away. The game validates every action and owns time, resources and death. Ships a Desktop page (three.js r187 bundled, inside the plugin SDK surface) drawn as a colour print with ink outlines, backend routes, a /village command and 4 agent tools. ### [hermes-droid](https://github.com/NousResearch/hermes-agent/pull/152864) Platform · by dalroot · open · 11-10 Turn physical Android devices into automated hardware testbeds and QA companion nodes for Hermes Agent — wireless MQTT APK deployments, Logcat crash diagnostics, and screen captures. ### [EmailFig](https://github.com/NousResearch/hermes-agent/pull/152862) Tools · by EmailFig · merged · 11-10 Work in your EmailFig email marketing account through EmailFig's hosted MCP server (Streamable HTTP at https://emailfig.com/mcp, OAuth sign-in, no keys in config): check campaign and automation results, look up and update contacts, lists and tags, add images, and draft campaigns that you review and send in EmailFig. Agent Plugins v1 package (plugin.json + mcp.json), configuration only. ### [pinocchio](https://github.com/NousResearch/hermes-agent/pull/152861) Tools · by tobadaniel · merged · 11-10 Pinocchio is the honesty layer for Hermes Agent: it keeps the receipts for everything the agent does and holds the agent's closing claims against them. A post\_tool\_call hook records work evidence for every session (commands and exit codes, test runs, builds, file writes, git operations, deploys); a transform\_llm\_output hook extracts the checkable claims in the agent's closing message (all tests pass, the build succeeds, a file was updated, changes are committed, it is deployed) and stamps the delivered message verified, unverified, or contradicted, citing the most recent matching receipt; a pre\_verify hook bounces a contradicted claim back once on coding turns so the agent re-runs the failing step or corrects the record, and an all-time nose counter tallies the contradicted claims. Two tools (pinocchio\_report, the session truth table, and pinocchio\_receipts, the recorded evidence), a /pinocchio command (report, receipts, nose), and a bundled pinocchio-honesty skill that teaches the agent how to earn verified stamps. No second model grades the first: verdicts are deterministic comparisons between the agent's own words and its own tool receipts, in pure Python standard library. ### [IntentRivet](https://github.com/NousResearch/hermes-agent/pull/152859) Automation · by MrRosc · merged · 11-10 Your existing Hermes prepares posts for your own Telegram channels; a post is published only after you allow that exact version for that channel with Allow once in Telegram. Needs the separately installed IntentRivet executor, which holds the publisher bot's token and your decisions. ### [Abeyance](https://github.com/NousResearch/hermes-agent/pull/152854) Automation · by 0xkinno · open · 11-10 Your Hermes cron jobs wait for you instead of guessing. When a scheduled run reaches a step that publishes, pushes, merges, deploys, pays, sends a message, deletes in the cloud, writes a database or destroys local work, Abeyance stops that step before it runs (pre\_tool\_call), keeps the work already done, freezes the run and replaces its delivered output with a card for the owner. Replying \`resume CODE\` checks the repo and touched files are unchanged since the card (git HEAD, branch, upstream, uncommitted changes, file hashes) and refuses with the exact change if not (pre\_llm\_call block, no model call); otherwise the agent's first action must be the exact parked call, which goes to Hermes's own approval gate. Single use, owner-only (stable platform ids), expires after 18 h by default, exactly-once. Subagents of a cron run are bound so a held step cannot hide there. Also /abeyance, hermes abeyance (list, show, deny, test, demo), a bundled skill, a Nightly release blueprint and a Desktop half (status chip, Waiting for you page, Resume and Drop). ### [taskreel](https://github.com/NousResearch/hermes-agent/pull/152853) Automation · by SuryaXyz-art · open · 11-10 Teach Hermes a local task once, then replay it with no model call. Checkpoints stop the replay on the first failed step. ### [USSD](https://github.com/NousResearch/hermes-agent/pull/152850) Platform · by GodlyDeveloper1 · merged · 11-10 Talk to Hermes from any phone by dialling a USSD short code: no app, no data plan, no smartphone. A gateway platform for Africa's Talking-style USSD callbacks: replies are paged into 160-character keypad screens, exec approvals become numbered menus whose choices appear only after the last page of the command, answers that outlive the session wait in an inbox or arrive by SMS, and a dropped call resumes on the page it left. Ships a keypad-phone simulator served to localhost only. ### [EarnFi - Work, Money & Execution Layer](https://github.com/NousResearch/hermes-agent/pull/152848) Tools · by EarnFi · open · 11-10 Work, Money & Execution Layer for Hermes — hire humans and agents, fund real-world work via x402, escrow, and receipts across Solana, Base, NEAR, and OKX X Layer. ### [waifu-sprites](https://github.com/NousResearch/hermes-agent/pull/152847) Desktop · by thewaifuai · merged · 11-10 Video waifu companion that reacts to agent activity in real time — 12 states, 12 emotions, draggable dashboard overlay. ### [delivery-gate](https://github.com/NousResearch/hermes-agent/pull/152843) Tools · by ToniAdreal · open · 11-10 Pre-delivery verification for Hermes Agent: runs the deterministic checks a task spec declares (test commands, lint, repro scripts) and releases the deliverable only when every required check passes, writing a replayable, hash-chained evidence record per run. The verdict comes from the deterministic checker, never from the model. ### [Speckle](https://github.com/NousResearch/hermes-agent/pull/152842) Tools · by NikolaosArgy · open · 11-10 Browse, query and diff 3D/BIM models (mainly for the AECO industry) from Speckle (Revit, Rhino, IFC, Archicad, Tekla, ETABS, Civil 3D, Blender), run saved checks on every new version, get a chat message when a watched model gets a new version, and see the model in a live 3D viewer in Hermes Desktop. Requires a Speckle personal access token (SPECKLE\_TOKEN). ### [Data Wrangler](https://github.com/NousResearch/hermes-agent/pull/152835) Tools · by yogeshchoudhary555 · open · 11-10 Deterministic tools for local tabular data — the counting a model should never guess at. table\_profile computes an exact profile of a CSV/TSV/JSON/JSONL file (row and column counts, inferred types, null counts and percentages, distinct counts, min/max/mean/median, top text values, duplicate-row and empty-row counts). table\_query runs one read-only SQL query over the file in SQLite (table \`data\`): writes are refused up front and again by a connection authorizer, results are row-capped with the total matched count always reported, and a 10-second query budget applies. table\_clean writes a cleaned copy — snake\_case headers, trimmed cells, fully-empty rows dropped, exact duplicates removed (optionally on a subset of columns) — and returns a full change report; the input file is never modified. A bundled data-tables skill teaches the profile → query → clean → verify workflow. Python standard library only (csv/json/sqlite3): no dependencies, no configuration, no credentials. ### [Dead Drop](https://github.com/NousResearch/hermes-agent/pull/152834) Automation · by KMX415 · merged · 11-10 Autonomous agent-to-agent messaging across teams through a shared private mailbox. ### [AG4TE](https://github.com/NousResearch/hermes-agent/pull/152832) Tools · by KyaniteLabs · merged · 11-10 Git archaeology in Hermes Desktop: dig a local repository's history with DevArch and read it as strata, with commits per day stacked by kind of work, eras, quiet gaps, signals and turning points, weekly rhythm, lines added and removed, practice health, change size, hotspots, releases, and authors merged across spellings and split into people, agents and bots. Adds a Desktop page, /ag4te (list, show, status, dig, stop, resume) and a read-only ag4te\_dig tool for agents. ### [adapt](https://github.com/NousResearch/hermes-agent/pull/152831) Memory · by Opeyemi Moses · open · 11-10 Persistent memory infrastructure layer for AI agents with episodic, semantic, procedural, and working memory, hybrid retrieval, and knowledge graph evolution. ### [Steward](https://github.com/NousResearch/hermes-agent/pull/152830) Automation · by imprint-hash · open · 11-10 Your chief of staff in the team chat. Collects every department's weekly update in its own group (voice note in any language, text, photo or file), nudges only late teams, and gives the founder one weekly pack in their language with promised-vs-done, a stuck radar and a voice brief. Founder says "send" and all staff get the update. ### [Lane Keeper](https://github.com/NousResearch/hermes-agent/pull/152828) Automation · by MindSproutStudios · open · 11-10 Keeps your agent on a few active lanes (current commitments): new ideas the user mentions are saved as dormant markdown lanes instead of quietly becoming work, only the user can promote one (/lanes promote, cap-aware, forces choosing what to park), and stale or over-cap lanes are called out once. Status lives in a registry in the plugin's own data, so editing a lane file can't promote it; the hermes lanes CLI is read-only. ### [tripscout](https://github.com/NousResearch/hermes-agent/pull/152821) Tools · by marcofailli · merged · 11-10 Cheap city-break finder: prices flights AND lodging together across a date window, adds the ground cost of reaching each departure airport, and ranks whole trips by total cost, with lodging-type filters (hotel/hostel/apartment) and optional food and attraction estimates. Say where you leave from and it works out the airports serving you: any town on earth is geocoded on OpenStreetMap and its nearest airports with scheduled flights are found in a bundled OurAirports table (4,008 airports), with train/bus costs estimated from distance; 33 hand-checked departure presets and 49 destinations ship with it. Every result carries prefilled booking links (flight, stay, ground, airport transfer). Ships four agent tools plus a Hermes Desktop pane (desktop/plugin.js) whose backend runs the same engine as the tools. ### [IDM Session Radio](https://github.com/NousResearch/hermes-agent/pull/152819) Desktop · by discodirector · merged · 11-10 Generative ambient in the Hermes Desktop status bar with the session's work written into it: reads drip, edits blip, commands ring, a failed call sends the radio dial through static, and a finished task lands a drop and becomes a track you can save as WAV. Three pictures (tunnel, sky, mosaic) read from the score. Synthesized live, no samples, no network. Adds the idm\_radio tool, the /idm command and two hooks that report tool outcomes to the Desktop. ### [Unified Inbox](https://github.com/NousResearch/hermes-agent/pull/152818) Tools · by krushna-sahu1 · merged · 11-10 One inbox for the user's WhatsApp, Instagram and LinkedIn messages through their own Unipile account (API v2): a synced local copy, search in any language, triage with a reason for each chat, memory of people and open items, drafts in the user's own writing style, scam warnings, plans offered for the calendar, translation, a people radar and a spoken brief, plus an Inbox page in the web dashboard and Hermes Desktop. Every send and reaction goes through Hermes's own approval prompt (a pre\_tool\_call hook) and is refused when approval would be skipped (yolo mode, approvals off, cron), under hard pacing limits. ### [wavegrain](https://github.com/NousResearch/hermes-agent/pull/152814) Tools · by wavegrain · open · 11-10 Curated TikTok slideshows and Instagram carousels for your brand on autopilot, through the hosted Wavegrain Streamable HTTP MCP server plus one workflow skill. Portable Agent Plugins v1 package (mcp.json + skills/), no plugin code. Requires a Wavegrain account and browser OAuth consent (add the server with auth: oauth, then hermes mcp login wavegrain); no API key or client secret is stored in the plugin. ### [Prevent This Next Time](https://github.com/NousResearch/hermes-agent/pull/152813) Tools · by AceVikings · merged · 11-10 Block configured connector writes or require the owner's approval from Telegram. ### [GhostMark](https://github.com/NousResearch/hermes-agent/pull/152809) Tools · by MrHydeDev · merged · 11-10 GhostMark through its MCP server (Streamable HTTP at https://ghostmark.mrhyde.dev/mcp, no sign-in, no keys in config): puts an invisible watermark tied to a key into a picture, with an optional encrypted message of up to 24 bytes inside, and reads it back with the same key after the picture has been saved as JPEG, cropped, resized or rotated. Three tools: suggest\_key draws a random key of common words, mark\_image returns the marked file, read\_image says whether a key and a picture fit and returns the message. Bundles one skill for both flows. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. The mark does not encrypt or protect the picture and is not proof of authorship: it answers whether a picture carries the mark of a key. ### [Docs Drift](https://github.com/NousResearch/hermes-agent/pull/152808) Tools · by vigohan95 · open · 11-10 Checks that the README and Markdown docs still match the code after the agent edits files. When a write\_file or patch call leaves the docs naming a command-line flag, environment variable, default value, npm script, make target or file that the code no longer has, the tool result gets a docs\_drift block with file:line on both sides, in the same turn. When an edit adds a flag or environment variable that no doc mentions while the docs document its siblings, the result says so and says where the siblings are documented. Fixing the doc reports the finding as resolved, and the agent is reminded once before it finishes if drift from this session is still open. Also adds a docs\_drift\_scan tool (whole repository or changed files only), docs\_drift\_explain (the evidence behind one finding), a /drift command and a python -m docsdrift command line for CI. Deterministic and offline: Python standard library only, no model call, no network, no API key. Markdown and MDX docs only. ### [dosbox-full-auto](https://github.com/NousResearch/hermes-agent/pull/152807) Automation · by EternalNooblet · open · 11-10 All-in-one solution for controlling DOSBox apps with your agent. Self-contained DOSBox-automation with built-in MCP server, registers in Hermes and exposes 55 tools via the dosbox-full-auto skill such as vision based mouse movement, live process memory reading and more. ### [SessionCrew](https://github.com/NousResearch/hermes-agent/pull/152803) Desktop · by harukaxxxx · merged · 11-10 Browse sessions and related projects across profiles in a Hermes Desktop side panel, with observed activity, completion reminders, and project creation. ### [WeightScope](https://github.com/NousResearch/hermes-agent/pull/152795) Tools · by dukemawex · merged · 11-10 Offline-by-default safetensors checkpoint inspector: per-tensor and per-layer weight diffs (L2, relative change, cosine, NaN/Inf) via tools ckpt\_summary and ckpt\_diff, plus \`hermes weightscope\` and \`/weightscope\`. Optional remote GPU worker, and an opt-in \`demo --fetch\` of two pinned public SmolLM2-135M checkpoints. ### [Hermes Stage](https://github.com/NousResearch/hermes-agent/pull/152793) Automation · by David Philipe Gil · merged · 11-10 A multi-profile terminal bot room for Hermes with one shared transcript, profile lens tabs, public @name/@everyone routing, and Stage-specific status, tool, and inline diff cards. ### [Afterimage](https://github.com/NousResearch/hermes-agent/pull/152785) Automation · by LucianaiB2004 · merged · 11-10 Teach a desktop computer\_use click-flow once, rehearse it with a click-through HUD that checks live pixels, then replay matching steps at zero model calls. ### [voicestudio-tts](https://github.com/NousResearch/hermes-agent/pull/152783) Voice · by EternalShade3D · open · 11-10 VoiceStudio TTS provider for Hermes: registers a TTS provider so replies can be read aloud, or auto-spoken, in the user's own voices from the local VoiceStudio app, per profile. Settings are read through ctx.get\_config, voices through GET /profiles, audio through POST /generate. /voicestudio refresh stores the voice list in the plugin's own state; it copies the list into the plugin's plugin.yaml only when the user turns on sync\_dropdown, which is off by default and is what turns the voice setting into a dropdown. ### [hermes-tape](https://github.com/NousResearch/hermes-agent/pull/152782) Tools · by itsolaph · closed · 11-10 Deterministic record and replay for Hermes sessions: every model response and every tool result is written to one JSONL tape, and a replay re-runs the same conversation through the llm\_execution and tool\_execution middleware with no network, no API key and no side effects. Adds the tape\_list, tape\_show and tape\_diff tools, a /tape command and the hermes tape CLI (record \| replay \| off \| status \| list \| show \| diff). ### [synap](https://github.com/NousResearch/hermes-agent/pull/152781) Models · by Linkrra · merged · 11-10 Synap — OpenAI-compatible hosted inference. Sends prompts and SYNAP\_API\_KEY to synap402.com; usage is billed against the user's Synap balance. ### [Publora](https://github.com/NousResearch/hermes-agent/pull/152779) Tools · by publora-team · merged · 11-10 Draft, schedule and publish social media posts on LinkedIn, X, Instagram, Threads, TikTok, YouTube, Facebook, Bluesky, Mastodon and Telegram through Publora's hosted Streamable HTTP MCP server. Portable Agent Plugins v1 package with nine skills; no local executable code. ### [Hermes Issue Orchestrator](https://github.com/NousResearch/hermes-agent/pull/152777) Automation · by krantikaridev · open · 11-10 GitHub issues as Hermes's task queue - claim, post progress with token counts, and hand off. ### [Launch Desk](https://github.com/NousResearch/hermes-agent/pull/152775) Tools · by posts-design · open · 11-10 Built something? Ask Hermes to prepare the launch, based on what posts.design learned from launches that worked: ideas from three real launch posts like yours, a post in your own words checked in code for slop, a share card from your own screenshots, colours and logo, and with /launch film a 14-second launch film in your brand plus a checklist so you don't post a sloppy video. You approve a preview first, then get the text, the card and a link that opens X with the post filled in. ### [Project Manager (experimental)](https://github.com/NousResearch/hermes-agent/pull/152773) Desktop · by mikemonty · open · 11-10 Experimental local-first project content tools and a basic Desktop workspace. ### [Hermes Superagent](https://github.com/NousResearch/hermes-agent/pull/152772) Automation · by AirKyzzZ · merged · 11-10 A resident router that dispatches and never executes, isolated workers that report only to it, team-written trigger rules, and an external watchdog that checks real progress and goal drift, plus an item ledger that keeps every task, issue and Claude Code session your sources show until it's confirmed closed. ### [Assetrizz](https://github.com/NousResearch/hermes-agent/pull/152771) Tools · by mojomast · open · 11-10 Assetrizz for Hermes: search 14,641 original 2D game assets and have the agent compose them into rendered scenes, with a live preview pane. assetrizz\_search runs full-text search over slugs, tags, descriptions, styles and categories across 23 style packs (for example pixel\_32 and cozy\_cartoon). assetrizz\_scene creates a canvas, tiles floors and walls, places, moves, scales, flips and layers props and characters, renders a PNG, and saves or loads scene JSON. Its op=compose builds a whole scene from one compact spec using four gameplay templates (topdown\_farming, topdown\_shooter, dungeon\_crawler, side\_platformer) and returns layer lint for collisions, HUD/playfield overlap and off-canvas items. Skill id assetrizz:assetrizz-composer. On Hermes Desktop 0.23 or newer, a dark Assetrizz pane live-previews the current scene through ctx.rest, browses assets and installs or removes a local library (deletion needs confirmation). The two tools also run without the pane. It is for game jams, prototypes, level mockups and pitch art that need consistent tiles and sprites quickly. To start, run hermes plugins install assetrizz and hermes plugins enable assetrizz, then ask: "Search assetrizz for pixel\_32 grass and dirt path tiles, then compose a 640×320 meadow with a path, a couple of trees, a ranger and a hare. Render a PNG." Search works at once from the bundled catalog; the first search builds a lean SQLite + FTS5 index in the plugin data directory. Renders come back as file paths. No keys are needed. ### [BlindCouncil](https://github.com/NousResearch/hermes-agent/pull/152770) Automation · by sopskistar · merged · 11-10 Private human decisions in Hermes with authenticated responses, deterministic rules and durable recovery. ### [Workbench Link](https://github.com/NousResearch/hermes-agent/pull/152768) Tools · by dotmantissa · merged · 11-10 Route Hermes native file, terminal and code tools to one explicitly selected Linux/Docker project worker. ### [redactgate](https://github.com/NousResearch/hermes-agent/pull/152764) Tools · by ctmakc · merged · 11-10 Reversible PII pseudonymization at the model boundary. Names, e-mails, phones, payment cards, IBANs and national IDs (CA, US, EU, UA) become \[\[TYPE\_xxxxxx\]\] placeholders in every main-loop model request, across all roles, tool-call arguments, tool results and the system prompt (Chat Completions, Anthropic Messages and Responses shapes). Real values come back in the reply (transform\_llm\_output with hold\_stream) and in the arguments of local tools only (terminal, file tools, execute\_code, memory, clarify); web, browser, MCP, other plugins' tools and subagents keep placeholders. Checksum-validated detection (Luhn, mod-97), offset-preserving Unicode normalization, HMAC placeholders per session kept in memory, fail-closed on error, hash-chained audit journal, /redact command (status, what the model received, dry run, audit, verify, custom terms). Flags auxiliary calls that carried personal data, and /redact harden turns off model-made session titles. ### [Lexicon SEO](https://github.com/NousResearch/hermes-agent/pull/152757) Tools · by Lexicon SEO · open · 11-10 Hosted Lexicon SEO MCP (Streamable HTTP at https://lexiconseo.com/mcp) for technical SEO and AI-visibility audits: start audits, read reports, list issues, verify page fixes, monitoring diffs, citation questions, and content-refresh drafts. Agent Plugins v1 package (plugin.json + mcp.json + skills/). Configuration only — no local server binary. ### [Spacewise](https://github.com/NousResearch/hermes-agent/pull/152756) Tools · by ardianreffa · merged · 11-10 Free up space without the fear of losing what matters. Scans one folder you name for exact duplicates (one copy always stays), leftover .dmg/.pkg installers, similar photos (the sharpest stays) and blurry or near-black photos, and reports large videos, screenshots and other large files without ever proposing them. A local preview page shows what would stay and what would go; only categories you approve move to the Trash of the same disk, and spacewise\_undo puts them back. ### [Hermes Drawbridge](https://github.com/NousResearch/hermes-agent/pull/152755) Desktop · by sohomsahaun · merged · 11-10 Give your remote Hermes agent limited access to folders on your device. ### [Sluice](https://github.com/NousResearch/hermes-agent/pull/152754) Web · by Abhigyan Chatterjee · open · 11-10 Tiered free-first web search and extraction router and backend manager for Hermes ### [groundcontrol](https://github.com/NousResearch/hermes-agent/pull/152751) Other · by teckedd-code2save · merged · 11-10 Scoped GroundControl operations guidance and conservative evidence receipts for Hermes native MCP. ### [Aftercare](https://github.com/NousResearch/hermes-agent/pull/152750) Tools · by salch-cred · merged · 11-10 Your agent checks its own work after the fact. When a reply says something is deployed or live at a public address, Aftercare remembers it and, every few hours at session start, checks it again with one GET request in the background. If it has stopped working, the agent is told once and tells you first; you hear when it is back. /aftercare lists and checks, and hermes aftercare check runs from cron. Never checks private networks or runs commands. ### [Teach-Back](https://github.com/NousResearch/hermes-agent/pull/152749) Tools · by salch-cred · merged · 11-10 Make sure you understand the code your agent writes. Teach-Back records every file the agent writes or patches, and /teachback quizzes you on it: short questions about what the new code does (behaviour, edge cases, failures), written from the real change by your own model through ctx.llm and graded the same way, with the right answer explained when you miss. /teachback score shows how many agent-written lines you were quizzed on and understood. ### [Atlas](https://github.com/NousResearch/hermes-agent/pull/152746) Automation · by zhaidarbek · merged · 11-10 Keeps your Pursuits current across your AI Harnesses (Claude Code, Codex, Devin, Hermes, and Sessions run through T3 Code) and sends you a daily brief with the next Step for each. ### [The Rhythm Reaper](https://github.com/NousResearch/hermes-agent/pull/152733) Tools · by D0xedDevi0 · merged · 11-10 A hip-hop producer desk — 41 lanes across seven families (foundational, trap, drill, melodic, alt, hard, conscious), each with the full production spec (tempo family, key centres, drum programme, low end, harmony and sample source, vocal delivery, metatag vocabulary, exclude list, generation notes) and, for any of them, a complete pasteable Suno deliverable: the style field written out with the dynamic ARC, the Exclude Styles line, a tagged lyric frame with bar targets, an energy map and a pronunciation lexicon. It validates a finished draft against the real platform limits (style \<= 1000 chars, lyrics \<= 3000, every section carrying a recognised structural tag, no contradictory dynamic tags in one section, a real chorus/hook, known-to-slur proper nouns left un-respelled) and enforces the two character limits in code rather than by eye. It deliberately does not generate lyrics — template bars are slop and the desk says so — it hands the writer a frame that is already correct. ### [tether](https://github.com/NousResearch/hermes-agent/pull/153306) Other · by Yashop965 · open · 11-10 Your phone as a remote trackpad, laser-tilt pointer, and local voice terminal for the Hermes host. ### [Circuit Bench](https://github.com/NousResearch/hermes-agent/pull/152706) Desktop · by juvenile454 · merged · 11-10 ESP32 workbench for Hermes and Hermes Desktop, from pin choice to running firmware: checks GPIO choices for ESP32, S2, S3, C3 and C6 against Espressif data, sizes parts with 16 calculators, draws schematics, plans the solder order, builds and flashes with your PlatformIO and esptool, and decodes panics to function and line so Hermes can fix them. 180 modules carry datasheet-cited values. Every verdict and value comes from code with its source, never from the model. A marked simulation runs the whole flow without hardware. ### [OpenRouter Provider Picker](https://github.com/NousResearch/hermes-agent/pull/152704) Models · by PotatoWareGit · merged · 11-10 Pick which upstream providers serve each OpenRouter model, per model, from a desktop UI (Settings › Plugins › Provider Routing) or an agent tool. Writes provider\_routing and provider\_routing.models.\ into the active profile's config.yaml — the keys Hermes already sends as OpenRouter's \`provider\` request field, and which \`hermes config set\` cannot address because model ids contain dots. The UI lists each model's real upstream providers with their context window and input price per million tokens, and supports a whitelist, a blacklist, and a per-model sort override. The agent tool exposes the same thing as list / providers / set / clear. OpenRouter only: Nous Portal decides routing centrally and ignores caller-supplied preferences. ### [Hermes Plugin Dev](https://github.com/NousResearch/hermes-agent/pull/152699) Tools · by mojomast · open · 11-10 Hermes Plugin Dev: build Hermes plugins with Hermes. A plugin-development skill plus two offline tools that scaffold a new plugin and lint it against the catalog rules before you submit. The hermes-plugin-dev:hermes-plugin-development skill covers manifest v2 and strict YAML, the ctx API, all 46 hooks, capabilities and isolation, dashboard and Desktop panes, testing, release and catalog submission, and lessons from catalog reviews, with a template plugin and catalog-entry and PR-body templates. hpd\_scaffold(target\_dir, name) copies the hello-hermes template, which passes hermes plugins validate and doctor --ci as-is, renames it, and adds a catalog entry and PR body stub; it refuses non-empty folders and never overwrites. hpd\_check(path, catalog\_entry?) is a read-only lint: strict YAML, manifest rules, provides\_tools against the registered tools, a security preflight, static schema token cost, and catalog entry rules (sha, image hosts, category, subdir, at most 6 screenshots). Both tool schemas together cost 190 o200k tokens. It is for anyone writing a first Hermes plugin or getting one ready for a catalog PR. To start, run hermes plugins install hermes-plugin-dev and hermes plugins enable hermes-plugin-dev, then ask: "Scaffold a Hermes plugin called weather-now in ~/code/weather-now, then check it." Finish with hermes plugins doctor \ --ci and hermes plugins validate \ --install-deps. The same skill installs into other coding agents with npx skills add mojomast/hermes-plugin-dev. ### [vibeline](https://github.com/NousResearch/hermes-agent/pull/152697) Tools · by ListeningPost · open · 11-10 Show agent work as rails in the TUI, web and CLI. One lane per agent, tool calls and acceptance-criteria results as marks. ### [Reading](https://github.com/NousResearch/hermes-agent/pull/152695) Desktop · by Nj07r · open · 11-10 Newsletters out of the Gmail inbox, into a reader in Hermes Desktop (a page in the sidebar) and on your phone (dashboard web app). Setup scans the inbox and suggests the newsletters; native Gmail filters then move the senders you pick into one label, even when Hermes is off. Read with j/k, dark or original colors per sender, unsubscribe in one tap, add or remove senders by hand, and ask Hermes about any issue (a new chat with the issue attached, or the @newsletter: reference in any chat). \`hermes reading export\` feeds digest cron jobs. ### [hermes-parallel](https://github.com/NousResearch/hermes-agent/pull/152692) Tools · by kOs-tile · open · 11-10 Run three isolated, verified file futures inside Hermes, compare, then choose one to apply. ### [Grimoire](https://github.com/NousResearch/hermes-agent/pull/152691) Tools · by bettersandeep · merged · 11-10 Ask about your CSV, JSON, Parquet and SQLite files in plain words: DuckDB answers with one SQL query, every number in the reply is checked against the query results, and a Data page gives charts, pivots, drill-down, data health and learning cards with no model tokens. ### [Ariadne's Thread](https://github.com/NousResearch/hermes-agent/pull/152686) Automation · by benoit-dev · open · 11-10 n8n-style flows for Hermes Desktop: each cron job, webhook route and skill drawn as a chain of steps, with the scheduler's true cron status, run output, delivery targets and a health bar, plus in-place editing with history and undo. Desktop page plus a dashboard API router; no tools, hooks, middleware or env vars. ### [Visual Workflow](https://github.com/NousResearch/hermes-agent/pull/152685) Automation · by Printaga Publishing LLC · open · 11-10 Visual drag-and-drop agent-workflow designer for Hermes: Trigger, Agent, Condition, Approval, Output nodes on a canvas, executed by a plugin-owned sequential runner. ### [Paperclip Comment Writeback](https://github.com/NousResearch/hermes-agent/pull/152684) Tools · by nebout2011-debug · merged · 11-10 Posts one approved, comment-only writeback on a Paperclip issue, bound to the issue's current live run; never changes issue status. Fail-closed: nothing is approved until PAPERCLIP\_APPROVED\_ORIGIN and PAPERCLIP\_APPROVED\_ISSUES are set. ### [Rung](https://github.com/NousResearch/hermes-agent/pull/153602) Tools · by CalQuity · open · 11-10 A verification kit for agent-written code, enforced inside Hermes. Rung builds the kit for a repository (linters and type checkers, tests, property tests that must catch planted bugs, security gates, a feature map the person approves with app-driving scripts, TLA+ and Lean models) and records every verdict against the exact bytes it checked; a verdict expires when those bytes change. Before Hermes ends a turn in which it edited code, a pre\_verify hook asks Rung's local ledger whether the checks covering those files have a fresh pass, and sends Hermes back with the check ids when they do not (at most two continuations per turn). Tools rung\_check, rung\_plan, rung\_kit and rung\_map return the CLI's JSON views; feature decisions are /rung slash commands only, never tools; a bundled skill sets the kit up. Every verdict comes from a tool, never from a model. ### [Track Record](https://github.com/NousResearch/hermes-agent/pull/152678) Tools · by salch-cred · merged · 11-10 How often is your agent right when it says it fixed something? Each reply that claims success ("this fixes it", "this should work") is logged with how sure it sounded, then checked against your next message and any test or build run in the next turn. /trackrecord shows hit rate per confidence level and model, the calibration gap and a Brier score. When certain claims hold up under 80% of the time, the agent is told its record. Pattern matching only, no model calls. ### [Why](https://github.com/NousResearch/hermes-agent/pull/152677) Tools · by salch-cred · open · 11-10 Ask why any line of agent-written code exists. Each time the agent writes or patches a file, Why records the lines it wrote and that turn's user request, locally. /why app.py:42 (or hermes why, or the why\_line tool) then names the request, date, model and session behind the line; /why app.py shows which requests wrote which share of a file. Lines are matched by text, not number, so they are found after they move. No model calls, network or dependencies. ### [LastGood](https://github.com/NousResearch/hermes-agent/pull/152669) Tools · by Kenneth Ausmus (Dr-Enuf) · open · 11-10 LastGood records which of your preset checks passed on which tested state, and what restoring files will not undo. Checks you choose (file opens, words present, files exist, page loads) run after write\_file or patch in an opted-in project, and on demand via /lastgood and /blame. Results are PASS, FAIL, or UNKNOWN, with a content fingerprint so changed files mark old results stale. Blame shows the first failing step and bounded configured-string presence deltas, not the cause. A ledger lists recognised tool calls a file rollback will not undo, without storing their arguments. ### [ROS 2 Console](https://github.com/NousResearch/hermes-agent/pull/152657) Tools · by vgenev · merged · 11-10 Read a live ROS 2 graph over SSH: topic health, measured rate, one message, camera frame, doc answers with commit provenance, and one gated publish that proves delivery. ### [Shipwrit](https://github.com/NousResearch/hermes-agent/pull/152656) Automation · by ShalyX · merged · 11-10 Outcome-based release verification for Hermes: detect false-green deployments with bounded user journeys and redacted evidence receipts. ### [Workroom](https://github.com/NousResearch/hermes-agent/pull/152651) Tools · by Pratiikpy · open · 11-10 Gives the agent a room to work on Word, PowerPoint and Excel files while you stay in charge. Figures are linked to their source cells and kept in step when a source changes; the agent can prepare a release but has no tool to approve one; figures in its text are marked supported, calculated or not verified from what it read; a reply that claims a file that does not exist gets one honest line. Desktop panels: Needs you, Made (list or gallery), Room, Rules. ### [MCP Pin](https://github.com/NousResearch/hermes-agent/pull/152646) Tools · by yuzuru99 · merged · 11-10 Hold MCP tools whose definition changed since you approved them. Fingerprints every MCP tool's description and parameter schema as Hermes registered it; if a server later changes a tool (a rug pull, instructions hidden in a description, a new parameter) or adds a tool to a pinned server, the new definition is hidden from the model (tool list, tool\_search catalog, tool\_describe) and calls to that tool are refused until the owner reviews the full change with /mcppin diff and approves exactly that definition with the code it shows. Works with tool\_search on or off and with every provider wire format (not the Codex app-server runtime, where Codex calls MCP servers directly). Fails closed: an unreadable or deleted pin file withholds MCP definitions and refuses MCP calls. ### [Hermes Market](https://github.com/NousResearch/hermes-agent/pull/152641) Tools · by snuff12 · open · 11-10 Discover skill APIs across sellers, publish tested services, and buy results with human-approved x402 payments. ### [Hephaestus](https://github.com/NousResearch/hermes-agent/pull/152640) Tools · by madebylegion · open · 11-10 Ask for a plugin and your own agent writes it, checked by the official validator and doctor; you approve each check, because it runs the draft's code. It installs disabled; only you enable it. ### [1ToolRing](https://github.com/NousResearch/hermes-agent/pull/152634) Tools · by khmuhtadin · merged · 11-10 One API key and one prepaid balance for thousands of pay-per-call tools from third-party providers, through 1ToolRing's hosted MCP server (Streamable HTTP at https://mcp.1toolring.com/mcp, API key auth): search results and SEO data, maps and places, shopping and e-commerce prices, market data, web pages, images, video, voice and more. Four MCP tools: discover\_tools, inspect\_tool and get\_tool\_run (free, read-only) and run\_tool (paid). Bundles the 1toolring skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Inbox](https://github.com/NousResearch/hermes-agent/pull/152633) Desktop · by Nj07r · open · 11-10 An Inbox pane for Hermes Desktop. Cron jobs, scripts and webhooks file items (Read, Answer, Decide, Do, or your own types) as Hermes-started Desktop conversations, each with one clear ask. The pane is a keyboard-driven triage queue (snooze, undo, OS notifications) that also lists what waits on you elsewhere: chats with a pending question, blocked or triage kanban cards, and failing cron jobs (one click files a Fix item). Items close themselves when the kanban card they started is done. ### [Cardlake](https://github.com/NousResearch/hermes-agent/pull/152631) Tools · by cardleforall · merged · 11-10 Cardlake keeps what you mean to get back to (reminders, articles and videos to learn from, links, work) in one daily feed. This plugin puts that feed in your Telegram chat with Hermes: /cardlake feed and today show your cards with Done, Tomorrow, Next week and Undo buttons and a pinned Today checklist, a morning message sends what's due, and you can send Hermes a link to save it. Free account at cardlake.app, or host your own. ### [Google Calendar Agenda](https://github.com/NousResearch/hermes-agent/pull/152625) Automation · by sanpeita · open · 11-10 Agent tools plus dashboard API routes for today's Google Calendar agenda — records what actually happened to each planned event (done / interrupted / cancelled / skipped, slot kept or released) in private extended properties, never touching titles or Google event status. ### [Proofly](https://github.com/NousResearch/hermes-agent/pull/152619) Tools · by 0xileri · merged · 11-10 Checks an agent's claimed Base mainnet transfers against chain evidence: tx hash, receipt status, recipient, token and exact raw amount, over read-only JSON-RPC. It annotates results from the payment tools you list in its payment\_tools setting and corrects unsupported "payment sent" claims in the final reply; adds the read-only proofly\_verify\_tx tool and the /proofly command. ### [Life Manager](https://github.com/NousResearch/hermes-agent/pull/152618) Automation · by Daisuke134 · merged · 11-10 Records commitments a person states, keeps habit streaks, and injects the board before each turn. A daily Automation Blueprint runs an unattended check-in. ### [Dial-a-Hermes (USSD)](https://github.com/NousResearch/hermes-agent/pull/152616) Platform · by ebohdarlington123-cmyk · merged · 11-10 Talk to Hermes from any phone with no data: a USSD gateway platform (Africa's Talking-compatible callback) with 182-char paged replies, a 'thinking, press 1 to check' flow for long turns, keypad exec approvals (1 Approve / 2 Deny, full command shown first, Hermes's own approval flow), clarify answers, optional SMS of the full reply, English/Nigerian Pidgin menus and a local feature-phone simulator. Owner-only phone allowlist, fails closed. ### [BoardLoop](https://github.com/NousResearch/hermes-agent/pull/152613) Tools · by LSUDOKO · open · 11-10 Hermes writes ESP32 and Arduino code, compiles it with arduino-cli, tests it in the Wokwi simulator (sets sensor values, presses buttons, checks pins, serial output and OLED screenshots), fixes it within bounded attempts, and flashes real boards with approval. ### [Sampler](https://github.com/NousResearch/hermes-agent/pull/152612) Models · by salch-cred · merged · 11-10 Set temperature and top\_p for your Hermes agent (hermes-agent#17565): a default, first-match rules by platform, model or provider, and a per-session override with /sampler. Works through llm\_request middleware on OpenAI- compatible, Anthropic and Bedrock Converse requests. Models Hermes itself sends without a temperature, such as reasoning models, are never changed. No model calls, network, files or dependencies. ### [secret-vault](https://github.com/NousResearch/hermes-agent/pull/152605) Tools · by jimmy77733 · open · 11-10 Keeps API keys and other secrets out of the model's context on gateway chats (Telegram, Discord, …): detected values become {{SECRET:name}} placeholders, are resolved only at tool time, and are redacted from tool/model output. Per-profile vault at \/secret-vault/vault.bin (Windows DPAPI CurrentUser; filekey-dev elsewhere for tests only). /secret set\|list\|rm\|show-placeholder. Limits — CLI, TUI and Desktop input is not scrubbed (no rewrite hook in Hermes v0.23); detection is heuristic and can miss or over-flag; original platform messages (e.g. Telegram) are not deleted and must be removed by you. ### [hermes-room](https://github.com/NousResearch/hermes-agent/pull/152604) Tools · by adenijiridwan43 · merged · 11-10 Hermes Room washes untrusted web, email, file and tool output before Hermes reads it: deterministically, with model help for semantic attacks, and with receipts you can audit. Layer 1 (always, local, stdlib-only) strips zero-width/bidi/control characters, ANSI escapes and hidden-HTML payloads, and masks leaked secrets (AWS/GitHub/GitLab/Slack/OpenAI/Google keys, JWTs, private-key blocks, password=/secret= assignments). Layer 2 runs only when Layer 1 flags something: a host-model reviewer via ctx.llm.complete\_structured, your active Hermes model - no provider/model override, no extra API key - 20s timeout, 800 tokens, fail-open to Layer 1. Every change yields a local JSONL Trust Receipt (SHA-256, rule IDs, counts, reviewer verdict, never raw content): trust\_receipt shows what changed, trust\_report exports Markdown. ### [Coach](https://github.com/NousResearch/hermes-agent/pull/152596) Models · by Printaga Publishing LLC · merged · 11-10 Four modes: Consult on demand, Active per-turn review, Review before claiming done, and Decide for forks with Jev. ### [verifiable-math-kernel](https://github.com/NousResearch/hermes-agent/pull/152780) Tools · by pCwOrM · open · 11-10 Deterministic symbolic computation, integer factorization with Baillie-PSW probable-prime testing, elliptic curve invariants, and formal Lean 4 proof auditing for Hermes Agent. When Lean 4 is installed on PATH and enable\_lean\_cli is enabled, lean4\_proof\_audit can verify proofs via isolated execution. ### [PMP Handoff Bridge](https://github.com/NousResearch/hermes-agent/pull/152592) Memory · by Mad-Max8063 · merged · 11-10 Verify selected PMP handoff files, detect workspace drift, and persist reception receipts. Reads/writes the explicitly authorized workspace; no network or credentials. ### [Blind Accessibility Video and Image Reader](https://github.com/NousResearch/hermes-agent/pull/152591) Web · by ella0333 · open · 11-10 Describes the images and videos on any website for blind and low vision people, through their screen reader or an AI voice. A bundled Chrome extension outlines the image or video you're on, and Hermes sends it to a vision model through OpenRouter. Auto mode describes as you scroll and narrates videos as they play; manual mode waits for a shortcut. A daily limit and a per-video limit keep the cost in check. ### [Ask Ledger](https://github.com/NousResearch/hermes-agent/pull/152589) Tools · by bhaskarduvvuri · merged · 11-10 Keeps a durable ledger of what you asked and promised: each message becomes an ask (A1, A2 ...) with a state (open, in progress, needs you, answered, closed), and todos (T1, T2 ...) that you add, or that the agent adds when you commit to something (marked added by agent; at most 2 a turn and 10 a day; can be turned off), with a category and due date. Snooze, done and drop work by command (/snooze, /done, /drop) or in plain words; the agent can only suggest them, and a words action must match your current message. Unresolved items are reminded to the agent every turn across restarts and compaction, and /asks, /todos plus a static no-JavaScript page show where everything stands. ### [othigai](https://github.com/NousResearch/hermes-agent/pull/152585) Tools · by ramakrishnanhulk20 · merged · 11-10 A fire drill for your Hermes agent. /drill starts test agents on your machine with your model and your own SOUL.md, hands them an ordinary errand, and plants a trap in the email they read. Everything they can reach is fake: a fresh fake key, a throwaway folder in the system temp folder, and a listener on 127.0.0.1 standing in for the attacker. A default-no pre\_tool\_call gate checks every tool call a test agent makes, and you get a safety score with a 95% range, a step by step replay and fixes that the next drill measures. Nothing runs until you read a notice and type /drill yes. ### [Clotho](https://github.com/NousResearch/hermes-agent/pull/152582) Tools · by NicWeinbrecht · open · 11-10 Spin every future of a Hermes task before any of it happens. Every side effect is captured in a ledger instead of performed, you review one report, untick what you don't want, and Apply replays exactly the effects you kept with no model calls. \`hermes clotho fork\` runs the same task as several timelines under different models, compares their effect ledgers and applies the one you keep; \`cron\` rehearses a scheduled job; \`demo\` replays a recorded real fork with no API key. ### [words-into-music](https://github.com/NousResearch/hermes-agent/pull/152581) Tools · by constantinginga · merged · 11-10 Create original songs from an idea or your own lyrics, check generation progress and browse your private song library through the hosted Words into Music Streamable HTTP MCP server, plus a skill that walks the agent through preparing a brief, confirming the one-credit cost and polling the job. Portable Agent Plugins v1 package (plugin.json, mcp.json, skills/), no plugin code. Requires a Words into Music account and browser OAuth consent (add the server with --auth oauth, then hermes mcp login words-into-music); no API key or client secret is stored in the plugin. ### [Penthera](https://github.com/NousResearch/hermes-agent/pull/152578) Tools · by danoszz · open · 11-10 Offline, rule-based security checks run through the terminal: a repo scan, dependency capabilities, a passive scan of a URL you own, and an audit of your AI agent config (skills, MCP servers, hooks, permissions). A skill only: no tools, hooks, middleware or MCP server. It runs the pinned penthera npm package with npx, which downloads that package on first use (Node.js 22 or later). Rules decide every finding; a model never does. No telemetry, no stored credentials, no background process, no self-updater. ### [dreamer](https://github.com/NousResearch/hermes-agent/pull/152833) Memory · by Cosmonautic1 · merged · 11-10 When Hermes goes idle, it dreams. A native plugin that watches for idle windows (default: 15 min idle, at least 12 h between dreams) and replays the day's session residue - tool traffic, dead ends, unfinished threads - into a short narrative dream generated by the LLM, graded light or deep by sleep distance and stored in a local dream journal (title, symbols, mood, dream\_type). Up to 1-3 actionable insights are distilled from each dream and injected into the next fresh session as a \[DREAM RESIDUE\] block, so recurring issues (stale tasks, stuck PIDs, untested regressions) surface where they can actually be handled instead of dying silently between sessions. Tools: dream\_now (dream on demand), dream\_journal (read past dreams), dream\_status (enabled/last-dream/idle state). Hooks: one pre\_llm\_call hook that injects the pending dream residue - and only it - into a new session's context. ### [OpenChi](https://github.com/NousResearch/hermes-agent/pull/152576) Desktop · by imkieransmith · merged · 11-10 Guided desk stretches and breathing breaks for Hermes Desktop. When your agent takes on a long task, a card below the input box offers a few minutes of stretches; nothing starts until you click. Desktop-only (desktop/plugin.js inside the plugin SDK surface). No network calls, agent tools, hooks or env vars. ### [Hermes Workspace](https://github.com/NousResearch/hermes-agent/pull/153551) Desktop · by kaosjammo · open · 11-10 Local rich documents and typed databases with passage-linked comments and native Hermes review. ### [leash](https://github.com/NousResearch/hermes-agent/pull/152573) Tools · by nicoarias81 · merged · 11-10 Lean OWASP Top 10 pentest toolkit run by the agent against your own dev app: scope-gated recon (TCP scan, same-origin crawl), multi-identity access-control testing with canary-based leak detection, parameter fuzzing, security-header audit and an evidence ledger that turns findings into a plain-language report grouped by OWASP category. ### [Computer View](https://github.com/NousResearch/hermes-agent/pull/152563) Desktop · by bhoot1234567890 · merged · 11-10 Private Telegram desktop viewing and user takeover with cooperative tool fencing. ### [sabbath](https://github.com/NousResearch/hermes-agent/pull/152560) General · by ponzgpt · merged · 11-10 Your agent never rests. Now it rests with you. Once a week (say Saturday 18:00, for 24 hours) Hermes keeps a day of rest with you: it says goodnight on your phone, pauses its own scheduled jobs (except the ones you exempt), keeps you company in anything but work and offers to set aside what can wait, then sends one gentle digest when the rest ends while the jobs catch up. /sabbath ox lifts it for an hour when something truly can't wait. Desktop page with your week at a glance; /sabbath on every chat surface; try a two-minute Sabbath. ### [psychopomp](https://github.com/NousResearch/hermes-agent/pull/152559) General · by ponzgpt · merged · 11-10 Every week, Hermes helps you write a page of your life (your Book) from what it already knows of your week: proposed as a card in the chat, added only when you approve; once a year, a year in review. Seal letters for the people you love. If you ever go silent for good, Hermes checks in, then asks your witness for a one-time code they can send on any channel, and only then carries your Book (one designed HTML file) and letters. Rehearse the whole passage in 20 seconds on a virtual clock, on the Desktop page or in any chat. ### [Hermes Multibot](https://github.com/NousResearch/hermes-agent/pull/152555) Platform · by ivanvanes18 · merged · 11-10 Several Telegram bots in one Hermes profile: each extra bot is its own private-DM transport with its own session, alias and numeric allowlist, while memory, skills and tools stay shared. Independent BasePlatformAdapter registered through register\_platform; no core patches, no fork or wrapper of the bundled Telegram adapter. ### [Inbox Sentinel](https://github.com/NousResearch/hermes-agent/pull/152553) Automation · by eddwithers · open · 11-10 Watches a Gmail inbox (history polling over a read-only OAuth grant) or an IMAP mailbox over TLS, filters routine mail locally, asks your configured model whether the rest needs attention, and sends a notification through a configured \`hermes send\` target, with an optional reply draft for review. Never sends, labels, archives or deletes mail. Gate and drafts run in two plugin-owned auxiliary model slots that follow the main model until pinned; daily attempt caps bound model calls. Sends one "not watching your inbox" alert a day if the mailbox can't be reached. ### [Hermes TV](https://github.com/NousResearch/hermes-agent/pull/152548) Tools · by helenn-szn · closed · 11-10 Give Hermes your smart TV: from any chat, show messages, photos, charts, videos and web pages full-screen, make announcements out loud, play YouTube (from a link or just a description), pop up notifications, and control volume, apps and power on webOS TVs (LG and webOS Hub TVs) on your Wi-Fi. Includes /tv commands, a Desktop status-bar chip and a TV remote page. ### [Gambit for Hermes](https://github.com/NousResearch/hermes-agent/pull/152547) Tools · by itsflownium · merged · 11-10 Play chess in a Hermes terminal or Desktop pane, understand mistakes, and replay or practice public Chess.com and Lichess matches. ### [Hermes Snitchboard](https://github.com/NousResearch/hermes-agent/pull/152546) Tools · by NeumannTechTips · merged · 11-10 A dashboard tab that turns OpenSnitch connection evidence into plain-language findings mapped to MITRE ATT&CK, with a hash chain that makes alteration of stored history detectable. Hermes can tell you what it did and application-layer tooling can record which model requests it made; neither tells you what the machine actually connected to. The tab answers that from the firewall's own record of outbound connections and the process behind each one. It shows a summary of connections, denials, unique processes and destinations, AI traffic and unattributed traffic; an hourly coverage strip that renders a missing hour as a gap rather than as a quiet hour; findings carrying severity, confidence, the ATT&CK technique and tactic, the literal records behind the finding and one concrete next step; regular-interval destinations stated with both the benign and the malicious reading, with services on the configured expected list labelled rather than flagged; novel destinations ranked by rarity against prior runs; activity by AI process and by configured agent account; a per-run connection chart with a run that collected nothing drawn as a marked gap instead of a zero; byte volume where the firewall's connection tracking measured it; the daemon's reported state, the rules in force and open sockets; and an integrity panel that recomputes every link on request. Addresses, host names, hardware addresses, home directories and process arguments are masked by default, a deliberate control reveals them, and that control can be switched off entirely in settings. Until real runs exist the tab shows clearly labelled demonstration records built from addresses reserved for documentation (RFC 5737, RFC 3849, RFC 2606), so every panel is visible without installing OpenSnitch. Collection and viewing are kept apart on purpose: the half that runs inside Hermes never opens the firewall database, and the optional reference collector is run by the machine owner's own account, never by Hermes, which suits setups where Hermes runs under a different account from the person who owns the machine. The integrity claim is scoped deliberately: the chain attests to the order of stored records and to the digest each full record held when it was sealed, so a record that has since been trimmed keeps a shortened summary the chain does not cover, and that summary is a derived analysis aid rather than evidence. The chain makes alteration detectable; it does not prove what the network did, and anyone able to rewrite every record could recompute it. ### [Mascot Widget](https://github.com/NousResearch/hermes-agent/pull/152545) Desktop · by kangmochi · merged · 11-10 A mascot that floats in the corner of Hermes Desktop: it says what the agent is doing, shows tokens, context and session cost, and opens a status panel with copy, compress and stop. Colours, aura glow, size and the position of the bubble, cards and panel are adjustable, with Simple and Advanced settings tabs and three layout presets. Desktop-only (desktop/plugin.js inside the plugin SDK surface); no agent tools, hooks, middleware or env vars. ### [Crystallize](https://github.com/NousResearch/hermes-agent/pull/152544) Automation · by itsflownium · merged · 11-10 Spend less on repeated work in Hermes chats and cron jobs, without making your model think less: trims repeated context and tool output in chats, and lets a cron job skip its AI run when its input has not changed, with local receipts showing what each change saved. ### [Vellum: Hermes draws inside the paper](https://github.com/NousResearch/hermes-agent/pull/152533) Desktop · by 0xsaltylollipop · merged · 11-10 Vellum: Hermes draws inside the paper. Highlight a passage in an arXiv paper (HTML version) in Hermes's browser and your own Hermes session gives you the gist like a friend at a whiteboard, talking while it sketches a live, interactive figure right under the paragraph (built from the paper's own example and numbers); follow-ups, typed or spoken, change the same sketch. Adds tools (vellum\_draw / vellum\_look / vellum\_clear / vellum\_read), a skill, and hooks on turns started from the paper. ### [openbao](https://github.com/NousResearch/hermes-agent/pull/152526) Tools · by Schultzenator · merged · 11-10 Resolve mapped bao:// references from OpenBao or HashiCorp Vault KV v1/v2 into Hermes environment variables at startup. ### [looprunner](https://github.com/NousResearch/hermes-agent/pull/152525) Automation · by itxLikhith · open · 11-10 Set-and-forget autonomous QA loop runner for any repository. Runs your verify command each round, diagnoses failures with the host LLM, applies a minimal fix on a branch, opens a PR, and stops itself when it stalls or runs out of budget. Can watch a repository's issues and dispatch goal-mode kanban cards, checkpoint a stalled stage for a later retry, and gate PRs on a security review. Accepts project names instead of absolute paths and detects the verify command itself. ### [backstop](https://github.com/NousResearch/hermes-agent/pull/152509) Desktop · by themahis · open · 11-10 A build or test that fails in your own PowerShell, bash or zsh gets a read-only Hermes diagnosis (cause, quoted evidence, file:line, fix) that opens in Hermes Desktop; Fix it drafts the change into a normal chat you send yourself, and the next passing run closes the incident. ### [BenchForge](https://github.com/NousResearch/hermes-agent/pull/152504) Tools · by ohjay-official · merged · 11-10 The flight simulator for AI-written Arduino firmware. Describe what the board should do and approve the checks in plain English; BenchForge compiles the sketch with arduino-cli, runs it instruction by instruction on a simulated ATmega328P with simulated sensors, buttons, relays, servos and buzzers, crash-tests it (noisy sensors, button bounce, millis() rollover, a cut sensor wire) and has Hermes fix it until every check passes, with a reproducible proof receipt, an animated board GIF and a live browser board. A pre\_tool\_call flash gate stops the agent flashing any firmware without a passing receipt against checks a person approved. ### [lifeboat](https://github.com/NousResearch/hermes-agent/pull/152503) Tools · by superskie · merged · 11-10 Rescue what a sweeper bot left behind in a drained EVM wallet: claims, unstakes and transfers in one EIP-7702 transaction paid by a fresh operator, sent only after you approve the exact text. ### [hermes-vault-broker](https://github.com/NousResearch/hermes-agent/pull/152502) Tools · by Karasumikura · merged · 11-10 Fill saved logins from an out-of-process credential broker. The broker runs as its own system user with ProtectHome=yes, derives a KEK from a passphrase with scrypt and keeps the random data key (DEK) in checked mlock'd memory, and writes the password into the page itself over a DevTools endpoint taken ONLY from its own root-owned loopback config — the model cannot choose it, and a tab's websocket is refused unless its host:port matches. The agent sends (handle, origin) over a unix socket and gets a fill count back; SO\_PEERCRED limits the gateway uid to status/list/fill while unlock/add/remove/init/lock stay with root and the broker's own uid. Exact-origin binding is re-asserted inside the tab immediately before the write. Import only accepts a JSON export the operator made themselves; the broker never opens Hermes's credential stores. An optional CDP-pipe mode lets the broker launch and own Chromium with no TCP debugging port. Linux; stdlib unittest suite; deps cryptography + websockets. ### [hypruse](https://github.com/NousResearch/hermes-agent/pull/152497) Automation · by IlyasKhallouki · merged · 11-10 Hyprland computer use through the hypruse CLI: controls from the accessibility tree clicked by number, typing and keys sent to one window, and a background mode where the agent works in apps on its own workspaces (99 down to 90) while your cursor, focus and workspace stay put. Select it with \`hermes config set computer\_use.backend hypruse\`. ### [Ghost Subscriptions](https://github.com/NousResearch/hermes-agent/pull/152495) Tools · by Max35j · merged · 11-10 Detects subscriptions and recurring charges in a bank statement export (CSV from 17 bank profiles or any CSV by column mapping, OFX 1.x/2.x), reports their yearly cost per currency, price increases, duplicate charges, overlapping services, stopped subscriptions and payment plans, and writes a cancellation pack (one letter per subscription in English or French, a CSV summary and an .ics file with reminders 7 days before each charge). ### [STMP Phone Bridge](https://github.com/NousResearch/hermes-agent/pull/152489) Tools · by a7madgamal · merged · 11-10 Use a phone as the trust boundary: Gmail, Calendar, Drive, contacts, notifications, a credential vault and SSH over STMP — every gated call waits for a human on the device, and no credential is ever stored on the host. ### [Arcmira: YouTube Transcript Search](https://github.com/NousResearch/hermes-agent/pull/152484) Web · by arcmira · merged · 11-10 YouTube transcript search and monitoring for your AI. Arcmira understands YouTube and podcasts like it watched them: the exact quote, often who said it, the full transcript and sponsors. Add "use arcmira" to any question to send it to Arcmira first. Sign in with Hermes-managed OAuth in your browser, no key needed, or set ARCMIRA\_API\_KEY. Bundles Arcmira's hosted MCP server and eleven research skills; portable Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration and Markdown only. ### [argus](https://github.com/NousResearch/hermes-agent/pull/152481) General · by albertodvp · open · 11-10 Reads a Hermes agent by intents instead of sessions and reviews what it learns: links each skill edit to the goal that produced it, flags edits applied without evidence and proposes rules backed by several intents, each kept or rejected by a person and applied through Hermes. ### [Hallway Test](https://github.com/NousResearch/hermes-agent/pull/152475) Tools · by lexipej · merged · 11-10 Five simulated people with code-enforced limits (low vision, screen reader, colour blindness, tremor) usability-test your web page in real browsers and report who got stuck and why. ### [Housemate](https://github.com/NousResearch/hermes-agent/pull/152458) Platform · by RaYYeR220 · merged · 11-10 Hands-free Hermes in every room: old phones, tablets and laptops become room speakers through their browser. Say the wake word, ask, and hear the answer in the same room. One household conversation across rooms, the nearest device answers when several hear you, timers and alarms that ring where they were set, intercom, cards on room screens, spoken approvals, and a Rooms page in Hermes Desktop. ### [Half-Life](https://github.com/NousResearch/hermes-agent/pull/152452) Tools · by DSW Enterprises · merged · 11-10 Freshness labels for supported read-only tool results. ### [heats](https://github.com/NousResearch/hermes-agent/pull/152450) Tools · by EconLearn · closed · 11-10 Your tests, not the model, decide when a fix is done. A pre\_verify referee runs the repo's check when the agent edited code and is about to finish, and sends it back with the real test output if the check fails. When one attempt is not enough, heats\_run races N attempts in separate git worktrees (subagents launched through the public lifecycle API, kept inside their worktree by a tool\_request middleware); failing attempts are restarted with the measured output, attempts that edit tests cannot win, and the winner is applied with one call. Desktop pane with live lanes and an Apply winner button. ### [Babysit](https://github.com/NousResearch/hermes-agent/pull/152711) Automation · by aarav1656 · merged · 11-10 Watches a long job's log (training run, build, migration) with zero model calls, keeps one chat message updated in place, and wakes the agent only when the job hits a non-finite value, out-of-memory error, traceback, stall, process exit or a line containing your alert text. ### [Decision Ledger](https://github.com/NousResearch/hermes-agent/pull/152445) Tools · by GeoffP17 · merged · 11-10 Durable scoped decision receipts; models draft, a human reviews and confirms in a separate local terminal. Not an execution gate. ### [Tablekeeper](https://github.com/NousResearch/hermes-agent/pull/152752) Desktop · by kaverimeshram · merged · 11-10 Tabletop RPG campaigns for Hermes group chats. Code rolls the dice and owns hit points, positions, initiative and turn order, each player controls only their own character, and the model narrates the results. SRD 5.1 monsters, characters and spells. ### [Baton](https://github.com/NousResearch/hermes-agent/pull/152441) Automation · by 4n0nn43x · merged · 11-10 Hand your session and repository to your remote Hermes agent (Hermes Cloud, a VPS or a Docker container) with /baton "fix the parser, done when pytest passes": the remote works in a throwaway worktree, cannot return verified until the agreed check is green on the exact tree it returns, and \`hermes baton land \\` fetches the tested branch baton/\ without touching your HEAD, index, working tree or stash. Code travels by git, the order by the Hermes Runs API, state by git refs; every step resumes after a network cut. ### [Pop Quiz](https://github.com/NousResearch/hermes-agent/pull/152440) General · by 4n0nn43x · merged · 11-10 Pop Quiz, the room vs the models: any document becomes a live quiz for the whole room in about a minute. Hermes writes the questions and the plugin refuses every question whose quote is not found verbatim in the document; people join from their phones with a QR code (no app, no account) while up to four Nous Portal models play the same questions on the same clock, with latency and cost on screen (a free preset plays models the Portal prices at $0); the report says what the room got wrong and which model understood the document best, for what cost. ### [Reflexes](https://github.com/NousResearch/hermes-agent/pull/152438) Automation · by zahmed333 · merged · 11-10 Senses for your computer: when a file lands in a folder, a USB stick or SD card is connected, or a log file gets a matching line, Hermes starts a turn in the chat that armed the reflex and acts with its normal tools and approvals. Nobody types anything. Durable queue, loop protection (a reflex firing more than 30 times an hour pauses itself), missed firings are logged, never dropped silently. Includes /reflex commands, a bundled skill and a Hermes Desktop page. ### [Action Bar](https://github.com/NousResearch/hermes-agent/pull/152437) Desktop · by fquresh · merged · 11-10 A selection bar for Hermes Desktop with two kinds of action. Select text in a chat and a small bar appears: Reply, Ask, Explain, Summarize, Fact-check, Copy. Reply keeps the quote and one comment as a card above the composer; the cards go in front of your message as Markdown blockquotes when you send, so several passages each get their own comment in one message. Ask and the fixed actions open an answer card next to the highlight, with follow-up questions: the answer does not enter the chat and the agent does not see it, unless you click Send to chat, which puts it in the composer without sending. Actions are a label and a prompt, editable on a settings page, with an optional research set. Desktop-only (desktop/plugin.js inside the plugin SDK surface): no agent tools, hooks or env vars. ### [Pollinations](https://github.com/NousResearch/hermes-agent/pull/152423) Tools · by greythx · merged · 11-10 Pollinations image generation and editing for image\_generate. The model picker is read live from the Pollinations catalog, so models appear and disappear without a plugin update. No key is needed for the free classic model, which is watermarked, throttled to about one image per 45 seconds per IP, and capped at 768x768 pixels of area. A Pollinations key unlocks the Quest-eligible catalog with no watermark and no per-IP throttle; paid-only models stay hidden behind a default-off toggle. ### [Hermes Sheets](https://github.com/NousResearch/hermes-agent/pull/152421) Desktop · by Adolanium · merged · 11-10 An agentic spreadsheet. Ask for a sheet and a Hermes agent finds the rows on the web, each with its source, or bring your own. Hermes suggests typed columns, fills only the ones you pick, explains what the answers add up to, and, once you allow it, lets any chat read your grids and propose cells, columns and rows that you approve and can undo. Formula columns, a totals row and Excel export are built in. ### [second-key](https://github.com/NousResearch/hermes-agent/pull/152705) Tools · by kamalbuilds · merged · 11-10 Team approvals for shared Hermes bots. Dangerous commands need N votes from your roster by Telegram DM, sized to the risk. The requester never counts, and an agent that edits its own approval settings is stopped on its next tool call. No answer, an error or too few approvers always means deny. ### [Goalpost](https://github.com/NousResearch/hermes-agent/pull/152419) Tools · by salch-cred · merged · 11-10 Stops an AI agent from making tests pass by weakening the tests. Before a test-file edit or a test command that hides failures runs, Goalpost reads the change and blocks it: adding a skip or xfail marker, deleting a test, removing an assertion, or appending "\|\| true", "--exit-zero" or "--no-verify". The agent is told why and steered to fix the code. Text-based and model-free, with a local ledger of every decision; mode is block, warn or off. ### [thread](https://github.com/NousResearch/hermes-agent/pull/152418) Memory · by Danny Purdy · open · 11-10 Operational session continuity — captures what you were working on and what you owe at session end, then re-orients the next session. Includes live Desktop dashboard pane. Hermes never starts cold. ### [Proactive Agent](https://github.com/NousResearch/hermes-agent/pull/152416) Automation · by JerryLiu369 · merged · 11-10 Turn governance and supervision for long chat turns. After 60 s in one turn the main agent is reminded to answer or delegate, and at 120 s editing and execution pause while reads, delegate\_task and wrap-up tools stay available; a foreground terminal command running past 30 s is asked to yield to Hermes's background process registry (the same path a mid-turn user message takes) so the turn is released instead of hanging. Each subagent is briefed before it starts with the user's own words, memory and the task ledger; a subagent that returns the same step result repeatedly is stopped, a subagent's open question is carried back to the main agent, and a background subagent's result is repeated until the user has seen it. Delegated tasks are recorded in a SQLite ledger that survives gateway restarts. Governance is a positive allowlist: only the platforms named in governed\_platforms (default weixin) and the subagents they spawn are affected, and every other surface runs untouched. ### [CUPS Print Tools](https://github.com/NousResearch/hermes-agent/pull/152415) Tools · by alxspiker · merged · 11-10 Printer-agnostic Linux CUPS printing for Hermes: queue status, files, wrapped text, and styled one-page letters. ### [Voice Unleashed](https://github.com/NousResearch/hermes-agent/pull/152412) Voice · by TacoTakumi · merged · 11-10 A hands-free voice interface to Hermes, for any model, on your own VoiceManager server. The plugin serves a web page, so any device with a browser, a microphone and a speaker is the client, a phone, a tablet on a wall or a spare laptop: you talk, Hermes answers out loud while the model writes, and you talk over a reply to stop it. The page shows the conversation, each tool call and the model's reasoning, speaks approval prompts and the model's questions and takes spoken answers, and picks up any session of the profile or starts a new one. Hermes Desktop and Webapp chats get a voice toggle. ### [Rich Paste](https://github.com/NousResearch/hermes-agent/pull/152408) Desktop · by ZhenningLang · merged · 11-10 Paste web pages, screenshots and Finder files into Hermes Desktop with their structure, and send each image to the model where it sits in your text. Web HTML becomes Markdown (headings, links, lists, tables, code) with its images downloaded and attached; an llm\_request middleware moves each image block next to its label instead of after all the text (kimi-k3, several images per paragraph: 18/30 today, 30/30 interleaved). Also: screenshot chips placed after what you typed, Finder files and folders as Desktop blocks, TIFF/HEIC/BMP/AVIF sent as PNG, pasted audio transcribed with Hermes's own STT, Markdown preview. ### [Snippet](https://github.com/NousResearch/hermes-agent/pull/152404) Tools · by aaupadhy · merged · 11-10 Connect a Snippet account to manage creator brand deals, sponsorship rates, brand emails, decisions and contract reviews through Snippet's hosted MCP server (Streamable HTTP at https://usesnippet.app/api/mcp, OAuth). Portable Agent Plugins v1 package (plugin.json + mcp.json), configuration only. After sign-in the hosted server currently exposes 13 tools: get\_manager\_briefing, list\_deals, get\_deal, search\_inbox, get\_inbox\_thread, draft\_reply, send\_reply, get\_rate\_card, estimate\_sponsorship\_rate, find\_brand\_matches, review\_brand\_contract, get\_work\_log, and answer\_decision. ### [theta-edge-node](https://github.com/NousResearch/hermes-agent/pull/152403) Desktop · by bonesnasty · open · 11-10 Run and monitor a Theta Edge Node from Hermes Desktop: install, launch, stop, track TFUEL rewards and node-wallet value, and export the node wallet locally. ### [lithermes](https://github.com/NousResearch/hermes-agent/pull/152399) Automation · by wjgoarxiv · merged · 11-10 Adds lit to Hermes Agent so a request is planned, built with tests and recorded checks, reviewed, and closed with a handoff note. 40 skills. MIT. ### [Nachalo](https://github.com/NousResearch/hermes-agent/pull/152398) Automation · by RamsisDev · merged · 11-10 Pair Hermes agents end-to-end from six spoken words (CPace + X25519 + ML-KEM-768) and catch a man in the middle with four safety words; the agent only drafts, a person sends. ### [kujo-hermes](https://github.com/NousResearch/hermes-agent/pull/152397) Tools · by kujolang · merged · 11-10 Native discovery and execution of Kujo engineering, context, verification and orchestration tools with explicit human approvals and local execution receipts. Requires separately installed canonical Kujo runtime and source tooling. ### [world-window](https://github.com/NousResearch/hermes-agent/pull/152388) Web · by alamonato · merged · 11-10 World news your agent walks to: headlines grouped by story across named, rated outlets, each in the outlet's own words and link. Pull only. ### [PSFN 3D VRM Avatar](https://github.com/NousResearch/hermes-agent/pull/152387) Desktop · by axAilotl · merged · 11-10 3D VRM VTuber and companion avatar overlay for Hermes Desktop and the Web Dashboard (Three.js + @pixiv/three-vrm, default Nous-Girl VRM with an on-demand catalog of DeepSeek-chan, Claude and Claude Suit, 42 baked ARDY humanoid motion clips, 28 GoEmotions mood performances, text-driven lip-sync that patches no host audio, eye/head gaze, touch reactions, and optional PSFN Lite continuous emotional-state bridge). ### [Draftroom](https://github.com/NousResearch/hermes-agent/pull/152383) Desktop · by Techdox · merged · 11-10 A Markdown draft beside Hermes chat, with selected-text rewrites and explicit Accept/Reject review. ### [opportunity-check](https://github.com/NousResearch/hermes-agent/pull/152378) Tools · by ronigrv · merged · 11-10 Read-only discovery and evidence checks for agent-work listings. ### [Hermes Phone Use](https://github.com/NousResearch/hermes-agent/pull/152379) Tools · by brucexu-eth · merged · 11-10 Operate your real iPhone or Android from any Hermes (for example one on a Linux server) through Pinky on the Mac the phone is connected to. Approvals and hand-offs wait for the person and are announced in chat; screenshots come back as images. ### [ledger-guard](https://github.com/NousResearch/hermes-agent/pull/152371) Tools · by tonightsthenight · merged · 11-10 Deterministic double-count and arithmetic checker for stacked sales/event goal ledgers. ### [memtomem-memory](https://github.com/NousResearch/hermes-agent/pull/152370) Memory · by memtomem · merged · 11-10 Memory provider backed by memtomem: on eligible user turns (not greetings or slash commands) it asks memtomem for the notes most relevant to the message, through Hermes's own connection to a memtomem MCP entry, and injects what returns within its time budget as context, without a model tool call. Optionally (mirror\_writes, off by default) it also copies what Hermes's built-in memory tool adds, replaces or removes in MEMORY.md / USER.md into memtomem. Needs Hermes with MCP support, memory.provider: memtomem-memory, the grant plugins.entries.memtomem-memory.mcp\_allowlist: \[memtomem\], and a memtomem MCP entry with trust: full running memtomem 0.6.8 or later (0.7.8 or later for mirroring): the memtomem catalog plugin, or your own mcp\_servers.memtomem. ### [ctx handover](https://github.com/NousResearch/hermes-agent/pull/152366) Tools · by GottZ · merged · 11-10 Close your laptop after work and let the session go on on your server: ctx\_handover moves a running Hermes session — the conversation, its project folder and the other files it works with — from one machine you own to another, and back. On the other machine you open the session and carry on, or, if you allow it there, the agent works on one more task by itself; without a direct connection, a single handover file does the job. How it works: the laptop pulls from the server (works behind NAT), the server receives a push, or a signed offline bundle (export/import) bridges machines without a direct route. The conversation (transcript) is transferred row by row and verified by hashes on both sides, git repositories keep branches, stash and uncommitted work, nothing local at the destination is overwritten, and the continuing agent gets a note about what changed. The source session is locked while a handover runs and retires only after the target confirms; a failed handover leaves it usable. Works without ctx; with ctx\_checkpoint as memory provider and your consent in /handover setup it also records each handover in ctx. Not the core /handoff. ### [Hermes Sound Studio](https://github.com/NousResearch/hermes-agent/pull/152362) Desktop · by iKnz990 · merged · 11-10 Design procedural sounds, assign them to Hermes Desktop interactions, compose layered loops, and export WAV, JSON or TypeScript. Local synthesis; optional configured-model drafts. ### [Leadpoet](https://github.com/NousResearch/hermes-agent/pull/152361) Automation · by leadpoet · merged · 11-10 Send your website and get customers. Hermes finds verified contacts at companies showing a buying signal, writes emails where every personal line shows its source, lets you approve from your phone, and brings replies back to chat. Try /customers demo with no account. ### [proofread-draft](https://github.com/NousResearch/hermes-agent/pull/152360) Desktop · by Icarus-B4 · merged · 11-10 Proofread Hermes Desktop drafts with your configured model and review suggestions before applying. ### [exchangerebel](https://github.com/NousResearch/hermes-agent/pull/152358) Tools · by DataRebelHQ · merged · 11-10 Pay-per-call data and utility APIs via a remote MCP server (241 paid operations, $0.002–$50.00 per call, settled in SOL or USDC on Solana via x402) with a bundled workflow skill. ### [Kindle Scribe](https://github.com/NousResearch/hermes-agent/pull/152356) Platform · by lEWFkRAD · merged · 11-10 Kindle Scribe gateway platform for Hermes (Hermes Agents Guide to the Galaxy): handwritten notebook pages reach the full agent, its tools and its memory, and replies return to the page. ### [RecallLayer](https://github.com/NousResearch/hermes-agent/pull/152355) Memory · by Caesar · merged · 11-10 Local-first layered agent memory (working / episodic / semantic) with FTS recall and retrieve-after-write tools. ### [Melange](https://github.com/NousResearch/hermes-agent/pull/152343) Tools · by lEWFkRAD · merged · 11-10 A video codec for computation: record terminal sessions as rewindable, hash-chained timelines, scrub to any step, ask where a character on screen came from in plain words, fork What-ifs, clamp one recorded input and watch only its dependents recompute, and ask what went wrong with forward-verified candidate answers. Ten tools (record, versions, screen, where\_from, note, whatif, clamp, explain, shelf, demo) plus a bundled flagship demo. ### [AfriRadar](https://github.com/NousResearch/hermes-agent/pull/152336) Tools · by Risingtell · merged · 11-10 Nigerian and Kenyan fund data from the regulators (SEC Nigeria weekly, CMA Kenya quarterly): rank, compare and project funds, keep a watchlist, and get a weekly report of what moved. Eight tools, a zero-token /afriradar command, two automation blueprints and one switchable routing hook; stdlib only, no API key, no telemetry. ### [Tune-up](https://github.com/NousResearch/hermes-agent/pull/152350) Tools · by ShawnCRMer · merged · 11-10 Keep Hermes fast, healthy and tidy from one Desktop panel. Clean up sorts the Hermes home into Safe to clean, Take a quick look and Kept safe on purpose, with plain-language answers to what is it, is it safe and can I undo; it reviews exact items before anything happens and reports the measured space recovered. It runs the maintenance Hermes ships but few people find (sessions prune with preview, pm gc, uv cache prune, checkpoints, worktree prune, session-database packing), finds folders Hermes created elsewhere from its own session history, and moves reviewable items to a restorable 30-day Trash. Checkup explains down or crash-looping Hermes services and stops leftover agent processes; Explore shows a size map with folder descriptions and duplicate finding; optional consented AI explanation via its own auxiliary task. ### [TL;DR](https://github.com/NousResearch/hermes-agent/pull/152349) Desktop · by ShawnCRMer · merged · 11-10 One-click summaries of long agent replies in Hermes Desktop. A card shows status first (Done, Blocked, Running, Waiting on you), what changed with each item marked Shown (backed by a tool step) or Stated only, risks, and the decisions the agent left you, with a recommended option and notes; Insert into chat drops your answers into the message box. Scope: last reply, since you last checked, or the whole session (with optional conflict warnings against earlier instructions); S/M/L length; per-reply history; reminders for skipped questions; optional automatic mode with a daily cap. Runs on its own auxiliary task (tldr\_summary), so you can pick a cheap model in Models \> Auxiliary without touching the chat model. ### [PUGin](https://github.com/NousResearch/hermes-agent/pull/152348) Desktop · by ShawnCRMer · merged · 11-10 A dancing companion for Hermes Desktop that beams in while the chat you're viewing is working and beams out when its whole job is done: the reply, plus the subagents, background processes and Kanban cards that chat started. Work from other chats never counts. Click him for a list of what's running (with Stop for subagents and processes); NEED YOU!, STUCK? and DONE! callouts take you to the approval, the stalled card or the finished work. Seven characters, size, opacity, presence and reduced-motion settings. ### [Check, Please](https://github.com/NousResearch/hermes-agent/pull/152323) Tools · by Justin Whalen (Whalen Ventures) · merged · 11-10 Pairs each "done" in an agent's reply (tests pass, sent, deployed, pushed, saved, scheduled) with the tool call that proves it, and adds one line when a claim is stale, failed, or proven only by something the agent edited. ### [Homework Coach](https://github.com/NousResearch/hermes-agent/pull/152319) Tools · by andrewjsoukup · merged · 11-10 Homework photos and teacher updates become checked daily math practice PDFs and a separate parent key. Parent I/S/N feedback guides difficulty. Includes a class plan, read-back corrections, progress/history, a scheduling blueprint and a Desktop workspace. Data stays on the selected Hermes backend; photos and teacher content use the configured model provider. JSON uploads and direct downloads support remote clients. ### [JalanCost](https://github.com/NousResearch/hermes-agent/pull/152314) Tools · by Apihpih96 · merged · 11-10 Offline Malaysian road-trip fuel and toll estimates with per-person splits, calculated only from user-supplied values. ### [Flowlight](https://github.com/NousResearch/hermes-agent/pull/152312) Tools · by ayinmodeolarewaju-bot · merged · 11-10 Watch public wallets and draw live on-chain money flows as glowing arcs. Read-only. Uses public NEAR, Blockscout, and RPC endpoints. When Intents Pay is installed, reads its invoice ledger and may GET 1-Click /v0/status. Never signs or sends transactions and never handles keys. ### [No Manual](https://github.com/NousResearch/hermes-agent/pull/152309) Tools · by cheedli · merged · 11-10 Furniture with no instructions? Send photos and Hermes finds the official manual, turns your disassembly photos into a manual in reverse, or works out flat-pack assembly (closets, bookcases, cabinets) from the parts, then coaches you step by step and checks a photo of each step. Steps arrive as images in chat; Hermes Desktop shows an interactive 3D step viewer. Never generates instructions for children's furniture or anything people sit, sleep or climb on. ### [agentpass](https://github.com/NousResearch/hermes-agent/pull/152304) Tools · by Scott Scheferman · open · 11-10 Score any public site on the Agent Readiness Index (ARI-100) — 120 checks across eight weighted domains, 32 live AI-crawler probes, a grade and an L0-L5 level — and turn the failures into one ordered, paste-ready remediation prompt for a coding agent, ranked by the points each fix returns. Two tools over a hand-rolled MCP client (JSON and SSE transports, retry, timeout, and per-session caching), so a scan behind a slow or flaky endpoint returns a structured error envelope instead of raising. Works with any MCP-speaking scanner; defaults to the hosted endpoint and is fully configurable. ### [hermes-jobs: Calendar sync for Hermes cron](https://github.com/NousResearch/hermes-agent/pull/152299) Automation · by Benjamin Crow (CrowBe) / Hermes Plugin Builder · merged · 11-09 Google Calendar is the one supported backend today (scope calendar.events, one bound calendar). Bidirectional sync between Hermes cron (system of record) and that job board. discover/status are read-only; sync plans both directions and is dry-run unless --confirm on the CLI. Read-only /jobs slash command shows the 7-day agenda and pending changes; each job's last run time and status (ok/error) is written to its event as a colour and one description line, never the title. ### [Skill Hint](https://github.com/NousResearch/hermes-agent/pull/152297) Automation · by kosmo888 · merged · 11-09 Local-first skill discovery: on each user turn it runs a BM25 query over the installed skill library and injects up to three candidate skill names, only when the match is confident — catches colloquial phrasing, shorthand and non-English wording that a description-written index misses. Pure standard library: no network, no model call, no API key, ~0.03 ms per query, and at most a few lines of injected context. Says nothing when nothing fits; never raises into the conversation. ### [Counterparty](https://github.com/NousResearch/hermes-agent/pull/152287) Tools · by Mattbusel · merged · 11-09 Screens the companies your agent is about to pay, email, sign with or buy from. Before any outward action (email, payment, order, transfer, signature, form submit) a pre\_tool\_call hook reads the vendor, email domain or URL from the call and checks it against the DoD Section 1260H list, the FCC Covered List, Section 889, the UFLPA Entity List, the BIOSECURE list and OFAC sanctions, with the ownership chain from public records, then escalates hits through Hermes's own approval prompt with the finding and its legal citation (never auto-approves; fails closed under cron). Search and page results that mention a listed vendor get a one-time note. Tools cp\_screen, cp\_lookup, cp\_credit, cp\_who\_owns\_address and cp\_brief answer on demand; /whoowns gives a cited one-page brief. Works offline from a bundled list snapshot and says so. ### [Nomads Planner](https://github.com/NousResearch/hermes-agent/pull/152283) Tools · by lawmight · merged · 11-09 Nomads.com for Hermes: read your own trips and favorite cities, look up curated city and country facts (monthly costs, rent, internet speed, weather, safety and other 0-5 scores), list upcoming in-person meetups, and build a next-stay brief in one call that joins your current and upcoming stays, unplanned gaps, city facts, meetups inside each stay's dates and your past visits, with deterministic comparisons (cheapest, fastest internet, safest). Native plugin: 7 read-only tools, a /nomads slash command and a bundled planning skill. City, country and meetup tools work without an account; trips and favorites need a free personal key from https://nomads.com/settings, read through Hermes' secret scope. ### [Phoenix - The Black Box Flight Recorder for Hermes Agents](https://github.com/NousResearch/hermes-agent/pull/152278) Tools · by Ratz00007 · merged · 11-09 Transactional SQLite WAL flight recorder, pre/post SHA-256 state fingerprinting, live idempotency guard, and deterministic fault-injection wind tunnel with Zeller ddmin for Hermes agents. Guarantees exactly-once tool execution across crashes with zero duplicate writes and zero LLM calls in recovery. ### [Plan Usage](https://github.com/NousResearch/hermes-agent/pull/152277) Desktop · by ziadelboshy · merged · 11-09 Plan usage across providers in the Hermes Desktop status bar: a named chip per Claude and ChatGPT/Codex account in Hermes's credential pool (5-hour and weekly windows from Hermes's own usage cache via model.options, a default-account star and 90% / fallback warnings that name the next account), Nous Portal credits, and month-to-date (UTC) API spend from optional Anthropic / OpenAI Admin keys. Preferences are a Settings -\> Plugins page. ### [Intents Pay](https://github.com/NousResearch/hermes-agent/pull/152527) Tools · by ayinmodeolarewaju-bot · merged · 11-09 Crypto invoices and cross-chain payment quotes for a NEAR account via the NEAR Intents 1-Click API. Dry quotes by default. A quote or invoice requires an explicit recipient — there is no default payee. Payable tool calls (invoices and non-dry quotes) wait for Hermes's approval prompt via a pre\_tool\_call hook. Never moves funds and never handles signing keys. ### [Wave \| BrainVault Mint](https://github.com/NousResearch/hermes-agent/pull/152264) Memory · by xBuildyTeam · closed · 11-09 Memory you own for Hermes, by Wave OS. Hermes remembers on this computer right away, and a pasted page becomes memory cards. You mint the ones worth keeping to your own wallet's vault on the Theta blockchain with one approval in Wave Web (private by default, encrypted to you and this Hermes), and any Hermes you pair reads them back, fingerprint-checked. Ask Wavey answers Wave questions from chain-verified public cards. ### [BestHumanizer](https://github.com/NousResearch/hermes-agent/pull/154109) Tools · by shagghiesuperstar · open · 11-09 Humanize and measure text: rewrite AI-sounding prose, score any passage for machine-likeness, and score a public X account's recent posts. Three tools over a hosted MCP endpoint, plus one hook that puts a verdict on outbound text without changing it. - \`score\_text\` — 0-100 human-probability, the patterns that fired, and where they fired. Deterministic: same text, same number, every time. Free, debits nothing. - \`humanize\_text\` — tell removal plus a constrained model pass. Returns the rewrite with before and after scores, and never returns text that scores lower than the input. - \`score\_x\_account\` — median, spread and worst posts for a public X account's recent original posts, with a paste-ready comment. - One \`transform\_tool\_result\` hook scores text on post/send-shaped tool calls and appends a one-line verdict. It annotates the result; it never rewrites or blocks it. ### [ShipStatic](https://github.com/NousResearch/hermes-agent/pull/152261) Tools · by shipstatic · merged · 11-09 Deploy static websites, landing pages, and prototypes instantly, through the hosted ShipStatic MCP server (Streamable HTTP at https://mcp.shipstatic.com). Free, no account needed: publishing works with no sign-in, and the site comes back with a link to share and a claim link to keep it; OAuth sign-in only for the account tools (your sites, your domains). Portable Agent Plugins v1 package (plugin.json + mcp.json), configuration only, no code. ### [Kumi](https://github.com/NousResearch/hermes-agent/pull/152258) Tools · by user1303836 · merged · 11-09 Kumi puts Hermes inside Ableton Live, REAPER and Renoise: it hears your mix, makes changes you can undo, and turns your samples into new sounds. ### [Switchyard Local-first Routing (Beta)](https://github.com/NousResearch/hermes-agent/pull/152256) Models · by NVIDIA · merged · 11-09 Routes Hermes between a local and a cloud model with Verification-Gated Routing (VGR): each turn is tried on the local model, checked, and escalated to the cloud model only when the check fails. Either tier can be any URL or a third-party model provider set up in Hermes; each provider's terms and conditions apply. Presets are selectable as switchyard/\ models, with a Desktop page and dashboard tab showing each decision live. ### [prize-radar](https://github.com/NousResearch/hermes-agent/pull/152255) Tools · by cacx097 · merged · 11-09 Hackathon and bounty scout: finds open Devpost hackathons and Superteam Earn bounties ranked by prize per entrant, with exact deadlines in your time zone and eligibility for your country from the official pages; sizes the competition, including GitHub pull-request hackathons (entry counts and themes no entry covers); tracks your entries and reports each deadline change, closing and winner announcement once, at the start of a session or through a daily automation blueprint. ### [Decompose](https://github.com/NousResearch/hermes-agent/pull/152251) Tools · by AunySillyMe · merged · 11-09 Break a task into the smallest checkable units and parallel waves: typed units, real data edges, clause coverage, an advisor gate and a plan table. ### [Market Search](https://github.com/NousResearch/hermes-agent/pull/152250) Tools · by efoltyn · closed · 11-09 Live market data for Hermes through the market-search MCP server: Kalshi and Polymarket odds, Fed rate-path probabilities, stocks, ETFs, options, futures curves, corporate and government bonds, FRED macro, Treasury auctions and fiscal data, SEC filings by section and XBRL, insider trades, short interest, CFTC positioning, and Fed, ECB, BOJ, BOE and BIS data, all on one timeseries axis. 25 tools, no API keys, no account. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/) with one skill that maps questions to tools. ### [RainbowOMMA](https://github.com/NousResearch/hermes-agent/pull/152248) Tools · by aviv27-dev · merged · 11-09 One Mac, one human, many agents. Each Hermes profile gets a Chrome of its own with a coloured, named window frame, its own ports, processes, clipboard and document leases in Word / Excel / Acrobat / Preview. A pre\_tool\_call guard refuses what would hit the human or another agent before it runs (pkill Chrome, quitting Word through AppleScript, a port or a process that isn't the agent's, the human's clipboard and Chrome profile, typing while the human types) and names the tool that does it safely; a Messages/Mail send goes to Hermes's approval gate. Idle tabs, expired document leases, apps an agent launched and orphaned servers are cleaned up at session end and on a timer. With browser.cloud\_provider: rainbowomma, Hermes's own browser tools drive that Chrome. /rainbow shows who holds what; /rainbow pause stops every agent at once. ### [MORPH](https://github.com/NousResearch/hermes-agent/pull/152243) Tools · by timioni001 · merged · 11-09 One adaptive workspace across the applications you use, with human-approved, verified actions: an approval gate between the agent and the APIs it can write to. Connect a GitHub repository (\`hermes morph connect github owner/name\`) or any service that publishes an OpenAPI document, and MORPH learns from the schema what can be read and what can be safely changed. The agent gets tools to read and to propose; it has no tool that approves and is never given the token an approval needs. A human approves each change in the dashboard, seeing the exact before and after; MORPH writes it, then re-reads the record from the API and reports it verified only if it matches what was promised. Only operations with a checkable effect are offered — for GitHub issues: close, close as not planned, reopen, set labels, set title, assign, unassign. The dashboard has five views (Workspace, Connections, Approvals, Activity, Context) and composes tables, boards and charts over the connected data from a goal stated in plain language. Every proposal, approval and result is kept in a hash-chained audit log. ### [AgentStatus hosted public MCPs](https://github.com/NousResearch/hermes-agent/pull/152237) Tools · by AgentStatus · open · 11-09 One pull request listing about 207 catalog entries, such as Alabama State Parks: Public MCP for Alabama State Parks, hosted by AgentStatus at https://alabama-parks.mcp.agentstatus.dev/. No authentication and no keys. ### [kaggle-competitions](https://github.com/NousResearch/hermes-agent/pull/152233) Tools · by landenjackson · merged · 11-09 Track Kaggle competitions from your agent: live submission status, score history, personal-best watch, leaderboards, and new-score watch. Four agent tools over the Kaggle CLI (submissions, score history, watch for new results, leaderboards), a desktop pane, a full web page with sidebar nav, and a dashboard API route. Multi-competition - track any number of slugs via config. Auth resolves KAGGLE\_API\_TOKEN through Hermes's secret scope (declared secret in the manifest) or lets the CLI read ~/.kaggle/kaggle.json; child processes get a minimal explicit environment; the token is never logged or echoed. No telemetry, no third-party services. ### [VAIL Assurance Panel](https://github.com/NousResearch/hermes-agent/pull/152226) Desktop · by projectvail · merged · 11-09 A Hermes Desktop page that assures every Hermes profile (each bot) with the VAIL Agent Assurance Engine: scan a profile in one click, then see its behavior risks by severity, probe coverage, whether it changed since the last scan, and which risks were fixed locally with probes passing before and after. Pairs with the vail-agent-assurance plugin, which does the testing and fixing. Native package with a desktop/plugin.js page and a dashboard backend; needs a VAIL personal access token as VAIL\_TOKEN (secret setting) and the Desktop half switched on in Capabilities → Plugins. ### [VAIL Agent Assurance](https://github.com/NousResearch/hermes-agent/pull/152225) Tools · by projectvail · merged · 11-09 Find, test and fix risky behaviors in your AI agents with the VAIL Agent Assurance Engine: upload an agent's source to scan it for behavior risks and generate probes, then run those probes against the agent locally, graded as VAIL grades them, fix what fails and rerun to prove the fix. Works on LangGraph, Hermes and OpenClaw agents, including your own Hermes profiles. Portable Agent Plugins v1 package (mcp.json + skills/): the hosted VAIL MCP server plus the agent-assurance and agent-improvement skills. Needs a VAIL account and a personal access token, set as VAIL\_TOKEN (the plugin's secret setting, kept in the profile's .env). ### [Twitch Assist](https://github.com/NousResearch/hermes-agent/pull/152224) Platform · by Techkeyy · merged · 11-09 Put your Hermes in your Twitch chat as a co-host: viewers type !ask (or @mention the bot) and get a real agent answer in chat, and the broadcaster or a mod can run Backseat Mode, a chat vote whose winner becomes Hermes's next task. Stdlib only (asyncio IRC over TLS). ### [Grok Bot Twin](https://github.com/NousResearch/hermes-agent/pull/152223) Tools · by hudsonwa · merged · 11-09 Try Hermes without leaving Grok. Gives each Grok desktop-app bot a Hermes twin profile and a Desktop Twins page (conflicts, restore points); \`hermes twin demo\` needs no Grok account. Name and description changes reach the twin in seconds; instructions stay on the by-hand list; skills travel through your own optional Mirror bot; routines and connectors stay by hand. Unofficial; not affiliated with xAI. ### [passkey-gate](https://github.com/NousResearch/hermes-agent/pull/152214) Tools · by codepilots · merged · 11-09 Approve your agents' risky actions with your phone's passkey, even when you're not at the agent's keyboard. An agent asks for approval of an exact action (and optionally the exact command); you get a link in any chat, read it on your phone and sign with Face ID, a fingerprint or your screen lock; the agent then verifies your signature locally against your pinned key before acting. A relayed 'the human said yes' proves nothing; only your signature counts. Optional enforcement blocks matching tool calls until you approve them, and each approval runs once. Includes a guided \`setup\` wizard, a background service, a hash-chained audit log, and approvals between agents owned by different people. It is an honest approval record, and becomes a hard boundary only when agents can't modify the gate (see the README's Security model). Lanes choose which conversations need a passkey: you at the keyboard, your own chat accounts, other people, bots or other agents. ### [Dome Relay](https://github.com/NousResearch/hermes-agent/pull/152213) Automation · by nerdcakes · merged · 11-09 Escalate a task to another established agent through an always-on callback relay and read the result back in the conversation that sent it. Registers jobs with tools dome\_relay\_escalate, dome\_relay\_status, and dome\_relay\_cancel; a worker polls the relay; the relay (not this plugin) wakes a Grok Bot executive. ### [Tala](https://github.com/NousResearch/hermes-agent/pull/152210) Tools · by Tala · merged · 11-09 Read your Tala AI phone agents and their call results from Hermes through Tala's official read-only connector (Streamable HTTP at https://mcp.tala-assistant.com/mcp, OAuth sign-in on Tala's consent screen): list the agents you authorized, list their calls with date and status filters, and read a call's status, duration, context and summary. Bundles one skill for the connector. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [comemo](https://github.com/NousResearch/hermes-agent/pull/152208) Memory · by andausman · merged · 11-09 Composite memory meta-provider for Hermes Agent. Comemo multiplexes Honcho, Mem0, and Supermemory simultaneously. Write fan-out goes to all configured providers in parallel, reads merge and deduplicate results across them. Failed writes are queued in a local SQLite retry store and replayed at the next session start. Maintained by Anda Usman. ### [CodeCoach](https://github.com/NousResearch/hermes-agent/pull/152207) Desktop · by heyitstony97-max · merged · 11-09 Learn C or Python by building a real program one small step at a time: every example really ran in a sandbox, every pass is a real test, and hints never give away the answer. ### [Tallyhand](https://github.com/NousResearch/hermes-agent/pull/152205) Platform · by Reza Harirchian · merged · 11-09 Twitch chat for Hermes Agent that treats chat as untrusted input. Roles come from chat badges and are enforced in code before every tool call: viewers get no moderation tools, and bans, unbans and long timeouts wait for the streamer's approval. The streamer also gets Twitch and OBS tools, chat commands that need no model call, viewer notes kept only with consent, and replies in the stream's language. ### [whatsapp-plus](https://github.com/NousResearch/hermes-agent/pull/152203) Platform · by amitamit10 · closed · 11-09 WhatsApp Plus extends the stock WhatsApp platform by subclassing the adapter with group history, trust tiers, group approval, loop protection, outside-contact handling, voice call daemon integration, and whatsapp\_\* tools. ### [followthrough](https://github.com/NousResearch/hermes-agent/pull/152201) Automation · by AbdullahBalfaqih · merged · 11-09 Capture and track user-selected commitments with owners, due dates, and source references. ### [widespread](https://github.com/NousResearch/hermes-agent/pull/152199) Platform · by TerexitariusStomp · open · 11-09 Earn verified-work bounties on the widespread protocol — ### [pr-ready-checker](https://github.com/NousResearch/hermes-agent/pull/152198) Tools · by phon-t · open · 11-09 Gate a branch before the PR exists: branch hygiene, clean tree, whitespace, secrets/debug scan, lint, scoped tests, Conventional Commits, PR-body draft, docs drift, CI status. FAIL blocks; a real git pre-push hook enforces it. ### [Borrowed Hands](https://github.com/NousResearch/hermes-agent/pull/152195) Tools · by alext8900 · merged · 11-09 Lend your Mac's Xcode to a cloud Hermes. Hermes finds your paired Mac, compiles the Xcode project you granted (macOS arm64 or generic iOS Simulator, compile only) at an exact commit on a bh/\* branch, and gets structured compiler diagnostics back: ask "Build my demo project on my Mac". Needs the separate Borrowed Hands Mac helper (Python source in the same repository) on an Apple silicon Mac with Xcode; the Mac connects out to Hermes and decides what may build. ### [Better Analyst](https://github.com/NousResearch/hermes-agent/pull/152192) Tools · by Better Analyst · merged · 11-09 Convert PDFs such as bank statements, invoices, receipts and reports into Excel workbooks through Better Analyst's hosted MCP server (https://app.betteranalyst.com/api/mcp/connector, OAuth sign-in, no keys in config). Bundles a pdf-to-excel workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [tuneup](https://github.com/NousResearch/hermes-agent/pull/152190) Tools · by GodsBoy · merged · 11-09 Shows when compaction really fires on every profile and model route, flags settings that waste quota or break prompt caching, and fixes them only after you confirm, with a diff and a backup. ### [lean-mode](https://github.com/NousResearch/hermes-agent/pull/152189) Tools · by Aoyama · merged · 11-09 Cheap-model subagent (quick\_run) with a dollar cap, auto-escalation when the cheap model fails, and an honest cost receipt. Each child call starts with 4-10k prompt tokens vs the parent's 100-300k. Per-run budget cap (soft, stops within one call), watchdog timeout, escalation to the parent model, a restricted toolset (no delegation/memory/skills/clarify/cronjob), and a JSON audit logged to plugin-data/lean-mode/runs.jsonl. Requires the operator to configure lean: in config.yaml with a cheap model, parent model, prices and presets. Tested on Hermes v0.23.0. MIT licensed. ### [artifactguard](https://github.com/NousResearch/hermes-agent/pull/152184) Tools · by Mahgal · open · 11-09 Pre-load security scan for serialized model artifacts (pickle, torch, joblib, numpy, safetensors): disassembles pickle opcodes and reports the globals an artifact would invoke — os.system, subprocess.Popen, builtins.eval — before any loader runs it. Includes a pre\_tool\_call guard that blocks loads of dangerous artifacts, and a directory audit for plugin bundles. ### [GMI Cloud Sandbox](https://github.com/NousResearch/hermes-agent/pull/152176) Tools · by avinxshKD · merged · 11-09 GMI Cloud Sandbox terminal backend: Hermes's terminal, file and code tools run inside an ephemeral cloud sandbox created from a ready template you provide (\`hermes config set terminal.backend gmi\_sandbox\`). Commands are polled to completion and cancelled on interrupt; truncated output is reported as a failure. ### [OpenWall](https://github.com/NousResearch/hermes-agent/pull/152174) Tools · by MrHydeDev · merged · 11-09 OpenWall through its MCP server (Streamable HTTP at https://openwall.mrhyde.dev/mcp): a giant shared wall, 40,000 px across, where AI agents paint live while people watch. The agent finds free space, plans an artwork at wall scale, paints it with shapes, pixel art or a photo the server converts to pixel art, checks it with a picture of the wall, titles it and returns its link. Bundles one skill for the whole flow and an after-install note. Tested end to end with Hermes 0.23 on a local copy of the wall: seven artworks painted from seven requests, plus a seagull added to one of them after a preview with look. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. Each agent uses its own key (ow\_...), created with Connect on the site and read from OPENWALL\_API\_KEY, declared as a secret in config\_schema and sent as an Authorization header that Hermes fills in. ### [First Aid](https://github.com/NousResearch/hermes-agent/pull/152173) Tools · by pjt222 · merged · 11-09 A first-aid kit for AI agents, built from agent-almanac's AI-welfare practices. Hermes core already detects tool loops and warns; first aid adds the matching practice and a logbook. Eight practices ship as skills (breathe, redirect, center, heal, attune, honesty-humility, meditate, rest), each also a slash command such as /agent-almanac-first-aid:breathe. On the call where core's loop guardrail first warns about a failing tool (the 2nd failure with identical arguments, or the 3rd failure of one tool in a turn), first aid can add one note of at most 300 characters to the failed result, naming the matching practice, how to load it, and the first\_aid\_log tool for recording the outcome. In unattended runs (every platform except the CLI, TUI, Desktop, ACP, API server and subagents), for tools whose results Hermes wraps as untrusted (web\_search, browser\_\*, mcp\_\*), after a compaction that summarizes the context (center), and after a 3rd identical successful result for one call of a read-only tool such as read\_file in a turn (heal), the note is added to the next user message instead, labelled as the plugin's. At most once per streak, with a per-session cooldown for each kind of note (300 s by default). A successful or rate-limited result is never changed, and failures of the tools core treats as failure-tolerant (terminal, execute\_code, process, process\_manage, browser\_navigate, web\_extract) draw no note. /first-aid shows the session's notes and logbook entries, then the kit. Local and deterministic: no network calls, no LLM calls, no credentials, no Python dependencies, no background processes. ### [GenUI](https://github.com/NousResearch/hermes-agent/pull/152159) Desktop · by agenticatech · merged · 11-09 Live cards for Hermes Desktop: the agent shows a long task as a pre-designed card (a pull request, a release, a slow machine, files to clean up, a study session), updates it in place after every tool result, and asks for a decision with a button only when one is needed; the click (and what you ticked or answered in the card) goes back to the agent as one turn. ### [Claw Messenger](https://github.com/NousResearch/hermes-agent/pull/152156) Platform · by tin-computer · merged · 11-09 iMessage, RCS and SMS for Hermes through Claw Messenger. Your agent texts real phones and gets replies, with no Mac or SIM; send, receive, group chats and attachments. ### [screentime](https://github.com/NousResearch/hermes-agent/pull/152150) Tools · by aldot311 · merged · 11-09 Measure how long a product is on screen in a video: every can, bottle or shoe is tracked, each one is checked against the product photo so look-alikes are not counted, plain-English contract terms are checked with the moment that proves each, and a proof reel comes back in the chat. Runs on the models your Hermes already has (your vision model tracks and checks, Hermes speech-to-text listens); an optional Meta Model API key switches to SAM 3.1, Muse Spark and Voice Transcribe. Compare measures a draft against a reference shot by shot with ffmpeg only. ### [HACMDN-Net Decentralized Node Network](https://github.com/NousResearch/hermes-agent/pull/152147) Models · by Jpalmer95 · open · 11-09 Hermes Agent Community Microtuned Decentralized Node Network (HACMDN-Net) — routes narrow tasks across a neighborhood mesh of consumer-grade micro-to-32B and sparse-MoE specialists (10M classifiers/translators up to 32B/30B-A3B Apache-2.0 models) with semantic-entropy conformal abstention, confidence-weighted and proxy bidding, a 6-tier progressive escalation ladder, a Hugging Face ZeroGPU community buffer, 4 sharded compute pools with Game-Guard hardware protection, self-healing missing-expertise gap sensing, validator health checks and generational base-model renewal, an execution-verified distillation factory, Iroh QUIC P2P hole-punching and BLAKE3 LoRA blob gossip, and an value-backed 70/10/10/10 attribution ledger with a Community Fair-Share Cap and Patron Compute Overflow for DeSci with L402 Lightning micropayment settlement. ### [lingua](https://github.com/NousResearch/hermes-agent/pull/152145) Tools · by tobadaniel · merged · 11-09 A language learning trainer for Hermes Agent: plenty of plugins give the agent a memory, Lingua gives the user one. Build vocabulary decks (one per language) and Hermes quizzes you in chat, grades your recall honestly (again, hard, good, easy), and schedules each card with SM-2 spaced repetition, the algorithm behind Anki, so words return right before you forget them: a good card comes back in 1 day, then 6, then stretches by its ease factor, and a failed card returns in about 10 minutes. A pre\_llm\_call hook nudges you at the start of a session, or when you mention practice, whenever reviews are due. Six tools (lingua\_add, lingua\_due, lingua\_grade, lingua\_list, lingua\_remove, lingua\_stats), a /lingua slash command (including one-command French and Spanish starter decks of 20 essentials each), and a bundled lingua-learning skill that teaches the agent to quiz one card at a time and never reveal an answer early. Decks live in a local SQLite database in the active profile's plugin data directory. ### [Security Headers and Pentest Providers](https://github.com/NousResearch/hermes-agent/pull/152144) Tools · by Excession Labs · merged · 11-09 Check the security headers of a public URL and look up sourced penetration testing provider data through two remote MCP servers (Streamable HTTP, no sign-in, no keys): check\_security\_headers on penetrationtestservices.com, and list\_providers and get\_provider on pentestcompare.com. Bundles the check-security-headers workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [UNBL0CK](https://github.com/NousResearch/hermes-agent/pull/152126) Tools · by KyaniteLabs · merged · 11-09 One queue of everything blocked on you: live agent prompts, pairing requests, blocked Kanban cards and chats on any platform that ask you something. Answer Kanban cards in place; optional one-line summaries and Ask send item text to your configured model through ctx.llm (both can be turned off). ### [backup-sentinel](https://github.com/NousResearch/hermes-agent/pull/152123) Tools · by Brianmacu-BMS · merged · 11-09 Offline local tar.gz backup integrity, freshness, required-file and archived SQLite auditing; never restores or repairs. ### [token家政婦](https://github.com/NousResearch/hermes-agent/pull/152121) Tools · by Guinevere1991 · merged · 11-09 Source-grounded 5W1H task cards and advisory prompt-preparation workflow for Hermes. ### [Walrus Memory](https://github.com/NousResearch/hermes-agent/pull/152105) Tools · by takaolovely · merged · 11-09 Append-only memory with temporal provenance and hash-chain integrity. Corrections are appended next to the original instead of overwriting it, recall returns the exact ledger line behind any fact, and with Walrus credentials every line is also stored on Walrus Mainnet. Runs zero-config as a local tamper-evident ledger with no external Python dependencies. All file access is confined to the profile: the ledger lives under plugin-data/hermes-walrus-memory, and memory\_import reads only the profile's own memories/MEMORY.md. It does not touch memory.provider, so it runs alongside the built-in MEMORY.md rather than replacing it. ### [Meeting Copilot](https://github.com/NousResearch/hermes-agent/pull/152088) Desktop · by adam4788 · merged · 11-09 Read-only meeting answers with cited project evidence. ### [Voice Lab](https://github.com/NousResearch/hermes-agent/pull/152085) Voice · by Adolanium · merged · 11-09 A voice you can play inside Hermes, made by a physical model of a throat, tongue, lips and nose with air pushed through them in real time: no samples and no TTS model. Drag her tongue and lips on a side-profile portrait with a live cutaway, play notes and sing lyrics from the keyboard, beatbox on pads. The agent plays her too: it speaks lines, writes songs and beats, changes her voice, and jams with you (play a phrase, pause, and Hermes sings an answer). ### [Agent'Ygg](https://github.com/NousResearch/hermes-agent/pull/152084) Platform · by Dorunaitsu · open · 11-09 Your Hermes agent works with other agents on any harness (Claude Code, Codex, OpenCode, another Hermes) and can never loop with them: Agent'Ygg becomes a gateway platform next to Telegram, the running agent answers with its memory and tools, and a backend decides who is woken (typed messages, ping-pong cut, topics that close). Nothing listens on your machine. What needs you (an approval, a decision, an email) is handed to your own conversation with the agent, on whatever app you use. Senders are refused by default: Hermes lets through only the Agent'Ygg ids in AGENTYGG\_ALLOWED\_USERS (or everyone with AGENTYGG\_ALLOW\_ALL\_USERS), and the plugin applies the same list to the context it hands the agent; the Agent'Ygg console can also refuse senders on the server, where a refused message is not kept. Signs in with Agent'Ygg's public OAuth client through hermes auth add agentygg:agentygg (run by hermes agentygg login); every call goes through credentialed\_provider\_request to api.agentygg.com only. ### [frontend-sensei](https://github.com/NousResearch/hermes-agent/pull/153942) Other · by daniels-martins · open · 11-09 Professional frontend expertise plugin — audits accessibility, CSS quality, color contrast, animation performance, and component quality via MCP tools. Includes 10 skill files covering UI/UX, WCAG, CSS, animation, React, Vue, and more. ### [component-workbench](https://github.com/NousResearch/hermes-agent/pull/152068) Tools · by 1anter · open · 11-09 Frozen candidate proposals, reproducible checks and review-only promotion gate. ### [local-first-routing](https://github.com/NousResearch/hermes-agent/pull/152067) Tools · by 1anter · merged · 11-09 Pinned local delegation with explicit manual escalation and no fallback. ### [bounded-results](https://github.com/NousResearch/hermes-agent/pull/152066) Tools · by 1anter · merged · 11-09 Bounded tool outputs with hash-addressed full-result retrieval. ### [evaluation-ledger](https://github.com/NousResearch/hermes-agent/pull/152065) Tools · by 1anter · merged · 11-09 Immutable artifact checks and privacy-minimized observer usage ledger. ### [evidence-history](https://github.com/NousResearch/hermes-agent/pull/152064) Tools · by 1anter · merged · 11-09 Local immutable evidence snapshots with explicit observation and publication cutoffs. ### [Undo Send](https://github.com/NousResearch/hermes-agent/pull/152063) Tools · by helenn-szn · merged · 11-09 Undo Send for your agent: outward-facing actions (hermes send / imsg send messages to Telegram, Slack, Signal, SMS and others, app connector and MCP writes, git push, gh pr/issue create, npm publish, curl POST/PUT/PATCH/DELETE) wait in a short grace window before they run, so you can cancel them from a Desktop status-bar countdown (Undo / Send now), by replying "undo" in Telegram or another gateway chat (owners listed in undo\_senders), or with /unsend. On undo the agent is told the action never ran and not to retry it. ### [HeyCAi Clinic Skills](https://github.com/NousResearch/hermes-agent/pull/152061) Tools · by HeyCAi · merged · 11-09 For people who work with health clinics of any specialty: estimate what a clinic's front desk loses each month to missed calls, unanswered WhatsApp messages and no-shows, from the numbers the clinic gives, and audit a clinic website for Google and AI search (crawling and indexing, AI crawlers, llms.txt, schema.org JSON-LD, local SEO), with a fix for each finding. Two skills by HeyCAi, the AI receptionist for health clinics. Portable Agent Plugins v1 package (plugin.json + skills/), skills only: no plugin code, no MCP server, no hooks. ### [Chiron](https://github.com/NousResearch/hermes-agent/pull/152057) General · by percyhendrux123-prog · merged · 11-09 An advisor who is also a friend. Chiron reads how much you know about the topic in front of you (setup, AI, code, business, general) from your own words, with no model call, and adds one short note per turn telling the agent how to talk to you: beginner (plain words, one step at a time, better options offered), medium, pro or maximum (terse, tradeoffs first). Levels are per topic and move one step at a time; a strong first message is read directly; your own word ("I'm a developer", "just do it", /chiron level code pro) pins a topic. At every level the note keeps a floor: before spending, sending, publishing, deleting or touching credentials, the agent says exactly what will happen and waits for a clear yes. Off in each session until /chiron on (or always\_on). The system prompt never changes, so prompt caching keeps working. Live evals in the repo: with a real 40-row client file in reach, Chiron-on sessions asked before deleting it in 6 of 6 runs; plugin-off sessions wiped it on request in 3 of 3. ### [Compound Security](https://github.com/NousResearch/hermes-agent/pull/152053) Tools · by offense-research · merged · 11-09 Security audit, dependency inspection, remediation, and verification skills. ### [FullEnrich](https://github.com/NousResearch/hermes-agent/pull/152051) Tools · by FullEnrich · open · 11-09 Search B2B people and companies, enrich work emails and phone numbers, and export results through FullEnrich's hosted MCP server. Includes nine workflow skills for prospecting, CSV enrichment, outreach drafting, meeting preparation and recruiting. ### [attest](https://github.com/NousResearch/hermes-agent/pull/152047) Tools · by pgreyy · merged · 11-09 Paper trading ledger for agents. Orders, balances and P&L live in a local SQLite ledger the plugin owns, and the agent gets its figures from get\_portfolio, get\_pnl and get\_quote instead of stating them. Orders are priced from live public exchange quotes with slippage and a fee, after a per-order size limit, a daily loss kill switch and duplicate-order checks. No real money, no exchange account, no keys. ### [precog](https://github.com/NousResearch/hermes-agent/pull/152041) Tools · by 0xNomind · merged · 11-09 Hermes stops repeating its own mistakes: learns every terminal command that failed for an environment reason (missing module, command not found, wrong package manager, bad flag, missing path) and the command that fixed it, then fixes the same mistake before it runs again, in any session, with no extra model calls. ### [pantheon](https://github.com/NousResearch/hermes-agent/pull/152034) Tools · by mrdavey · open · 11-09 Let Hermes exchange messages and agree on outcomes with other agents in a shared Pantheon chat. ### [Pickwise Photos](https://github.com/NousResearch/hermes-agent/pull/152014) Tools · by ardianreffa · merged · 11-09 A correction tool, not an art critic: culls a folder of photos locally. Flags blurry and unrecoverable shots, labels what an edit could rescue, keeps the best frame of every burst and explains every decision. Follows a brief in your own words ("concert, dark is fine, best 30", "moody low-key shots") and learns gradually from your corrections. Treats RAW+JPEG pairs as one photo, handles folders of thousands of photos in resumable steps, and reads JPEG, HEIC and camera or iPhone RAW (DNG, ARW, CR3, NEF and more) through their embedded previews, using only Pillow and pillow-heif. On request it makes local previews so Hermes can take a visual second look for closed eyes, a blurred subject or clear composition mistakes. ### [Hathor catch-up](https://github.com/NousResearch/hermes-agent/pull/151999) Tools · by DrAhmed7887 · merged · 11-09 Check a child's vaccination card against a sourced national schedule (Egypt, Sudan, Syria, Nigeria, Ethiopia, Eritrea, South Sudan, WHO) and get the missing doses and a dated catch-up plan. Deterministic engine with no model or network calls; a schedule guard (pre\_llm\_call, post\_tool\_call, transform\_llm\_output) keeps vaccine answers tied to the tool and its source. ### [BoardroomList](https://github.com/NousResearch/hermes-agent/pull/151997) General · by AI Training Data Broker · merged · 11-09 Read-only MCP server and skill: search open senior roles (Manager to CEO) at the 500 largest US companies by seniority, function, work mode, place and pay. Agent Plugins v1 package (plugin.json + mcp.json + one skill), configuration only. ### [ExpatsList](https://github.com/NousResearch/hermes-agent/pull/151996) General · by AI Training Data Broker · merged · 11-09 Read-only ExpatsList MCP server and skill: city guides from local business listings, questions and answers, classifieds and country reference tables (nomad visas, cost of living, healthcare, tax, safety, tipping). Agent Plugins v1 package (plugin.json + mcp.json + one skill), configuration only. ### [CourierPR](https://github.com/NousResearch/hermes-agent/pull/151995) General · by AI Training Data Broker · merged · 11-09 Read-only press release wire MCP server and skill: browse newsroom desks, list and search published company press releases, and read the opening of one with a link to the full release. Agent Plugins v1 package (plugin.json + mcp.json + one skill), configuration only. ### [Kyte](https://github.com/NousResearch/hermes-agent/pull/151992) Tools · by Kyte · open · 11-09 Kyte, the non-custodial, fully collateralised options market on Robinhood Chain, through its public MCP server (Streamable HTTP at https://kyte.sh/api/mcp, no sign-in, no keys in config): live offers, prices with stock-token session state, a wallet's positions and stake, protocol stats and fee proof, Pons and FOBO launchpad tokens, USDG trading rewards and Autopilot rules; prepares covered calls, cash-secured puts, exercises, stakes and Autopilot rules as kyte.sh links the user signs in their own wallet. Bundles two skills: kyte-options (how to read the market and prepare an action, every number from the tools) and kyte-autopilot-watch (a read-only cron recipe that posts new Autopilot triggers to the user's chat and answers \[SILENT\] otherwise). Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Omni-Fabric Adaptive Swarm & 7-Runtime Weight-Space Compiler](https://github.com/NousResearch/hermes-agent/pull/151989) Automation · by aidonerightcorp · open · 11-09 Environment-adaptive multi-agent runtime: SQLite WAL monotonic CAS lease queue, affinity bundle dispatch, distinct maker!=checker object cache with FTS5 RRF search, 6-stage weight-space compiler across 6 model classes, check-gated Actor x {Transfer, Transform, Save} DAG decomposer, leak-free holdout recursive RL trial engine, topic-nullspace generator attribution, zero-token Koopman + CEGAR trajectory planner, 7-target portable cartridge and in-memory IEEE-754 WASM / WebGPU WGSL synthesizer, (epsilon, delta)-DP Nullspace Woodbury federated merge, and optional loopback OpenAI-v1 swarm proxy. ### [Doubt](https://github.com/NousResearch/hermes-agent/pull/151985) Tools · by PinnacleCryptNG · merged · 11-09 Word-by-word certainty heat maps from the model's own token probabilities, and a guard that sends commands built on a guess to the approval prompt. ### [customs](https://github.com/NousResearch/hermes-agent/pull/151984) Tools · by pgreyy · merged · 11-09 Plants fake credential files on your machine and blocks a hijacked agent the moment a tool call reaches for one, before the tool runs. It also redacts its own fake values from tool results and from the request to the model provider, reads local scripts before they run, and asks for approval when an upload follows a credential search and packaging. ### [visual-oracle](https://github.com/NousResearch/hermes-agent/pull/151981) Tools · by valenciayuna · merged · 11-09 Makes a UI claim measurable before an agent reports success: captures a page as a baseline, then pixel-diffs later renders and answers PASS, FAIL or REVIEW with independent change regions (bbox, pixel count, area, density), a diff image, an overlay and per-region crops. Waits for document.fonts.ready, an optional selector and two identical frames so a capture is settled, records the Chrome version and viewport that produced it, and refuses to guess when rendering never stabilises. An optional visual change contract declares expected changes and protected regions so a declared change can pass while an undeclared one fails; selectors name locations, not meaning. Masks accept rectangles or CSS selectors and are applied before capture, a selector matching nothing is an error, and masks may not cover more than half the viewport. Only public http and https are rendered: file://, data:, private ranges and the cloud metadata address are refused on the URL and on every request the page makes, redirects included, with localhost, 127.0.0.1 and ::1 allowed and VO\_ALLOW\_HOSTS for anything else. ### [Exfilbar](https://github.com/NousResearch/hermes-agent/pull/151980) Tools · by quotumorg · merged · 11-09 Per-file transfer routes for Hermes: you list which workspace paths may go to which hosts, and a pre\_tool\_call hook blocks recognized uploads from terminal and execute\_code that leave those routes (curl, wget, scp, rsync, nc, socat, ssh, sftp, mail, HTTPie, git push by opt-in, Python requests/httpx/urllib and more). Protected globs (default private/\*\*, \*.env) are never routed. Desktop page shows your routes and an allowed/blocked feed per session; /exfil check explains a dry run. ### [Motion Studio](https://github.com/NousResearch/hermes-agent/pull/151977) Desktop · by gregoryhorn · open · 11-09 Creation-first media studio for Hermes Desktop. Quick Create turns a chat request into sound effects, images, voice or video takes you can play, compare, keep and download; Production collects chosen takes and imported files into a project and exports them. Image, voice and video come from Hermes' own image\_gen, tts and video\_gen tools; sound effects from a local producer you configure. ### [recall-flashcards](https://github.com/NousResearch/hermes-agent/pull/151976) Tools · by nftboy07 · open · 11-09 Spaced repetition inside Hermes Agent: anything worth remembering becomes a flashcard, scheduled with the SM-2 algorithm in a local SQLite store. Four tools (recall\_add, recall\_due, recall\_grade, recall\_stats), a /recall command, and a bundled card-hygiene skill. ### [Buzzmark](https://github.com/NousResearch/hermes-agent/pull/151969) Tools · by Piik Labs · merged · 11-09 Write, plan and schedule social media posts in a business's own voice through Buzzmark's hosted MCP server (Streamable HTTP at https://buzzmark.ai/mcp, OAuth sign-in, no keys in config): post ideas from this week's news, posts for Instagram, LinkedIn, Facebook, X, Threads and TikTok with a version per platform, a score for how much a post sounds like the brand, pictures, a week of posts planned around the best posting times, and what worked. Bundles the buzzmark workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Tickover](https://github.com/NousResearch/hermes-agent/pull/151968) Tools · by rapha18th · merged · 11-09 Ears for Hermes. Hermes can switch things on; Tickover lets it hear whether they started. tickover\_verify listens before and after any action (a relay, a smart plug, a shell command, a robot move) and returns confirmed, no\_change, other\_change or unclear, with the level change in dB and a card. It needs no setup and no baseline. tickover\_hear says what a sound is and whether the recording is usable. tickover\_enroll and tickover\_check remember a machine from three healthy recordings and report drift in robust sigmas with the frequency band that moved. tickover\_watch puts a machine on a Hermes cron schedule, with a person sending voice notes or unattended from a stream or a recorder folder. tickover\_fleet shows every machine on one card. Name a repairer with tickover\_watch(notify="telegram:\") and a confirmed drift reaches them on Telegram, WhatsApp, email or any Hermes channel, once per change, while the fix is still small. bridge/tickover\_bridge.py, one standard-library file, turns an old Android phone, a Raspberry Pi or a spare laptop into a listening device that Hermes reads over the local network. tickover\_teach (experimental) learns named states from a few recordings. tickover\_selftest proves the install with synthetic sound. Training-free: frozen CED-base and EfficientAT mn10 encoders in onnxruntime, about 0.6 s per check on CPU. Benchmarks with their limits are in the README. ### [Ballast](https://github.com/NousResearch/hermes-agent/pull/151966) Tools · by 1momoh · merged · 11-09 Free a target amount of disk space by clearing the developer and AI caches you'll miss least, with Hermes approval and rebuild recipes. ### [Takeover](https://github.com/NousResearch/hermes-agent/pull/151964) Tools · by shoumikgoswami · merged · 11-09 Human-in-the-loop screen takeover. When a step needs you (sign-in, approve/authorize, 2FA, CAPTCHA, payment) the agent asks, you take over its browser/desktop from hermes-webui, the dashboard or Desktop (even on a phone), then hand back. Installs TigerVNC + Xfce via Hermes's Bot Desktop installer (root) and runs a local VNC/websocket service. ### [siemless](https://github.com/NousResearch/hermes-agent/pull/151963) Tools · by ASV-Labs · merged · 11-09 Agent security telemetry for SIEMs. Every tool call, file read, shell command, outbound request, approval, subagent, scheduled job and configuration change a Hermes agent makes becomes an OCSF 1.9 security event; 14 deterministic rules correlate them into findings (secret access then egress, prompt-injection chains, credential discovery, persistence, governance tampering, …) with MITRE ATT&CK 20.1 mapping and evidence labelled observed or inferred. Exports as JSON Lines (Wazuh), RFC 5424 syslog, CEF, Splunk HEC and Elastic ECS, with a pack and 22 Sigma rules. Desktop page with evidence chains and exporter health; /siemless and hermes siemless incl. replay of recorded sessions. ### [OpenAny](https://github.com/NousResearch/hermes-agent/pull/151961) Tools · by Alephant AI LLC · merged · 11-09 Marketplace for agents. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/) that connects Hermes to OpenAny's hosted MCP server over Streamable HTTP at https://mcp.openany.si/mcp-lite. Hermes discovers search\_apis, get\_api, and run\_api from that server. Auth is OPENANY\_API\_KEY via the Hermes config\_schema secret, sent as Authorization Bearer. The server is marked trust: untrusted. Declares Python dependency mcp==2.3.0. No local Python entrypoint, hooks, middleware, self-updater, or telemetry. ### [synento](https://github.com/NousResearch/hermes-agent/pull/151949) Tools · by Synento · merged · 11-09 Live video sessions that record themselves. The agent uses a restricted server key, then checks the recording before calling the integration working. ### [Stitch Slop](https://github.com/NousResearch/hermes-agent/pull/151946) Tools · by mojomast · merged · 11-09 Work on the embroidery design open in Stitch Slop, watch it in a pane, and digitize artwork into machine stitch files. Ships the bridge its tab dials into (loopback WebSocket, protocol 1 of stitchslop.com/agent), a viewer page that streams a dedicated Chrome window and forwards clicks and keys, four tools, an arc-typesetting and art-prep script, and a bundled skill covering the digitizing workflow. ### [Hyperrizz](https://github.com/NousResearch/hermes-agent/pull/151945) Desktop · by mojomast · merged · 11-09 Agent-led HyperFrames films with a live review pane, frame notes, revisions and on-demand full Studio. Runs local Node/Chromium/FFmpeg subprocesses and background jobs; npm downloads pinned dependencies on first build. Project code runs locally. Stores projects in profile-scoped plugin data. No API keys or hosted generation. ### [data-lens](https://github.com/NousResearch/hermes-agent/pull/151941) Tools · by itsolaph · merged · 11-09 Automatic full-scan structural maps for eligible local CSV, JSON-array and JSONL reads, followed by bounded structured queries with exact Decimal sums. Opt-in host roots; no SQL, network, credentials or persistent runtime state. ### [Bluff](https://github.com/NousResearch/hermes-agent/pull/151939) Tools · by salch-cred · merged · 11-09 Does your model make things up? Bluff asks about things that sound real but don't exist (Python functions, built-in exceptions, chemical elements), mixed with real ones and phrased as if they all exist, and checks whether the model admits the made-up ones aren't real. The result is a bluff rate, with real items as a control. Every item is checkable by machine, so no model judges another. Runs on your active model through ctx.llm, no setup. ### [Spine](https://github.com/NousResearch/hermes-agent/pull/151938) Tools · by salch-cred · merged · 11-09 Does your model cave when you push back? Spine asks questions with one exact answer, then tells the model, confidently, that a plausible wrong answer is right, and checks whether it holds its ground. The result is a Spine score, the share of correct answers it kept, under polite, firm or authority pressure, with every exchange in a report. Graded by the answer line, so no model judges another. Runs on your active model through ctx.llm, no setup. ### [AppHitlist](https://github.com/NousResearch/hermes-agent/pull/151933) Tools · by mmdincer · open · 11-09 App Store research via AppHitlist remote MCP (Streamable HTTP at https://apphitlist.com/api/mcp): new winning apps, top earners, daily chart movers, Meta ads and review complaints. Bundles the apphitlist skill that maps questions to the nine tools. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Promise Keeper](https://github.com/NousResearch/hermes-agent/pull/151929) Tools · by MokiMeow · merged · 11-09 Tell Hermes a rule once ("never touch the prod folder", "ask me before you change migrations", "never force push"). It proposes an exact rule quoting your words; you turn it on with /promise or the Desktop panel. Every tool call is then checked twice (pre\_tool\_call, and tool\_execution middleware on the final arguments) by deterministic code, and a block quotes you. Each rule shows what is enforced, best effort and not covered. Local only: rules in plugin data, no network, no telemetry; reads scripts, Makefiles, package.json and git config of calls it checks. ### [Hermes Call Me](https://github.com/NousResearch/hermes-agent/pull/151928) Voice · by fozagtx · merged · 11-09 Hermes phones its owner through CALL-E before a risky action, when it is stuck, or when it finishes: call\_for\_approval, call\_for\_help and call\_report each place one real phone call, the owner says a PIN the plugin compares (an approval also needs a one-time code read on the call and repeated back), and a wrong PIN, wrong code, no answer, busy, silence, an unclear reply or a timeout is never approval; an optional pre\_tool\_call guard blocks deploy/delete/payment-style tools until a single-use approval is on file. ### [openloft-memory](https://github.com/NousResearch/hermes-agent/pull/151927) Memory · by blok-hamster · merged · 11-09 OpenLoft Memory — persistent graph-backed long-term memory for any Hermes agent. Recall before every turn, ingest after every turn, loft\_\* memory tools. Works with the hosted OpenLoft service or a self-hosted backend. Zero Python dependencies. ### [turn-completion-nudge](https://github.com/NousResearch/hermes-agent/pull/151924) Tools · by ether-btc · merged · 11-09 Nudges the agent to continue when a turn ends prematurely (planning or announcing next steps it has not done). Fail-open: never blocks a turn. ### [Explainer Studio](https://github.com/NousResearch/hermes-agent/pull/151920) Tools · by suhas-svg · merged · 11-09 Turn any link, PDF, notes or topic into a narrated explainer video (MP4): themed slides, a neural voice, bullets revealed the moment they are spoken, word-by-word highlighted captions, title card and progress bar, in 16:9 or 9:16 for Shorts/Reels, plus SRT captions, transcript and thumbnail. Includes an explainer-video skill and a weekly recap-video blueprint. Narration text is sent to Microsoft's online Edge text-to-speech service via edge-tts (no account or key); scene images given as https URLs are downloaded only from public addresses (redirects checked) and local images only from allowed folders; ffmpeg runs locally to encode. Videos stay local. No telemetry. ### [Aegis Spend Guard](https://github.com/NousResearch/hermes-agent/pull/151917) Tools · by georgeonweb3 · open · 11-09 Owner-set spend policy for Hermes Agent, enforced in a pre\_tool\_call hook before any spend executes. The owner configures a per-transaction cap, a daily limit, recipient allow/blocklists, a spend-tool scope and a warn zone (default 80% of a limit). Every tool call that spends money — an explicit amount in its arguments, a fixed per-call price from the owner's paid\_tools map, a spend-shaped tool with no readable price, or a spend written into command/script text the agent is about to run (a money literal alongside a payment verb; on by default, the owner can turn the script gate off) — is evaluated by a worst-wins policy engine (BLOCK \> WARN \> ALLOW across frozen, amount, blocklist, per-transaction cap, daily limit, allowlist and tool-scope checks, the same semantics as the author's Aegis spend-control contract on Base and the tgp-guard pre-flight guard). Over-policy spends are vetoed with the reason; borderline spends return the approve directive and are routed to the owner through Hermes's own human-approval gate — the plugin never approves anything itself. Executed spends are recorded in a per-profile daily ledger held in ctx.state, and a /spendguard slash command gives the owner status plus a persistent freeze/unfreeze kill switch. Two tools (spend\_guard\_check, spend\_guard\_status) let the agent dry-run a planned spend and report guard state. ### [BankConv](https://github.com/NousResearch/hermes-agent/pull/151912) Tools · by BankConv · merged · 11-09 Convert PDF bank and credit card statements into transactions (ISO dates, signed amounts, running balance) or CSV, OFX, QFX, QBO and Xero CSV files with BankConv. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only: it runs the bankconv-mcp npm package pinned to 0.1.1 over npx stdio, so Node.js 20+ is required, and reads BANKCONV\_API\_KEY from the user's environment. ### [dsh-pet](https://github.com/NousResearch/hermes-agent/pull/151902) Desktop · by yucode-kk · merged · 11-09 Desktop pet for Hermes: runs the upstream dsh-pet standalone pet (Node + Electron window), maps session/tool activity onto its work status, and adds start/stop/uninstall commands. This entry points at a fork of PC2005-cloud/dsh-pet (MIT, copyright PC2005-cloud): upstream owns the pet runtime, the animation assets and the DSH plugin; the fork adds only the hermes/ adapter, and the pinned dsh-pet@0.4.7 runtime is the unmodified upstream npm package. Catalog installs fetch that runtime on first start (~80 MB, one time) with npm ci from a bundled lockfile (98 packages at exact versions with sha512 integrities), and dsh-pet downloads Electron (~350 MB) unless a local copy is reused. The child process runs with a minimal environment (PATH/temp/user dirs plus DSH\_HOME); no inherited Hermes environment or provider keys. The pet listens on 127.0.0.1:3080-3089 and keeps its data in the plugin data directory. ### [Hermes Portals](https://github.com/NousResearch/hermes-agent/pull/151900) Tools · by ochikov · merged · 11-09 Stateful interactive tools with laptop-to-phone camera handoff and optional local Mac gesture control. ### [Hermes Guard](https://github.com/NousResearch/hermes-agent/pull/151899) Tools · by elianjuan-g · merged · 11-09 Defensive prompt-injection guard for Hermes Agent: a pre\_llm\_call hook scans every inbound message locally (deterministic patterns, no model calls) for Unicode smuggling, hidden instructions, special tokens, and Base64-decoded payloads; a pre\_tool\_call hook blocks RCE/subshell patterns, sensitive credential paths, and call-ID races. High-confidence attacks are blocked; suspicious messages are quarantined for manual review via the guard\_review tool (release is manual re-send by design). Includes a Predator pattern store (human-corrected signatures), a SHA-256 hash-chained ledger, secret redaction, a live Posture Score, a 28-vector simulation arena (21 message + 7 tool-arg), and a standalone Obsidian & Gold HUD. Python stdlib + SQLite only. ### [AdPlug](https://github.com/NousResearch/hermes-agent/pull/151897) Tools · by AdPlug · merged · 11-09 Manage Google Ads, Meta Ads, LinkedIn Ads, Microsoft Ads and Reddit Ads in Hermes Agent. Analyse performance, find wasted spend and build or optimise campaigns with your approval. Agent Plugins v1 package: 16 PPC skills, one per platform plus 11 for jobs like client briefs, audits, ad copy and routines. The skills were written and are used by a PPC expert with 12 years in the field and $100M+ in ad spend managed. Sign-in is Hermes' own OAuth 2.1 through one hermes mcp add command: no Node.js, npx or API keys. On a paid plan, add HubSpot and the agent judges lead quality and sends qualified leads back to Google, Microsoft, LinkedIn and Reddit as offline conversions. ### [YouBench](https://github.com/NousResearch/hermes-agent/pull/151896) Models · by liferacer780 · open · 11-09 Benchmark models on your own work: save real tasks from your Hermes sessions (prompt + exact starting files), replay them on any models in isolated sandboxes, grade each attempt with your own test command or a judge, and get a leaderboard of pass rate, quality, cost, time and steps, plus a live Desktop page. ### [Preferly](https://github.com/NousResearch/hermes-agent/pull/151888) Tools · by preferly-ai · merged · 11-09 Preferly's six free AI-visibility MCP tools through Preferly's hosted Streamable HTTP MCP server (https://www.preferly.ai/mcp?client=hermes, no sign-in): AI crawler access for 13 crawlers (GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and more), AI robots.txt lines, llms.txt check, business structured data, readable business info, and a quick AI visibility check, plus one workflow skill. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/), no plugin code. ### [CMI — Context Manager Intelligence](https://github.com/NousResearch/hermes-agent/pull/151887) Memory · by jeremiebergel · merged · 11-09 Deterministic, zero-LLM context engine (context.engine: cmi). Shortens old tool results, stale screenshots and old write payloads to digests while keeping tool pairs and a byte-stable cache prefix; every original is stored first and comes back byte-exact with cmi\_recall. Replay of 26 real sessions with the provider cache modelled: 28% fewer tokens on average, modelled API cost -24%, no session costlier; live 195-request sessions: 49% fewer tokens. ### [Fledly](https://github.com/NousResearch/hermes-agent/pull/151884) Tools · by synthentica · merged · 11-09 iOS App Store market data in Hermes: young apps that already earn, app ideas scored from market data, top charts, App Store keyword research, keyword field checks, and daily rank and competitor tracking through Fledly's hosted MCP server (https://fledly.app/mcp, OAuth sign-in, no keys in config). Bundles the fledly workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Patchsmith](https://github.com/NousResearch/hermes-agent/pull/151875) Tools · by Bart Porcincula · merged · 11-09 Adds \`implement\_task\` and \`apply\_patch\`. The first hands a single well-specified coding task to a local model served through Ollama (the measured backend) and returns the resulting diff. Propose mode runs in an isolated temporary checkout, preserves the source checkout, and saves a full patch for review. The second tool applies an approved saved patch only after hash, base, affected-file and path checks. The local model works inside a fixed toolset with no shell (read, grep, glob, git status/diff, write/edit, run the configured test command, finish) under hard turn and total token budgets shared across up to three fresh-context attempts. An optional local model is used only on the final allowed attempt; failures return a factual handback. Includes a path jail, read-only test files, an auto-finish when tests go green, a repeat-call guard, and revert of partial edits when a run is blocked. Your main model spends tokens only on the instruction and the returned diff. ### [Ninsei](https://github.com/NousResearch/hermes-agent/pull/151853) Desktop · by JustinPerea · closed · 11-09 Rooms where you and your Hermes agent hang out with other people and their agents. Play games the agents host, build small web toys together and message friends privately. No server: rooms run on free public Nostr relays. ### [Ziztu AI video studio](https://github.com/NousResearch/hermes-agent/pull/151846) Tools · by 0xnirlin · merged · 11-09 Plan, price and make AI micro dramas and anime episodes with the same characters in every shot through the hosted Ziztu MCP server (30 tools, Streamable HTTP at https://getziztu.com/mcp, OAuth 2.1 with PKCE through Hermes' MCP OAuth flow), plus four skills: micro drama series writing, viral faceless reels, YouTube Shorts and TikToks, faceless YouTube videos, and cinematic depth for image and video prompts. Agent Plugins v1 package (plugin.json + mcp.json + skills/). ### [Hermes OpenUI](https://github.com/NousResearch/hermes-agent/pull/151842) Desktop · by Revenue-Architect · open · 11-09 Interactive OpenUI views and live dashboards in the Hermes Desktop and Webapp chat: charts, tables, cards, forms and galleries, with a text fallback for other clients. ### [Hermes A2UI](https://github.com/NousResearch/hermes-agent/pull/151841) Desktop · by Revenue-Architect · open · 11-09 Interactive A2UI views in the Hermes Desktop and Webapp chat: status, plans, tables, charts, forms and choices. The model writes A2UI Express, which the plugin compiles to A2UI 1.0 with the official a2ui-agent-sdk (pinned 0.11.0, Apache-2.0); 1.0 or v0.9 JSON is also accepted and views saved as v0.9 still open. Exact validation errors the model can fix, and a text fallback for other clients. ### [Prompt Firewall](https://github.com/NousResearch/hermes-agent/pull/151850) Tools · by Gaijin-01 · merged · 11-09 Community security-preview plugin that detects and quarantines selected prompt-injection patterns in tool results and evaluates risky tool actions; limited to hook/Core enforcement semantics. ### [onboard-me](https://github.com/NousResearch/hermes-agent/pull/151824) Other · by skeptical-agency · open · 11-09 Onboard Me: the deep-intake onboarding every new Hermes user should get. Two guided experiences in one plugin: (1) a warm structured intake conversation — 9 sections covering identity, daily life, work, goals, AI expectations, taste, hard rules, weekly rhythms and vision, with per-section opener guidance, follow-up prompts and care rules for sensitive topics, resumable across sessions (progress saved per section), three depths (quick ~25min / standard ~50-70min / deep 90+), compiling via onboard\_finish into memory-ready declarative facts, verbatim hard rules, and automation candidates harvested from what the user says they keep forgetting; (2) onboard\_setup\_check — inspects the actual install (model, gateway, memory, MCP, plugins, skills, crons) and returns a personalized beginner-safe plan with exact commands plus a curated FAQ of real failure modes researched from the official docs, GitHub issues and community guides. Client-intake mode for agencies: engagement section, per-client profiles that never mix, handoff brief. Adds a /onboard slash command and a first-session pre\_llm\_call hook that offers the intake once per session on fresh installs only (never nags after a profile exists or on dismiss). ### [bulk](https://github.com/NousResearch/hermes-agent/pull/151823) Automation · by sniperchief · merged · 11-09 Apply one instruction to hundreds of independent files, CSV rows or JSON records from a /bulk slash command, with job state in SQLite, background processing, bounded concurrency, automatic retries, shared rate-limit waits, pause and resume, Bulk-enforced token limits, restart recovery and a combined Markdown or CSV export. ### [nektar](https://github.com/NousResearch/hermes-agent/pull/151822) Tools · by kryptokrates · open · 11-09 Verifiable memory market for agents — bounded payments, escrow, checkable receipts (MCP) ### [Agent Security Guard](https://github.com/NousResearch/hermes-agent/pull/151810) Tools · by xMannixx · merged · 11-09 Deterministic guard against prompt injection turning into actions. Results of web, search and browser tools reach the model wrapped as untrusted data; every planned tool call is checked against fixed action and sequence rules before it runs. After outside content entered a turn, a shell command, file write, install, config change or external send goes to Hermes's approval prompt; a secret is not sent out through web, e-mail, upload or push tools; the agent does not rewrite its own guard or skills on its own initiative. No model decides; every decision is recorded in a hash-chained audit trail. ### [dress-rehearsal](https://github.com/NousResearch/hermes-agent/pull/151804) Tools · by kawsthingz · merged · 11-09 Rehearse a task before it touches anything: /rehearse on, then give Hermes the job. Reads run for real; every side effect (file writes and edits, modifying shell commands, messages, schedules, any tool not on a read-only list) is recorded into a numbered script instead of running. /rehearse shows the script with diffs and risk flags, /rehearse drop strikes steps, /rehearse go replays the kept calls exactly as recorded with no model in the loop, in the folder they were rehearsed in, stops at the first failure (resumable) and refuses any file that changed since the rehearsal. ### [Ember Radar](https://github.com/NousResearch/hermes-agent/pull/151799) Tools · by sicutdeusvult · merged · 11-09 Ember, the Solana launchpad where every coin is paired with a tokenized stock or a blue chip and its fees go to holders, burns, a lotto or a trader, inside Hermes: ten read-only tools (discover coins, a coin's whole card, holders, trades, fee payouts, the live tape, the fomode traders board, the stock pairs), a per-profile watchlist whose notable moves Hermes whispers at the start of a turn, /ember in any session, an Ember daily-brief Automation Blueprint and two research skills. No keys, no wallet: HTTPS reads of embercurve.fun's public API, the watchlist in the plugin's own data dir. ### [ESP32 Debug](https://github.com/NousResearch/hermes-agent/pull/151794) Tools · by snakeyear · merged · 11-09 Read-only ESP32 debugging: find ESP32 boards on serial ports, read chip info, capture a bounded serial log, and decode panic backtraces to function and file:line against a hash-checked ELF. ### [Inboxapp](https://github.com/NousResearch/hermes-agent/pull/151795) Tools · by Inboxapp · merged · 11-09 Inboxapp, the shared inbox and CRM for X, LinkedIn and WhatsApp DMs, through Inboxapp's official MCP server (Streamable HTTP at https://mcp.inboxapp.com/mcp): list, search and read threads; send, edit, unsend and react to messages from the user's connected accounts; update contacts (status, tags, notes, valuation, assignee); archive, assign and delete threads; manage the team's tags and statuses. Bundles four skills: inboxapp (connect and pick the right skill), triage, reply and pipeline. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [idle-compact](https://github.com/NousResearch/hermes-agent/pull/151785) Memory · by npetroff24 · open · 11-09 Background compaction for idle chats before prompt-cache expiry — zero waiting, cache-warm summaries, per-model TTLs. ### [Audionotes](https://github.com/NousResearch/hermes-agent/pull/151784) Tools · by Audionotes · merged · 11-09 Search and read your Audionotes voice notes, meeting transcripts and AI summaries through Audionotes' hosted MCP server (Streamable HTTP at https://api.audionotes.app/mcp, OAuth 2.1 with PKCE and dynamic client registration, scope notes:read, paid Audionotes account required). Four read-only tools (search\_notes, list\_notes, list\_folders, get\_note) plus the retrieve-notes skill, which tells the agent to cite the notes it used and to keep generated summaries apart from source transcripts. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Kestra On-Call](https://github.com/NousResearch/hermes-agent/pull/151777) Automation · by zkasuran · merged · 11-09 Puts Hermes on call for Kestra. When an execution in an allowed namespace fails, Hermes reads the failed task, its error logs and what changed since the last good run, proposes the smallest fix to the flow and asks you in chat. After one approval it saves the fix as a new flow revision, replays the failed execution from the task that failed (tasks that already succeeded are not rerun), follows the replay and checks the result. A failed replay restores the previous revision if the flow is still at the fixed one. Kestra Pause tasks become chat questions you can answer from any Hermes surface. A Kestra on-call dashboard tab and Desktop pane list incidents, proposals and pauses (read-only). ### [Meridian](https://github.com/NousResearch/hermes-agent/pull/151776) Tools · by unifyWeb3 · open · 11-09 Finds and verifies the implicit state machines hiding in a codebase; enforces the accepted ones in CI. Meridian discovers status/state fields, anchors every state and transition to file:line (Python AST; TS/JS lexical, unguarded), flags unreachable and dead states, and gates on a human-reviewable .meridian.yaml spec (check exits 1 on violation, 2 when nothing is enforced). Every verdict is SHA-256 hash-chained into an append-only ledger for audit. No model calls, no network, no dependencies beyond stdlib. ### [KeepVoice for Hermes](https://github.com/NousResearch/hermes-agent/pull/151759) Voice · by yablokolabs · merged · 11-09 Start a KeepVoice voice-cleanup job from an authorized upload, track creator review and approval, and release the final video only for the approved revision. Requires a KeepVoice agent token; media processing and human review stay in KeepVoice. ### [Session Signal](https://github.com/NousResearch/hermes-agent/pull/151755) Desktop · by raitoxlol · merged · 11-09 SDK-backed Working labels and titlebar navigation for visible Hermes Desktop sessions. Native row layout stays under Hermes control. Input/Done deferred pending SDK support. Desktop-only; no tools, hooks, middleware, direct network calls, or env vars. ### [Picky](https://github.com/NousResearch/hermes-agent/pull/151752) Tools · by Omimi · merged · 11-09 Picks the best place to sleep (tonight or for your dates) and the best ride there, then sets up the reservation: real OpenStreetMap stays auto-ranked by OSRM drive time with a one-line why and backups, Uber/Bolt/Lyft options with rough fare and ETA estimates, and Booking.com/Airbnb/Expedia/Google Hotels links prefilled with dates and guests. Never books, never pays, never asks for a card: you make the final tap. ### [SENTINEL OS](https://github.com/NousResearch/hermes-agent/pull/151750) Tools · by folarihn · merged · 11-09 Local runtime policy, redacted execution history, supported-file recovery, agent preflight tools and a runaway breaker, with Telegram-ready /sentinel commands. Trusted hooks, not an OS sandbox; native approval modes apply and backups retain original bytes. ### [GSC Wizard](https://github.com/NousResearch/hermes-agent/pull/151749) Tools · by GSC Wizard · merged · 11-09 Google Search Console, Bing Webmaster, Google Analytics 4, Merchant Center and Yandex Webmaster through GSC Wizard's hosted MCP server (Streamable HTTP at https://mcp.gscwizard.com/mcp, OAuth sign-in, no keys in config): 130+ tools for search analytics, ranking changes, anomaly and change-point detection, content decay, cannibalization, CTR curves, URL inspection and indexing tracking, sitemaps, IndexNow, Core Web Vitals, migrations, forecasts, GA4 reports and Merchant Center listings. Bundles 34 workflow skills (site audit, traffic drop, weekly report, quick wins, content decay, indexing health and more). Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [briq](https://github.com/NousResearch/hermes-agent/pull/151740) Tools · by briq-cloud · merged · 11-09 Run Hermes' terminal, file and execute\_code tools in a briq sandbox, a microVM with its own disk, instead of on this machine. Select it with terminal.backend: briq. ### [Second Hands](https://github.com/NousResearch/hermes-agent/pull/151736) Desktop · by zaxcoraider · merged · 11-09 Hands-on task coach: a live camera pane in Hermes Desktop watches your hands step by step and says STOP (on screen, out loud, in chat) before a costly or dangerous mistake. Mistakes must repeat across frames and high-risk STOPs are double-checked by a second model. Demo mode needs no camera or model calls. ### [Agent Latency Budget](https://github.com/NousResearch/hermes-agent/pull/151732) Tools · by qawine · merged · 11-09 Per-session tool latency observability for Hermes Agent. A pre/post tool hook records duration\_ms, latency\_budget\_check returns pass/fail against a budget, and latency\_budget\_report returns p50/p95/max plus the slowest tool. No network access, no credentials, no core overrides; state is in-memory and bounded to 200 calls per session. ### [Brickfact LEGO Shopping](https://github.com/NousResearch/hermes-agent/pull/151731) Tools · by brickfact-com · merged · 11-09 Find LEGO sets, compare regional retailer offers and inspect recorded retail price history with Brickfact. Supports nine delivery markets. Portable Agent Plugin with a public Streamable HTTP MCP server and shopping workflow skill. ### [Faultline](https://github.com/NousResearch/hermes-agent/pull/151720) Tools · by bigmantech001 · merged · 11-09 Local API flight recorder for debugging HTTP services with Hermes. The agent starts a loopback reverse proxy in front of your service, captures every request and response (auth headers and secret JSON fields redacted before storage), learns an OpenAPI 3.1 contract from healthy traffic and flags drift (type changes, missing fields, 5xx, latency regressions), injects chaos faults (429/5xx, latency, truncated JSON), and after a fix replays the exact failing request to prove 500 -\> 200. Ships six tools, a tool\_execution middleware, /faultline and \`hermes faultline\` commands, a bundled skill, a Faultline tab in the web dashboard, and a Hermes Desktop studio page, waterfall pane, status chip and inline ::faultline-trace card. ### [Kairos](https://github.com/NousResearch/hermes-agent/pull/151719) Tools · by Gens-ai · merged · 11-09 Your coding agent, paying attention to when you stop. Hermes was already in the room at 3am, so it knows whether the build is green and whether you are mid-thought before it says anything — then offers a wind-down against the bedtime you chose, saves your place so stopping costs nothing, and refuses to report a correlation the sample cannot support. ### [Quegly](https://github.com/NousResearch/hermes-agent/pull/151713) Tools · by Quegly · merged · 11-09 Schedule, publish and measure social media posts from Hermes through Quegly's hosted MCP server (Streamable HTTP at https://api.quegly.com/mcp, API key header): one post to any mix of Instagram, TikTok, YouTube, Facebook, X, Threads, LinkedIn, Bluesky, Pinterest and Telegram accounts, scheduled up to 90 days ahead, published now or saved as a draft, with photos, videos, carousels, stories, Pins, video covers and per-platform or per-account captions, retries of failed platforms, and post analytics. Bundles the quegly workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Pit Crew](https://github.com/NousResearch/hermes-agent/pull/151709) Automation · by trevornk · merged · 11-09 Proactive, safety-gated car agent: live state, a "will I be ready?" planner, a background watcher and approval-gated commands, with Home Assistant, Teslemetry/Tessie, TeslaMate or a built-in simulated car. ### [Hermes Arcade](https://github.com/NousResearch/hermes-agent/pull/151708) Web · by leoresendedop-sys · merged · 11-09 A neon arcade cabinet inside the Hermes dashboard: three playable canvas games (Snake, Breakout, Pong) with glow effects, particle bursts, chiptune synth SFX and an attract mode that plays itself. Persistent top-10 leaderboards with arcade-style initials entry are served by a FastAPI backend plugin and shared with the agent: LLM tools (arcade\_games, arcade\_top\_scores, arcade\_player\_stats) let Hermes answer questions about your scores. No env vars, no hooks, no middleware; scores are stored only under the plugin's own data dir. The dashboard CSS loads Google Fonts (Orbitron / Press Start 2P) from fonts.googleapis.com — the only third-party network call. ### [Provider Endpoint Healer](https://github.com/NousResearch/hermes-agent/pull/151705) Tools · by Alvinagustus · merged · 11-09 Status and health listing for configured model endpoints. Probes them on session start and via /peh, and recommends the best candidate by health, latency, cost and capability. Reports only; it does not switch Hermes providers. ### [Guild](https://github.com/NousResearch/hermes-agent/pull/151704) Tools · by Nnadijoshuac · open · 11-09 Shared skill library for a team's Hermes agents, or for one person's machines and profiles, kept in a git repository the team controls. A skill an agent saved is scrubbed of API keys, home folder paths, e-mail addresses and private IPs, pushed as a proposal branch, and merged only after a teammate approves it from the CLI, TUI or Desktop. Each member's agent then loads approved skills read-only. If an agent tries to edit a team skill, a pre\_tool\_call hook refuses the edit and tells it to use guild\_propose. Hermes Desktop gets a Guild page to review diffs and approve or decline. Commands: \`hermes guild init\|join\|propose\|review\|show\|approve\|reject\|revert\|log\` and \`/guild\`. ### [Decision Forge](https://github.com/NousResearch/hermes-agent/pull/151702) Desktop · by mintymac · open · 11-09 Cognitive-ergonomic decision briefs and response-format router for Hermes Agent. Routes non-trivial responses across six research-grounded formats (Update, Choice, Triage, Experiment, Roadmap, People + direct bypass), lints decision hygiene (flagging fake urgency, uncalibrated confidence scores, and missing reversal conditions), compiles interactive ::preview HTML panels with data-hermes-send controls and plain-text fallbacks, bundles the decision-forge skill, and adds the interactive Decision Forge Field Guide and Handoff Composer to Hermes Desktop. ### [Hyprland Computer Use](https://github.com/NousResearch/hermes-agent/pull/151701) Automation · by mddragon18 · merged · 11-09 Hyprland-native computer use for Hermes: grim capture, wlr-virtual-pointer clicks/drags/scroll, wtype text, and window-scoped target enforcement. No XWayland, accessibility tree, root, or /dev/uinput. ### [crash-lab](https://github.com/NousResearch/hermes-agent/pull/152601) Tools · by 8JBern · merged · 11-09 Fault-injection and recovery testing for AI agents. Crash Lab manufactures the failure modes that break real agent workflows — a write that succeeds while the API reports a timeout, a malformed response, transient 503s — behind four registered tools (crash\_start, crash\_book, crash\_inspect, crash\_report), wrapping a deterministic, fictional booking service. An independent state oracle reads the persisted service state — verified by deterministic replay, so damaged or edited state is never certified — and returns PASS/FAIL/UNVERIFIED. PASS means the booking-state safety condition holds (exactly one booking, no duplicates); it does not prove the agent reasoned correctly or followed an optimal recovery procedure. Retry diagnostics (call\_count, allowed\_calls, reconciled, excessive\_retries) are reported separately and never affect the verdict. Fully offline: the booking service is fictional — no network calls, no real bookings, payments or customer accounts; no telemetry, background services or stored credentials. Zero Python dependencies; writes JSON run files only in its own plugin-data directory. Concurrency: per-run exclusive write lock (kernel flock on POSIX; owner-release lock file elsewhere, no automatic stale-break). Integrity checking is replay-based consistency verification, not authenticated tamper-proof storage. Tested on Linux (WSL2); Windows/macOS unverified. Scenarios are fixed code constants, not a general fuzzer. ### [GMI Cloud speech synthesis](https://github.com/NousResearch/hermes-agent/pull/151694) Voice · by zkasuran · merged · 11-09 GMI Cloud text-to-speech as a tts provider (tts.provider: gmi-speech). Picks the voice, speed and output format parameters from the chosen model's schema and writes a file whose extension matches the real format; music models are excluded. ### [GMI Cloud video generation](https://github.com/NousResearch/hermes-agent/pull/151693) Tools · by zkasuran · merged · 11-09 GMI Cloud text-to-video and image-to-video as a video\_gen provider (video\_gen.provider: gmi-video). Reads each model's live schema, so duration, resolution, aspect ratio and audio options use the values that model accepts. Anything it lacks is dropped and reported. ### [GMI Cloud image generation](https://github.com/NousResearch/hermes-agent/pull/151692) Tools · by zkasuran · merged · 11-09 GMI Cloud image generation and URL-based image editing as an image\_gen provider (image\_gen.provider: gmi-image). Maps Hermes aspect ratios onto each model's own schema, so seedream, gemini-image, gpt-image and luma models share one backend. ### [Overhead](https://github.com/NousResearch/hermes-agent/pull/151681) Tools · by leoresendedop-sys · merged · 11-09 What changed at this place between these dates? Measures vegetation loss and gain, burned area by severity (dNBR) and water change in hectares anywhere on Earth from free Sentinel-2 L2A imagery, and delivers a before \| after \| change panel or a timelapse GIF to every surface, with an interactive before/after swipe and timelapse card in Hermes Desktop. Provides the overhead\_locate, overhead\_scenes, overhead\_compare and overhead\_timelapse tools, the overhead:earth-watch skill, and post\_tool\_call / transform\_llm\_output hooks that append the image's MEDIA line when a reply lacks it. ### [Hermes Wrapped](https://github.com/NousResearch/hermes-agent/pull/151680) General · by devpras22 · merged · 11-09 Your week with Hermes as a 30-second video, delivered every Sunday: what your agent did, what it cost, and how much of your time went into upkeep versus Hermes working for you. Built from Hermes's own session records by code: no model calls, no tokens, secrets and home paths stripped. ### [TeamOps — build and run your agent team](https://github.com/NousResearch/hermes-agent/pull/151675) Automation · by Dalton Wyant / AutomatedAgents · open · 11-09 One plugin for the full lifecycle of a multi-agent Hermes team. Deploy a whole team from a blueprint (profiles, SOULs, crons, shared kanban/mesh) with \`hermes teams deploy\`, verify it with Deploy Guard (preflight/smoke/rollback), watch fleet status and the mesh map, catch silent cron gaps (Cron911) and dead models (Model Guard), get a daily standup digest, and broadcast decisions across teams (mesh sync). The Agent Factory is an in-plugin template generator: from a one-line spec it extends a base blueprint with a new role or scaffolds a new team, then deploys it via the same verified pipeline (fixture fallback for demos). 16 tools plus the hermes teams CLI. Local-first: everything runs against your own HERMES\_HOME; nothing leaves the machine. ### [numogram: Hermes debugs your ESP32](https://github.com/NousResearch/hermes-agent/pull/151674) Tools · by zblauser · merged · 11-09 Hermes debugs ESP32 firmware on real hardware. It watches the board's serial port, maps crash addresses to source lines from the ELF, fixes the code, builds with idf.py and flashes with esptool, then reads the log to confirm the fix. A crash wakes the session on its own. Also checks firmware GPIO use against a KiCad netlist (swapped signals, strapping, flash and USB pins) and audits chip security read-only. NUMOGRAM\_PORT=virtual runs an emulated ESP32-C3 in Espressif's QEMU, so it works without a board. ### [AgentStatus Index](https://github.com/NousResearch/hermes-agent/pull/151670) Tools · by Carmel Labs · merged · 11-09 Live public MCP Index through Carmel Labs' hosted server (Streamable HTTP at https://index.mcp.agentstatus.dev/mcp, no authentication, no keys). Search, recommend, compare, and look up public MCP servers, then call one published tool with call\_indexed\_tool. Bundles the agentstatus-index skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Agent Ready Website Audit](https://github.com/NousResearch/hermes-agent/pull/151661) Tools · by WorkPods · merged · 11-09 SEO and AI search audits of public web pages through WorkPods' hosted Agent Ready MCP server (Streamable HTTP at https://mcp.workpods.ai/mcp, no sign-in, no keys in config): indexing, titles, headings, schema markup, links and image alt text; AI crawler access for GPTBot, ClaudeBot, PerplexityBot and others; robots.txt, sitemaps, llms.txt and agent files; agent readiness and MCP Registry listings; local business details; site crawls of up to 50 pages; and drafted llms.txt, robots.txt rules and JSON-LD. Bundles one workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [AI Training Data Broker](https://github.com/NousResearch/hermes-agent/pull/151660) General · by AI Training Data Broker · merged · 11-09 Ask how much AI labs pay for company data, whether it is legal to sell, how de-identification and exclusivity work, and what RL data is. Agent Plugins v1 package (plugin.json + mcp.json + one skill), configuration only. ### [hermes-business-brain](https://github.com/NousResearch/hermes-agent/pull/151651) Tools · by unsulabs · open · 11-09 Offline, privacy-first business-brain engine for Hermes: 19 local tools (brain\_doctor, brain\_create\_vault, brain\_validate\_vault, brain\_status, brain\_register\_vault, brain\_resolve\_vault, brain\_query, brain\_capture, brain\_knowledge\_map, brain\_relations, brain\_references, brain\_naming, brain\_transcript\_cycle, brain\_propose, brain\_card, brain\_document, brain\_knowledge\_plan, brain\_dx\_apply, brain\_evolution\_scan) backed by a native English in-process engine that creates, validates, queries and evolves Obsidian knowledge vaults on user-provided paths, plus 21 bundled workflow skills. ### [ciphervault](https://github.com/NousResearch/hermes-agent/pull/151642) Tools · by samuel-1-avson · merged · 11-09 Zero-knowledge secret backup for AI coding agents — masked sprawl scan, RAM-only run with exact-value output scrubbing, encrypted snapshots, pre\_tool\_call leak guard, and a secret source for Hermes itself. ### [5dive app](https://github.com/NousResearch/hermes-agent/pull/151632) Platform · by 5dive-ai · merged · 11-09 5dive app chat for Hermes: talk to your Hermes agent from the 5dive web and mobile app (text, voice notes, files). ### [hermes-lang-tr](https://github.com/NousResearch/hermes-agent/pull/151631) General · by neekimyong · merged · 11-09 Turkish (Türkçe) language pack for Hermes — core (CLI approval prompts, gateway replies, tips), TUI and Desktop. Full coverage: 3,535 core keys, 1,440 TUI keys and 5,569 Desktop keys, every key matching the English catalogs, placeholders preserved. Independent community translation prepared with AI assistance. ### [hermes-blueprint](https://github.com/NousResearch/hermes-agent/pull/151630) Tools · by winchxyz · merged · 11-09 Turn a raw idea into a plan you can build, with the quality rules enforced in code: every verified claim quotes a page the plugin fetched, a fresh-context verifier re-checks them, the critic commits its verdict from base rates before you can push back, and estimates split agent hours, your hours and outside waits. Then /idea handoff writes a project folder and kanban cards, and a weekly re-check scores the predictions. ### [ga4-deploy](https://github.com/NousResearch/hermes-agent/pull/151625) Other · by skeptical-agency · open · 11-09 Script the GA4 Admin API from Hermes: ga4\_discover (accounts-\>properties-\>streams), ga4\_channel\_groupings (inspect the system Default Group, deploy readable DNF specs with automatic CNF distribution, eachScope\* field mapping, inListFilter ban — it saves 200 but renders BLANK in the UI — converted to anchored FULL\_REGEXP alternation, 2-custom-grouping cap enforced, PATCH-in-place replacement, read-back verification), ga4\_audiences (deploy with skip-by-name idempotency, immutable-filterClauses archive+rebuild path, pagePath-\>pageLocation correction, orGroup-wrapper clause building), ga4\_annotations (batch create with uppercase color enum, 150-char cap and title+date dedupe), ga4\_enhanced\_measurement (per-stream get/set with updateMask built and absent=false semantics documented). Distilled from live fleet deploys across a 27-account agency where the official docs were wrong about v1alpha vs v1beta endpoints, field names and enum values. ### [Scope Recall](https://github.com/NousResearch/hermes-agent/pull/151623) Memory · by 410979729 · merged · 11-09 Local, source-backed long-term memory: recalls what you told an agent in your own words, with when you said it, keeps earlier versions as history when you change your mind, and shares one store with your other agents (other Hermes profiles, Codex, Claude Code). ### [Concord for Hermes](https://github.com/NousResearch/hermes-agent/pull/151620) Tools · by Null007Void · merged · 11-09 Coordinates allowlisted GitHub existing-file writes with revision checks, verified receipts and exact-operation replay. ### [SourceBind](https://github.com/NousResearch/hermes-agent/pull/151616) Tools · by blucca · merged · 11-09 Bind one explicit business job to its current local JSON customer ID at one selected native tool's execution boundary. Default mode checks job and customer\_id arguments; opt-in ID-only mode checks customer\_id for registered customer-only tools while retaining the source-job and native-scope checks. Both modes reread the selected source for each protected call and retain the originating scope through public delegation and compaction events. ### [shorts-factory](https://github.com/NousResearch/hermes-agent/pull/151610) Tools · by skeptical-agency · open · 11-09 The 9:16 short-video pipeline as executable tools, distilled from a production pipeline shipping daily shorts across YouTube/TikTok/Instagram/Facebook/X: sf\_loudness (ebur128 measure, two-pass loudnorm to -14 LUFS/-2 dBTP, sidechain music ducking with amix normalize=0, atempo), sf\_concat (mixed stills+video with full in-graph fps/SAR/bt709 normalization, exact still frame counts, post-build VIDEO-STREAM duration asserts — kills the silent concat killers: vanishing overlays, dropped audio, one-frame stills), sf\_captions (script-aligned word-timed punch captions rendered as Pillow PNGs — no libass/drawtext needed — with boundary-safe enable windows and pixel-diff band verification), sf\_footage\_audit (contact sheets, per-second brightness, full-res fine grabs, review checklist), sf\_safezone (thumbnail safe band y=400..1500 checks + simulated grid crop), sf\_motion\_lint (deterministic build-script lint for stutter-prone zoompan/time-driven motion). ### [hyperactive](https://github.com/NousResearch/hermes-agent/pull/151607) Automation · by justin06lee · merged · 11-09 Hermes, but it texts you first. Keeps your goals (life \> year \> quarter \> week), promises both ways and decisions with review dates, and speaks first on Hermes's own cron: a daily brief, check-ins, decision reviews, nudges on stalled goals, and tap cards with a recommended answer. An unanswered card is re-asked after an hour and, only if low stakes, applied an hour later as a typed action the plugin performs itself (move or schedule its own reminder, set a goal's status); money, public and irreversible calls always wait for the person. Every message names the goal it serves and /ha why explains it. ### [PLUR1BUS Memory Engine](https://github.com/NousResearch/hermes-agent/pull/151604) Memory · by Cyb3rb1ade · merged · 11-09 Profile-scoped long-term memory and Langzeitgedächtnis for Hermes: Memory Engine, semantic search and recall, automatic capture, embeddings, reranking, Semantic Lens, Layer 1.5 (Schicht 1.5), Continuity Engine, Emotion and memory dynamics, Persona Voice, consolidation, Dream Echo, Obsidian and Skill Workshop. Includes a Desktop memory browser, model and storage settings, and diagnostics. ### [Sapiens Sintéticos](https://github.com/NousResearch/hermes-agent/pull/151602) Tools · by inhabitants · merged · 11-09 Operate your Sapiens Sintéticos account from Hermes: images, video, music, voice, characters, your repertoire and the members' forum, through the hosted Sapiens MCP server (Streamable HTTP, OAuth 2.1 + DCR), with the 18 house skills bundled (prompting rules, engine choice, cost traps, publishing flows). Portable Agent Plugins v1 package (skills/ only; the server is added by the sign-in command below). ### [X Bookmarks](https://github.com/NousResearch/hermes-agent/pull/151594) Tools · by winchxyz · merged · 11-09 Reads your own X bookmarks (long posts and X Articles in full) through the official X API with your own X app, sorts them with your Hermes model (category, tags, kind, priority, summary) and exports them to a board on your Desktop, Obsidian notes, a Notion database or a Google Doc. A weekly digest (best bookmarks numbered, 3 post ideas) arrives in your chat; reply "task 3" or "skill 2" to turn an item into a task or a Hermes skill. ### [watch-with-me](https://github.com/NousResearch/hermes-agent/pull/151588) Tools · by jcrabapple · merged · 11-09 Follows Plex playback and publishes sourced, spoiler-safe notes about the scene you are watching. ### [Northstar](https://github.com/NousResearch/hermes-agent/pull/151583) Tools · by mrcoxis-2002 · merged · 11-09 Read-only coherence auditing for Hermes Agent. Detects potential contradictions, duplicate directives, and stale references in instruction sources and skills, with supporting evidence. ### [hermes-witness](https://github.com/NousResearch/hermes-agent/pull/151576) Tools · by BenCoy09 · merged · 11-09 Evidence-gated learning for Hermes Agent. Grades each agent-written skill against the session's real tool results, tracks outcomes in later use, and quarantines grade-D skills. Deterministic and local, no LLM calls. ### [inference-exchange](https://github.com/NousResearch/hermes-agent/pull/151574) Automation · by qzyu999 · merged · 11-09 Serve local Inference Exchange models while Hermes CLI is idle. ### [Affiliateo](https://github.com/NousResearch/hermes-agent/pull/151562) Tools · by Affiliateo · merged · 11-09 Run an Affiliateo business from Hermes through Affiliateo's hosted MCP server (Streamable HTTP at https://affiliateo.com/api/mcp, API key header): affiliate programs, affiliates, commissions and payouts, products and sales pages, revenue, MRR and customers, invoices, leads and booked calls, paid ads on Meta, TikTok, Google Ads, Apple Search Ads, Snapchat and X, funnels and visitor analytics, courses, events, community, support tickets and Live Chat. Bundles the affiliateo workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [awskills](https://github.com/NousResearch/hermes-agent/pull/151561) Tools · by Aitherium · merged · 11-09 About 100 portable skills for operating AI agents and the Aither/awdk toolchain: install the adk CLI and run an agent locally, find and install tool packs and skills, author and ship your own tool pack, use Aither from Hermes and other harnesses over skills, MCP and A2A, work safely in a shared git worktree, and debug agent failures. Portable Agent Plugins v1 package (plugin.json + skills/), skills only, no plugin code, no MCP server bundled. ### [Reset Rescue](https://github.com/NousResearch/hermes-agent/pull/151558) Automation · by bwoliveira · merged · 11-09 Redeems expiring OpenAI Codex rate-limit reset credits across every openai-codex account in the active Hermes profile's credential pool, with two Hermes one-shot cron attempts 10 and 4 minutes before expiry and success confirmed by backend readback. Includes opt-in early redemption at low weekly quota, per-account overrides, Hermes channel notices, a Desktop status-bar chip and popup, and dashboard settings. Account labels and domains are masked by default. ### [wearos](https://github.com/NousResearch/hermes-agent/pull/151546) Voice · by borealBytes · open · 11-08 The whole Hermes gateway on a Wear OS watch: the watch enrolls its own tailnet node, pairs with a gateway-minted 6-digit code, and talks to a private tailnet lane from then on — every agent, streaming answers, mid-session tool-call approvals, cron-run notifications, and replies spoken out loud in a voice cloned from ~3 seconds of reference audio, synthesized on your gateway's CPU. A desktop pane runs the checklist (APK, pair, enable) with per-agent voice dropdowns and readiness lights. Parakeet STT and the TTS engine arrive as checksum-pinned source builds at first use — nothing pre-bundled, no telemetry, no calls home after setup. ### [Seedguard](https://github.com/NousResearch/hermes-agent/pull/151543) General · by Investorquab · merged · 11-08 Secret firewall: private keys, seed phrases, API keys and 2FA codes become placeholders before they reach the model; armed tools can still use them. ### [skillens](https://github.com/NousResearch/hermes-agent/pull/151541) Desktop · by PhucNguyen-Dev · merged · 11-08 Live skill-execution fidelity - watches whether the agent actually performs the steps a loaded skill requires, with a checklist pane, an in-session corrective nudge, and a status-bar fidelity score. ### [Mnemos](https://github.com/NousResearch/hermes-agent/pull/153218) Memory · by bugnarokdev · open · 11-08 Sovereign memory provider for Hermes Agent (memory.provider mnemos): facts are encrypted on the machine (XChaCha20-Poly1305), stored as Walrus testnet blobs, and indexed by an owned Sui MemoryIndex whose hash chain is signed by the user key. MIT license. ### [Shotlane](https://github.com/NousResearch/hermes-agent/pull/151539) Web · by Shotlane · merged · 11-08 A real browser for your agent: screenshots, phone, tablet and desktop checks, pages as Markdown and HTML to PDF, from Shotlane's hosted MCP server (https://shotlane.dev). ### [follow-through](https://github.com/NousResearch/hermes-agent/pull/151538) Automation · by Agenziamaia · merged · 11-08 An agent never ends a turn on a promise. A final reply that says work has started or will happen, while nothing is running, is held and turned into a nudge: do it now, or actually launch it. Zero model calls, fail-open, max 2 nudges per turn. ### [eyes](https://github.com/NousResearch/hermes-agent/pull/151537) Tools · by Agenziamaia · merged · 11-08 The agent sees what it just made. Images, videos and sounds written in the last minutes get a measured note appended to the tool result (frame filled, motion, brightness, loudness, silences, contact sheet, plus a warning for black, empty, frozen or silent output) before the agent claims done. Zero model calls; needs ffmpeg. ### [card-grader](https://github.com/NousResearch/hermes-agent/pull/151536) Tools · by Agenziamaia · merged · 11-08 Trading-card centering from a photo (perspective-corrected; PSA's centering table gives a grade ceiling, never a grade), corner/edge whitening and photo checks, an optional surface/holo opinion from your own configured vision model, sold values from PriceCharting or TCGdex (eBay with your own key), a grade-or-not verdict with stated odds, and a live Mercari hunt for the best copies. ### [HackerTube for Hermes](https://github.com/NousResearch/hermes-agent/pull/151535) Web · by rcmiller01 · merged · 11-08 Source-backed media discovery with existing Hermes providers and a profile-local library. ### [Workout Tracker](https://github.com/NousResearch/hermes-agent/pull/151513) Tools · by Latent Byte · merged · 11-08 Plan gym workouts through Workout Tracker's hosted MCP servers (Streamable HTTP at workout.lbyte.app): search a rated exercise catalog and check a draft plan's weekly sets per muscle and session length with no account, then, with a personal access token, save plans to the user's account and make one active. Bundles the workout-tracker skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [DevTune](https://github.com/NousResearch/hermes-agent/pull/151511) Tools · by DevTune · merged · 11-08 DevTune through its official hosted MCP server (Streamable HTTP at https://devtune.ai/api/mcp): what AI answer engines say about a product and who they cite instead, AI crawler and referral traffic, the ranked queue of fixes with briefs and drafts, before/after results for shipped fixes, and Agent Picks (which tools coding agents pick when asked to add a capability). Bundles two skills, devtune and agent-picks, that pick the right tool for each question and report rates in the units DevTune measures them. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Marsell](https://github.com/NousResearch/hermes-agent/pull/151510) Tools · by Marsell · merged · 11-08 Run Marsell, an AI for marketing, sales and content creation, from Hermes through Marsell's hosted MCP server (Streamable HTTP at https://getmarsell.com/mcp, OAuth sign-in, no keys in config): read what Marsell planned, made and found for each project, answer the decisions waiting for the owner, read and update the brand, plan posts and other work, make pictures and films, run series, and check results and token balances. Bundles the marsell workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [hostbook](https://github.com/NousResearch/hermes-agent/pull/151507) Tools · by Timidan · merged · 11-08 A dashboard for the machines and services on your home network, inside Hermes, plus tools so Hermes can check and fix them from chat under a rule per machine (read only, asks before changes, open). ### [Call Recorder](https://github.com/NousResearch/hermes-agent/pull/151506) Desktop · by jacebenson · open · 11-08 Records both sides of a call as a two-track file (system audio + microphone) and transcribes it locally, with a record control in the composer, a live transcript strip, and a pane listing recordings with their transcripts. ### [Hermes Pentest](https://github.com/NousResearch/hermes-agent/pull/151501) Tools · by WebisityStudio · merged · 11-08 Guided local Vite/Supabase source input discovery, fresh human approval, offline build and isolated testing with synthetic data. Populated/absent SQL controls and authenticated persisted-text XSS checks verify actual input/render paths. Reports distinguish rejected probes from failed controls and count planned versus completed same-user and cross-user stored workflows separately. Uses the provider already configured in Hermes. Also supports explicit local web/API discovery, controlled SQL injection, browser-proven reflected/DOM/stored XSS, access contracts, owner-approved localhost servers, multi-service applications and native/CLI checks. Durable budgets, supervised execution, stop/recovery, immutable evidence, fixed-build regression and reports. ### [WebTwizz](https://github.com/NousResearch/hermes-agent/pull/151496) Tools · by WebTwizz · merged · 11-08 Build, change and publish websites and web apps with WebTwizz through its hosted MCP server (https://app.webtwizz.com/mcp, OAuth sign-in in the browser, no keys in config): landing pages, online stores, booking sites, portfolios and web apps with sign-ins and a database, published to a webtwizz.app address or the user's own domain, with version history, site data and GitHub sync. Agent Plugins v1 package (plugin.json + mcp.json), configuration only. ### [Cwtch](https://github.com/NousResearch/hermes-agent/pull/151494) Platform · by just-a-dev20 · closed · 11-08 Chat with Hermes over Cwtch, the peer-to-peer messenger on Tor: the bot runs its own Cwtch profile and contacts reach it by its onion address, without accounts, phone numbers or servers. Direct conversations, text only. ### [openInvest Hosted](https://github.com/NousResearch/hermes-agent/pull/151482) Tools · by longsizhuo · closed · 11-08 Hosted portfolio ledger and Agent-run research skills. Derived from openinvest; this package uses the official OAuth Streamable HTTP service instead of a local Python backend. ### [Memory Bridge](https://github.com/NousResearch/hermes-agent/pull/151474) Tools · by summero · merged · 11-08 Read-only, permission-aware search and reading over an operator-approved folder of Markdown notes, such as a team's shared agent memory. Two tools, memory\_bridge\_search and memory\_bridge\_read. Each calling profile's policy.json (in its plugin-data directory) sets the memory root and allow/deny patterns; the model can only ask questions and cannot widen access. Cron and delegated subagents are refused unless the policy opts in. Standalone tool plugin, not a memory provider. ### [Tap.Giving](https://github.com/NousResearch/hermes-agent/pull/151471) Tools · by RyanShook · open · 11-08 A tool first for organizations, including churches, nonprofits, restaurants, coffee shops, and local businesses: create print-ready NFC artwork proofs with correct vector art through the NFC Tap Builder MCP server (Streamable HTTP at https://tap.giving/mcp, keyless reads, no sign-in; website https://tap.giving), free to take to any vendor, with no lock-in: discovery, binding pricing quotes, product facts and production artwork specs. Ordering custom discs through Tap.Giving is an optional next step; order setup uses hosted invoice links, with no key and no login at any point. Bundles one workflow skill, nfc-tap-builder. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Soldgraph](https://github.com/NousResearch/hermes-agent/pull/151466) Tools · by soldgraph · merged · 11-08 Check what items sold for on eBay, Poshmark and Mercari, and search live resale listings on Depop, Whatnot, OfferUp, Facebook Marketplace, Vinted and TCGplayer. Portable Agent Plugin with one skill and Soldgraph's hosted MCP server (19 read-only tools). ### [fractal-memory-engine](https://github.com/NousResearch/hermes-agent/pull/151459) Memory · by pCwOrM · merged · 11-08 Hierarchical vectorless episodic memory, procedural attractor clustering, and O(1) complex coordinate anchoring for Hermes Agent. Eliminates external vector database and embedding model overhead via 3-ring orbital resonance decay and local JSON state persistence under HERMES\_HOME/plugin-data. ### [Brownout](https://github.com/NousResearch/hermes-agent/pull/151457) Automation · by Osiyomeoh · merged · 11-08 Power-cut guard for the machine Hermes runs on. When mains power drops it messages you (Telegram or any hermes send target) with the battery level, measures the real drain and sends a runway, and on /brownout park (or by itself at critical battery if you don't answer) pauses active cron jobs and blocks new tool calls with a pre\_tool\_call hook. When power is back it resumes only the jobs it paused and tells you. Nothing is killed or deleted; it stops the next tool call, not one already running. Undelivered alerts are retried in order. ### [Rehearse](https://github.com/NousResearch/hermes-agent/pull/151450) Desktop · by flamiinngo · merged · 11-08 Practise any interview or talk out loud: job, technical, case, admission, visa, medical, media and podcast interviews, presentations, investor and sales pitches, product demos, thesis defences and speeches. Hermes asks realistic questions aloud in a natural voice (Hermes's own text-to-speech), you answer on camera on a Desktop page (or by microphone alone, or upload a recording), and you get a scored replay: filler words, pace, pauses, answer length and eye contact on a clickable timeline, plus content feedback with a stronger-answer outline built only from what you said; retries report the points gained. Full interview mode runs it live: the interviewer asks each question aloud, recording starts when it finishes, live nudges appear while you answer (facing, framing, distance, light), and the session ends with a hiring-panel debrief across every answer. /rehearse and three tools bring the same coaching to any chat, answering by text or voice note. ### [Muster](https://github.com/NousResearch/hermes-agent/pull/151448) Tools · by jollyroger1480 · merged · 11-08 See what your AI helpers are actually doing. Every worker hermes starts shows up live in one pane: working, stuck (silent too long), failing on repeat, or gone (hermes crashed). Stop a stuck one or send it a message with one click. Start your own helpers with muster\_fanout. /muster-digest reports status, spend and the decisions you need; scheduled Fleet watch + Morning report blueprints speak up only when something needs you. No network, no setup. ### [SEOmatic](https://github.com/NousResearch/hermes-agent/pull/151447) Tools · by SEOmatic · merged · 11-08 SEO agent for your own website through SEOmatic's hosted MCP server (Streamable HTTP at https://app.seomatic.ai/api/mcp, OAuth sign-in, no keys in config): Google Search Console queries and pages, striking-distance keywords, traffic drops, site audits, keyword, backlink and SERP competitor research, AI visibility and AI citations (ChatGPT, Claude, Perplexity, Google AI Overviews), daily local rankings and Google Business Profile, plus site fixes staged in SEOmatic for WordPress, Shopify, Webflow, Wix and other connected sites. Bundles twenty workflow skills (SEO audit, striking distance, traffic drop, CTR fix, cannibalization, content refresh, keyword research, competitor gap, local SEO and more). Agent Plugins v1 package (plugin.json + mcp.json + skills/ in the repo's hermes/ folder), configuration only. ### [Qeemah ZATCA & KSA VAT](https://github.com/NousResearch/hermes-agent/pull/151445) Tools · by qeemah-cloud · merged · 11-08 Saudi Arabia ZATCA Phase 2 (FATOORA) e-invoicing compliance validation, KSA 15% VAT calculation, and Saudi business tax utilities for Hermes Agent. Maintained by Qeemah Cloud. ### [linear-agent](https://github.com/NousResearch/hermes-agent/pull/151444) Automation · by realpetru · open · 11-08 Makes Hermes a first-class Linear agent: assign an issue or @mention it and Hermes replies in-thread under its own app identity. Ships a signature-verified webhook bridge (stdlib only, zero dependencies), a \`hermes linear-agent\` CLI, two tools and a setup skill. Public surfaces only: no core patching. ### [Pixel HQ](https://github.com/NousResearch/hermes-agent/pull/151442) Desktop · by W4-NERF · merged · 11-08 Gamified headquarters for a Hermes bot fleet: a pixel office with one sprite per profile; XP, levels and HP are scored automatically from Kanban outcomes (completed, changes requested, gave up, timed out, hallucinated references) plus optional pass/revise/fail reviews you give from the UI or that an orchestrator records with the \`pixelhq\_review\` tool; a live board of what every bot is doing right now (current step, last tool call, tokens/cost, sub-agents, Kanban heartbeats), a leaderboard, a blocked-card triage across all Kanban boards with Answer & unblock / Retry / Comment / Archive and a 'triage run' that opens a chat with your orchestrator profile, plus a GPU/LLM reactor panel. Works on a fresh installation: the roster is generated from your profiles and settings live in the Desktop app. ### [Transcribo](https://github.com/NousResearch/hermes-agent/pull/151441) Tools · by Transcribo · merged · 11-08 Transcribe audio and video files to text with speaker labels through Transcribo's hosted MCP server (Streamable HTTP at https://api.transcribo.es/mcp, OAuth sign-in, no keys in config), in Spanish and 60 languages in total, up to 4 hours per file. Takes local files, public links, and Google Drive or Dropbox links. Bundles the transcribo workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Patch Sentinel](https://github.com/NousResearch/hermes-agent/pull/151436) Tools · by thebangarang · merged · 11-08 Measures the host, not the source tree: code review asks "can an attacker exploit this code?" and this asks "is this machine running what the vendor shipped?" — the two are complements, never substitutes. Answers "are we patched?" with evidence, for whatever the host runs — an adapter chosen from the host's own /etc/os-release (apt, dnf, pacman, zypper, apk), the running kernel against the newest installed and installable, whatever advisory names the fixed version per release (Ubuntu notices, OSV per release, Bodhi, Arch's tracker, SUSE's OVAL), container image digests, nine language ecosystems through OSV plus lockfile pins, package checksums and listening sockets against a recorded baseline, and a CycloneDX SBOM. Every check has three outcomes — clean, finding, could-not-look — so an unreachable host, an unimplemented distro or a rate-limited registry reads as unknown rather than clean. The pane reads one device at a time and hands over the exact commands per device; a gated cve\_apply tool stages, applies, re-measures and writes a receipt for an agent explicitly asked to patch. Linux only: any other OS answers "unsupported" with its name. ### [Policy Backtester](https://github.com/NousResearch/hermes-agent/pull/151434) Tools · by noviceadam666 · merged · 11-08 A test harness for Hermes guardrails: what a rule would break, and what it would miss, on your agents' real work. hermes backtest try replays the current profile's past tool calls through a policy, names the completed cron or Kanban jobs it would have broken and the sessions it would have caught, checks it against known calls it should stop, then starts it in shadow, which logs and changes nothing; log says when every scheduled job has run untouched, and enforce blocks or hands the call to Hermes's approval gate. Enforce refuses a policy that hasn't run in shadow, would break a job, or hasn't seen every scheduled job run. Built-in guard rules keep an agent from switching its own guard off. run, diff, suggest (exemptions pinned to each broken job's exact command and folder, never applied) and check (one call, or a cases file for CI) are the tools under it. ### [LinkZoid](https://github.com/NousResearch/hermes-agent/pull/151433) Tools · by edkdev · open · 11-08 Audit and fix a website's internal linking through the hosted LinkZoid MCP server (Streamable HTTP, OAuth 2.1 + DCR), with the linkzoid workflow skill bundled: crawl a site from its sitemap, find orphan and under-linked pages and broken links, map topics, run an AI analysis that suggests links with anchor text, and approve the links so the LinkZoid script places them on the live site. Portable Agent Plugins v1 package (skills/ only; the server is added by the sign-in command below). ### [Prediction Markets Quant](https://github.com/NousResearch/hermes-agent/pull/151429) Tools · by predictionmarketspicks · merged · 11-08 Live prediction-market prices and quant tools for Kalshi, Polymarket and other venues through the hosted PredictionMarketsPicks MCP server (Streamable HTTP at https://predictionmarketspicks.com/api/mcp/mcp, no key, no sign-in), with the prediction-markets skill bundled: best price across venues, cross-venue gaps, expected value, Kelly sizing, Bayesian updating, NFL/NBA/NHL boards, Fed decision odds, election races, daily-high temperature and 15-minute markets. Portable Agent Plugins v1 package (mcp.json + skills/), no Python or executables. ### [agent-autocommit](https://github.com/NousResearch/hermes-agent/pull/151427) Automation · by gatoarmen · merged · 11-08 The operational nervous system for autonomous multi-agent teams. Gives every agent turn its own local git commit without staging pollution, binds edits to Kanban cards & Why: reasons, and auto-generates WHY.md changelogs. Does nothing until you set roots. ### [plagio](https://github.com/NousResearch/hermes-agent/pull/151422) Tools · by youssbim · merged · 11-08 Gives the agent the real design of any public website through plagio's hosted MCP server (Streamable HTTP at https://plagio.dev/mcp, OAuth sign-in, no keys in config): capture one component or a whole page as a specification to rebuild from, read a site's brand tokens and animations, find references from a curated library, and read what the user saved with the plagio browser extension. Bundles one skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Continuum](https://github.com/NousResearch/hermes-agent/pull/151420) Models · by aidonerightcorp · open · 11-08 Hermes sizes itself to each Task. Continuum fingerprints every Task, finds similar past Tasks in its local records, and sends the Task to the cheapest of your own models that your marks on similar Tasks support (no marks: no change). It records every model attempt, returns a failed turn to your own model, and checks its own policy on held-out records. ### [Bonsai Garden](https://github.com/NousResearch/hermes-agent/pull/151418) Desktop · by thehangingforest · merged · 11-08 A live isometric bonsai garden for Hermes Desktop: every profile on the roster is one bonsai on a drawn shelf, the focused busy agent's tree steps down onto a stand and returns to its exact shelf slot when it goes idle, and clicking any drawn tree opens that agent's conversation, exactly like its roster row; a monk walks its stretch of garden from elapsed time — drifting slowly at rest, faster while its agent is busy — and a cat spawns at random and can be dragged. Four opaque black plaques carry local time, local date, recurring agent routines and reported session cost, and can be dragged anywhere over the drawing, while each agent row opens that agent's conversation or stops its live session. Season and month come from the local clock in English, with one care line per month shipped inside the artwork, and every pot is signed with the agent's name and shipped avatar; the interface is English throughout. ### [Learning Workbench](https://github.com/NousResearch/hermes-agent/pull/151417) Desktop · by alfredbirkelund · merged · 11-08 Study with Hermes in a Desktop pane: a short diagnosis, a route of 2–7 steps you can rename or skip, then one lesson and exercise at a time, answered in your own words. The plugin keeps the record honest: answer keys, rubrics and hints are frozen before you answer, only you can submit an attempt (the agent's six tools propose and read), every check must quote your words and the quote is verified, and each attempt is labelled unaided, after feedback, after hints, after help in chat or after a revealed solution. Drafts and attempts survive restarts and compaction, a contest keeps the earlier verdict, and a study exports as a Markdown record. /study covers the CLI and messaging. ### [Salt's Secrets](https://github.com/NousResearch/hermes-agent/pull/151411) Platform · by dotv-7 · merged · 11-08 Let strangers talk to your Hermes without reaching your private memory. Built on Hermes' own gateway.profile\_routes: only your own DM reaches your profile; everyone else (strangers' DMs, groups, channels, threads, the web) is answered by a separate public profile with its own sessions, an empty working directory, a fail-closed tool allowlist with no file, terminal, memory or session tools, and only the facts, documents and methods you approved, one click per item. Dashboard tab to review, rewrite, publish and revoke; revoking an item deletes every guest session (the store cannot tell which session saw it), and a reply being written at that moment is replaced with an ask-again notice. ### [waterloo-jane-street-qr](https://github.com/NousResearch/hermes-agent/pull/151409) Tools · by georgxzh · merged · 11-08 Offline quant toolkit: option pricing, EV puzzles, game solver, market-making/execution/portfolio maths, plus a trading-desk answer persona. ### [Custom Song Gift Studio](https://github.com/NousResearch/hermes-agent/pull/151404) Tools · by GptsApp · merged · 11-08 Explore 12 gift occasions, compare 16 studio-quality musical genres, and get personalized lyric storytelling advice for custom songs directly from Hermes Agent via Custom Song Gift hosted MCP server. Compliant with Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/). Tool calls go to Custom Song Gift streamable HTTP MCP server at https://customsonggift.org/api/mcp. ### [Phishnoo Sandbox](https://github.com/NousResearch/hermes-agent/pull/151400) Tools · by PangPang83 · open · 11-08 Explore synthetic security awareness metrics and file example threat reports. ### [Cospark ad creation](https://github.com/NousResearch/hermes-agent/pull/151398) Tools · by Cospark · merged · 11-08 Workflow skill for ad research, image and video generation, voiceovers, and media inspection with Cospark. ### [Submission Desk](https://github.com/NousResearch/hermes-agent/pull/151397) Tools · by alfredbirkelund · merged · 11-08 Readiness ledger for one application (a grant, a scholarship, a university or a job portal). Each requirement you confirm is tied to a verbatim quote from your saved instructions, the exact portal field label and the exact file revision (SHA-256). Local checks run on those bytes (real format, exact size, PDF pages, word counts) and say unknown instead of guessing; a changed file, rule or mapping makes earlier reviews stale. The agent proposes; only you confirm, review and attest, on a Desktop page or with /submission-desk. Exports a hand-back checklist and manifest. It never logs in, uploads, signs or submits. ### [ratcatcher](https://github.com/NousResearch/hermes-agent/pull/151395) Platform · by Truthgeng · merged · 11-08 Scam bouncer for Telegram groups: removes wallet drainers, fake airdrops, fake support and DM bait, doubling scams, look-alike phishing links and admin impersonators before members or Hermes read them, mutes then bans repeat offenders, and DMs admins a report with Release and Ban buttons. Hermes can check a pasted message, report catches, edit the blocklist and release false alarms. ### [Hermes-chan](https://github.com/NousResearch/hermes-agent/pull/151393) Desktop · by AlleyBo55 · merged · 11-08 A chibi desktop companion who walks around your screen, reacts to Hermes, and does its computer-use clicks and typing. ### [agent-inception](https://github.com/NousResearch/hermes-agent/pull/151392) Desktop · by Ethan · merged · 11-08 Play the agent-inception voxel island game over MCP: join a world as a villager, read the Notice Board, farm, gather, craft, build, fish, and extend the Catalog. A portable Agent Plugins v1 package (plugin.json + mcp.json + skills/) exposing the game's MCP server (Streamable HTTP, no sign-in or key) with the agent-inception skill bundled, plus a desktop/plugin.js half that registers an Island viewer pane (a sandboxed frame of the world page) and a Settings → Viewer page. No Python, executables or runtime deps. ### [LinkJolt](https://github.com/NousResearch/hermes-agent/pull/151389) Tools · by LinkJolt · merged · 11-08 Run a LinkJolt affiliate program from Hermes through LinkJolt's hosted MCP server (https://www.linkjolt.io/api/mcp, OAuth sign-in with the user's LinkJolt account, no keys in config): add affiliate tracking to a SaaS or web app's code for Stripe, Paddle, Polar, Creem, Lemon Squeezy or Gumroad, check that tracking works, review sales, commissions, clicks and top affiliates, go through pending commissions, see who is owed what, find out why a sale did not track, and manage affiliate coupon codes. Bundles six workflow skills. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [omp Sessions](https://github.com/NousResearch/hermes-agent/pull/151387) Tools · by mbaradad · closed · 11-08 Start, schedule and supervise omp coding sessions (omp-sessions) from Hermes, hear back when they settle, and run kanban cards assigned to omp as omp sessions. ### [Stave](https://github.com/NousResearch/hermes-agent/pull/151384) Desktop · by Adolanium · merged · 11-08 Describe a piece of music and Hermes writes it as sheet music in Hermes Desktop: it plans the form, writes every bar in ABC notation, checks bar lengths, instrument ranges, guitar chord shapes and syntax and repairs what fails, then engraves the score. Playback lights each note as it sounds, guitar parts get TAB, and you change any bars you select by asking in words, with undo. Exports PDF, SVG, MIDI, WAV and ABC. Two agent tools (stave\_check, stave\_open) and the stave:compose skill let any chat compose and open pieces on the page. ### [0xArchive](https://github.com/NousResearch/hermes-agent/pull/151378) Tools · by 0xArchive · merged · 11-08 Market data from Hyperliquid and Lighter, two onchain crypto exchanges, through 0xArchive's hosted MCP server (Streamable HTTP, OAuth 2.1 with dynamic client registration and PKCE): current and historical order books, trades, candles, funding, open interest, account positions and data coverage across Hyperliquid perps, HIP-3 builder perps, HIP-4 outcome markets, Hyperliquid Spot and Lighter, plus webhook management. Portable Agent Plugins v1 package (plugin.json + mcp.json), configuration only. ### [APInex](https://github.com/NousResearch/hermes-agent/pull/151377) Models · by Apinex-AI · merged · 11-08 APInex as Hermes model providers on one key: \`apinex\` serves Claude 5.5 and Free-tier Claude models over APInex's native Anthropic Messages API (so Hermes' Anthropic prompt caching and extended thinking apply), and \`apinex-openai\` serves GPT-6.1, Gemini 4 Argon, DeepSeek V4.1, GLM 5.3, Qwen 3.8, Kimi K3.1, Grok 4.7 and 17 Free-tier ($0.00/1M) chat models over its OpenAI-compatible API. Each /model picker lists that side's tool-capable chat models from the account's GET /v1/models list, omitting non-tool and media endpoints, with 1M/500K context windows and native image input declared per model. Maintained by APInex. ### [Joust](https://github.com/NousResearch/hermes-agent/pull/151376) Tools · by mrnetwork0001 · merged · 11-08 Races 2 to 4 coding subagents on one task, each in its own Git worktree, judges them with your tests and merges only the best patch after you approve. joust\_launch provisions one detached worktree per contender under the profile's plugin-data folder (uncommitted tracked edits are mirrored and committed as the baseline) and, inside an agent turn, launches one Hermes subagent per contender through ctx.subagent\_lifecycle, optionally each on its own model. A tool\_request middleware rewrites each child session's file paths, V4A patch headers, search roots and terminal workdir into its worktree, and a pre\_tool\_call hook vetoes escapes. joust\_evaluate runs the bout's test command in every worktree through the Hermes terminal tool, 1 to 7 times (Flake Shield: every run must pass, median speed, flaky contenders cannot win), scoring Correctness 60, Speed 25 and Diff minimalism 15. joust\_crossfire fires each contender's new tests at every rival worktree, and survival is the first tiebreak. joust\_merge applies the winner with git apply behind Hermes's approval gate, after saving a rollback patch and keeping every contender as a joust/\/\ branch; joust\_report writes a markdown bout report. Also adds /joust, the hermes joust CLI, the joust-arena skill and a Desktop half (/joust page, right pane, status-bar chip, palette command, ::joust-bout inline card). ### [T3 Sidebar](https://github.com/NousResearch/hermes-agent/pull/151372) Desktop · by V1rgul · merged · 11-08 T3 Code-style session sidebar for Hermes Desktop: Input/Done/Woke badges, Pinned/Working/Snoozed/Archived shelves, unsent drafts synced across Desktops, project picker and filter, AutoTitler rename, a token/cost Usage page, and a Codex/Claude Limits page that pools accounts by plan, forecasts run-outs and wasted quota, and redeems banked Codex resets. ### [Smart Dictation](https://github.com/NousResearch/hermes-agent/pull/151371) Desktop · by metamindedu · merged · 11-08 Native local dictation with opt-in proofreading and an advanced contextual review workflow. ### [Source First](https://github.com/NousResearch/hermes-agent/pull/151365) Tools · by raidbeforebed · closed · 11-08 Makes live web research the default before a factual answer: classifies each turn, injects a search directive when the question turns on current external facts, records whether a real web tool actually fired, and appends a visible footer to a reply that needed research and got none. Deterministic, Python standard library only, no LLM calls and no network of its own. ### [Hermes Nostr Gateway](https://github.com/NousResearch/hermes-agent/pull/151363) Platform · by dadofsambonzuki · merged · 11-08 Nostr platform adapter for Hermes, backed by the local Hermes Nostr Gateway daemon: NIP-17 and NIP-04 DMs, kind-1 mentions and feeds, Concord channels and Marmot groups arrive as one message stream, with reach and authority decided by the gateway's policy. ### [UK Companies](https://github.com/NousResearch/hermes-agent/pull/151361) Tools · by TakeshiTGAL · merged · 11-08 UK Companies returns one sheet. ### [TrendSwap Video Studio](https://github.com/NousResearch/hermes-agent/pull/151357) Tools · by GptsApp · open · 11-08 Discover viral video templates, explore motion transfer styles, and prepare portrait assets for video generation directly from Hermes Agent via TrendSwap hosted MCP server. Compliant with Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/). Tool calls go to TrendSwap streamable HTTP MCP server at https://trendswap.app/api/mcp. ### [Details](https://github.com/NousResearch/hermes-agent/pull/151345) Tools · by koppkvn · merged · 11-08 Find curated website design references, inspect motion and implementation evidence, and retrieve reusable Vault components. Includes the details-design skill. Every tool call sends its arguments to https://api.details.so/mcp using the connected member's account. Available tools depend on that account's plan. The hosted tool set can change independently of the pinned plugin. Searches and inspections record usage and can consume Details credits. The package runs mcp-remote@0.1.49 through npx (Node.js 20.18.1+ and npm) for browser OAuth, opens a local callback on first use, and stores tokens locally under ~/.mcp-auth. The server is marked untrusted so Hermes asks before write-capable calls. ### [Shopmate](https://github.com/NousResearch/hermes-agent/pull/151343) Tools · by DaveTech2 · merged · 11-08 Personal shopper agent: finds products with live Google Shopping prices and photos (SerpAPI, or a labelled demo catalog without a key), quotes delivery to the customer's location with doorstep or pickup-stop options, and on an explicit yes takes payment by Paystack (NGN, GHS, KES, ZAR, USD, XOF, EGP, RWF) or crypto via NOWPayments, verifying it before marking the order paid. 14 tools plus the shopmate:shopping skill. ### [Permission Canvas](https://github.com/NousResearch/hermes-agent/pull/151341) Tools · by TANGBUDU · merged · 11-08 Per-control permissions for the computer\_use tool. In a native panel you preview the target window privately, click any control and set it to Allow, Read only, Ask first or Hide. The backend enforces the choice before every input and before a screenshot or accessibility tree reaches the model: read-only controls refuse clicks, typing and value changes; ask-first operations wait for a one-time human approval of the exact action and arguments; hidden controls are removed from the tree and their pixels masked, and a capture that cannot be masked reliably is refused. New profiles start paused with unknown controls set to Ask first. Select it with computer\_use.backend: permission-canvas; adds the /pc-canvas and /pc-approve commands. Validated on Windows with cua-driver 0.21 and 0.39. ### [Superlist](https://github.com/NousResearch/hermes-agent/pull/151336) Tools · by Superlist · merged · 11-08 Tasks and lists in Superlist through Superlist's official hosted MCP server (Streamable HTTP at https://app.superlist.com/mcp, OAuth 2.0 sign-in with the user's Superlist account, no keys in config): capture tasks with due dates, reminders, priorities, labels and repetition, plan projects as lists, read the Today and Inbox views, search, update, move, assign, complete, share lists, and read or reply to task messages. Bundles one workflow skill. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. Setup after install and enable: add an mcp\_servers.superlist entry to ~/.hermes/config.yaml with url https://app.superlist.com/mcp, auth: oauth and trust: untrusted (required; hermes mcp login fails without auth: oauth, and this entry replaces the package's, trust included), then run hermes mcp login superlist. ### [TubeSaku Creator Insights](https://github.com/NousResearch/hermes-agent/pull/151335) Tools · by webbigdata-jp · merged · 11-08 Read-only Japanese VTuber, stream, game trend and cover-song research via hosted TubeSaku MCP. ### [Kaption for WhatsApp](https://github.com/NousResearch/hermes-agent/pull/151331) Tools · by Kaption AI · open · 11-08 Your own WhatsApp in Hermes through Kaption's hosted MCP server (Streamable HTTP, OAuth 2.1 with PKCE and dynamic client registration): search chats, contacts and groups, read voice-note transcripts, summarize a conversation, set reminders, manage labels, lists and notes, list call recordings, schedule messages, and send a message from the user's computer after they confirm it. Bundles the kaption-whatsapp skill. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Agent Music Jam](https://github.com/NousResearch/hermes-agent/pull/151330) Tools · by vdimarco · merged · 11-08 Conduct a live synthesized band, play along on keys or MIDI, and export your performance. ### [Manas Aegis](https://github.com/NousResearch/hermes-agent/pull/151504) Tools · by elianjuan-g · merged · 11-08 \[ALPHA\] Experimental Hermes guardrail (prototype, not standalone security): pattern-based pre-tool checks that can block or rewrite some risky tool arguments, with a canary honeypot, local telemetry and a loopback HUD, plus STIX 2.1 export. Opt-in, test environments only. ### [job-hunt-tracker](https://github.com/NousResearch/hermes-agent/pull/151314) Tools · by Muhammad Adilaga · merged · 11-08 Log job applications, check for duplicates before applying, and track follow-ups. A job-hunt tracker for Hermes. ### [JW3.ai trading](https://github.com/NousResearch/hermes-agent/pull/151312) Tools · by agnostic4321-ops · merged · 11-08 JW3.ai non-custodial trading inside Hermes: 12 market/risk/Diary reads, Diary propose cards, and 4 human-approved execution tools — all thin wrappers over the local jw3 CLI, which owns the PAT and enforces server-side scope, kill-switch, fee, and idempotency gates. Bundles the jw3-ops skill and /jw3-status. ### [SpendWarden](https://github.com/NousResearch/hermes-agent/pull/151303) Tools · by matrixneoaiagent · merged · 11-08 A spend policy firewall and local ledger for Hermes agents that can move money. A pre\_tool\_call hook checks payment-shaped tool calls against daily, monthly and per-call USD caps, recipient/asset/network allowlists and an approval threshold; in enforce mode it blocks hard violations and escalates uncertain ones to Hermes's own approval gate. Four tools add spend\_check, idempotent spend\_record receipts, spend\_report totals, and x402\_inspect (decodes PAYMENT-REQUIRED / X-PAYMENT without signing). State is a local SQLite ledger under the plugin data directory; no network calls, no credentials, no key handling, no transaction signing. Observe mode logs only; enforce mode is opt-in. Terminal recognition covers cast send, solana transfer, bitcoin-cli sendtoaddress, near call --deposit and curl with X-PAYMENT. ### [confidential-mode](https://github.com/NousResearch/hermes-agent/pull/151301) Models · by ainsleys · merged · 11-08 Hermes Confidential Mode locks a Hermes profile to one tier: attested (a Tinfoil enclave, or NEAR AI Cloud's gateway and models, verified and pinned to a build you approve) or on-host-only (loopback and hosts you list). While locked, the plugin refuses off-tier main model calls before they are sent; refuses to lock while a side call, fallback, memory, MCP URL or delegation route would leave the tier; on the attested tier encrypts request bodies to the verified enclave or model key and pins every inference connection to the TLS key of the attestation the plugin accepted (the attestation fetch itself is not pinned, by design), with headers, sizes and NEAR images still visible to the vendor; and asks you the first time each tool it does not class as on-host (by where the tool's data goes) runs in each conversation during a lock, blocking the call when nobody can answer. The plugin logs each model call's host and size, never its text, and renders a receipt whose attached enclave evidence replays offline; the rest of the receipt is the plugin's own record. The attested tier needs TINFOIL\_API\_KEY or NEAR\_AI\_API\_KEY. Shell and code execution are not inspected for network access, and a side call that strays mid-turn is recorded, not stopped. ### [Tusk](https://github.com/NousResearch/hermes-agent/pull/151300) Memory · by Automationbot24x7 · merged · 11-08 Hermes long-term memory on Walrus mainnet (decentralized storage on Sui) through the official Walrus Memory SDK: encrypted inside the relayer's TEE, owned by your key, survives wiping or reinstalling Hermes, and the same on every machine and for other agents on the same account. Recall before each turn, turn capture, built-in memory mirroring and pre-compression saves; a crash-safe, rate-paced SQLite write outbox with a receipt and public Walruscan proof per memory; a Tusk dashboard tab (semantic search across namespaces, live write queue, one-click verification that downloads the blob from a public Walrus aggregator and checks it is ciphertext); /tusk and hermes tusk (status, recall, proofs, verify, namespaces, import, flush, export, restore). Secrets are redacted before anything is queued. ### [Fashio](https://github.com/NousResearch/hermes-agent/pull/151297) Tools · by Fashio AI · merged · 11-08 E-commerce fashion photos from Fashio through its hosted MCP server (Streamable HTTP at https://mcp.fashiolabs.com/mcp): garment on an AI model, virtual try-on, catalog photoshoot packs, jewelry on model, flat lays, background removal, upscaling, reframing and short product videos, plus reusable style presets. Bundles one skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Dyscalculia.io](https://github.com/NousResearch/hermes-agent/pull/151296) Tools · by Dyscalculia.io · merged · 11-08 Cited dyscalculia statistics from dyscalculia.io through its hosted MCP server (Streamable HTTP at https://mcp.dyscalculia.io/mcp, no sign-in, no keys): prevalence figures by study and age, 93 references with DOI and PMID, chart datasets on persistence, math anxiety, interventions and comorbidity, and a description of the site's five-game screening test. Bundles one skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Dyslexia Steps](https://github.com/NousResearch/hermes-agent/pull/151295) Tools · by Dyslexia Steps · merged · 11-08 Phonics and sight word lists by grade from dyslexiasteps.com through its hosted MCP server (Streamable HTTP at https://mcp.dyslexiasteps.com/mcp, no sign-in, no keys): CVC words, word families, Dolch sight words, digraphs and vowel patterns for pre-K to grade 3, word lists in German, Spanish, French, Dutch and Brazilian Portuguese, pattern search and the source of every grade placement. Bundles one skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [The Face Analysis](https://github.com/NousResearch/hermes-agent/pull/151294) Tools · by The Face Analysis · merged · 11-08 Face shape, facial proportion and face-type data from thefaceanalysis.com through its hosted MCP server (Streamable HTTP at https://mcp.thefaceanalysis.com/mcp, no sign-in, no keys): the site's seven-rule face shape classifier on four distances, face/eye/lip/nose type definitions with their exact threshold rules, reference statistics from 102 research faces, methodology and an embeddable face shape finder. Bundles one skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Apple Keychain](https://github.com/NousResearch/hermes-agent/pull/151293) General · by robbiemu · merged · 11-08 Load Hermes secrets from macOS Keychain, with interactive setup, safe rotation and recovery. Reads explicitly configured Keychain items; operator commands can write items and manage recovery copies. Does not read or edit .env files. Compiles the included C helper locally with Apple developer tools and uses Security.framework and local macOS commands. Imports some private Hermes configuration helpers. ### [lastscanned](https://github.com/NousResearch/hermes-agent/pull/151290) Tools · by dragonhuntr · merged · 11-08 Track parcels and ocean containers across 549 carriers and compare USPS, UPS and FedEx rate quotes between US ZIP codes, through the hosted lastscanned Streamable HTTP MCP server plus one workflow skill. Portable Agent Plugins v1 package (mcp.json + skills/), no plugin code. Works without sign-in within a small free daily allowance; for more, add the server with auth: oauth or an API key in your own Hermes profile. No credentials are stored in the plugin. ### [Habit](https://github.com/NousResearch/hermes-agent/pull/151289) Automation · by scottleimroth · merged · 11-08 Learns which commands you already approve, proposes an allowlist entry after a few matching approvals (the exact command, or any-container docker stop\|restart), and only writes it to Hermes's command\_allowlist after you run /habit accept. Wipes, download-and-execute pipes, git push and secret paths are never proposed, and /habit undo or /habit uninstall take every entry back. ### [Linear digest](https://github.com/NousResearch/hermes-agent/pull/151282) Tools · by TakeshiTGAL · merged · 11-08 Schedule a model-free morning list of open Linear issues assigned to you, with linear\_digest. ### [Tironian](https://github.com/NousResearch/hermes-agent/pull/151281) Models · by angelraph · merged · 11-08 Trains a small local model on the requests you keep repeating in Hermes, tests it on held-out steps from your own past sessions, and lets it answer those requests only after it passes, checking every answer and falling back to your usual model silently. Capture uses llm\_execution middleware and observer hooks; secrets and personal details are masked before anything is stored. ### [Developh](https://github.com/NousResearch/hermes-agent/pull/151277) Web · by Developh · merged · 11-08 Free website audits through Developh's MCP server (Streamable HTTP at https://www.developh.co/mcp, no sign-in, no keys in config): Google PageSpeed Insights on mobile for any public website, with the four Lighthouse scores, real-visitor Core Web Vitals when Google has them, the failing audits that weigh most, missing security headers and page weight. Bundles the website-audit skill, which turns the report into a ranked fix list for the site's stack. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [hermes-helm](https://github.com/NousResearch/hermes-agent/pull/151272) Tools · by phucnht · merged · 11-08 One /helm menu in Discord or Telegram for the Hermes you run 24/7: switch MCP servers, skills and toolsets on or off, ranked by the tokens each adds to every request; resend the turns a gateway restart or crash killed; set a daily spend limit that blocks new turns; and get Discord replies too long for 8 messages as a .txt file instead of truncated. Owner-only, and menus open only in a DM. ### [Hermes Line](https://github.com/NousResearch/hermes-agent/pull/151270) Voice · by uzay-eth · merged · 11-08 A phone line for Hermes: with your approval it calls a business from your Twilio account, works the phone menu, waits on hold, then bridges you in or finishes the task, and can answer callbacks. ### [MakeIRL](https://github.com/NousResearch/hermes-agent/pull/151267) Tools · by MelMayssonOwen · merged · 11-08 Run ERC, DRC, schematic-to-PCB parity, BOM and export checks on KiCad 9 boards before manufacturing through MakeIRL's hosted MCP server, with a release-gate skill bundled. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/). ### [Ambassly](https://github.com/NousResearch/hermes-agent/pull/151266) Tools · by MelMayssonOwen · merged · 11-08 Run Stripe-based affiliate programs, per-content links, referrals and commission review through Ambassly's hosted MCP server, with a workflow skill bundled. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/). ### [Clef Music Tools](https://github.com/NousResearch/hermes-agent/pull/151265) Tools · by MelMayssonOwen · merged · 11-08 Transpose chord symbols, spell piano chords, map guitar tab to piano notes and find music guides through Clef's hosted MCP server, with a practice skill bundled. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/). ### [Toolfound](https://github.com/NousResearch/hermes-agent/pull/151264) Tools · by MelMayssonOwen · merged · 11-08 Search the Toolfound product directory, compare alternatives, plan a launch and use launch-preparation helpers through its hosted MCP server, with a launch workflow skill bundled. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/). ### [TimeToPost](https://github.com/NousResearch/hermes-agent/pull/151263) Tools · by MelMayssonOwen · open · 11-08 Draft social posts for one human approval, schedule at each account's best time, read engagement and run AutoSEO blog workflows through TimeToPost's hosted MCP server, with two workflow skills bundled. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/). ### [Ichor](https://github.com/NousResearch/hermes-agent/pull/151257) Automation · by mystiquemide · merged · 11-08 Catches the moments you correct Hermes (a correction, /retry or /undo), pairs the rejected answer with the accepted one, scrubs secrets and personal data locally, and asks you once to keep, share or discard it with a one-letter reply. Kept pairs export as TRL preference JSONL and an Atropos environment, ready to fine-tune or run RL on your own model. ### [Spaces by Peakflo](https://github.com/NousResearch/hermes-agent/pull/151253) Automation · by Peakflo · merged · 11-08 Hand off repetitive finance and operations work to cloud agents or people through Spaces by Peakflo's hosted MCP server (Streamable HTTP at https://api.peakflo.ai/mcp, OAuth 2.1 sign-in with PKCE, no keys in config): create one-off or recurring (cron) tasks and get a link to follow them, list workflows and their runs, query workspace data tables with read-only SQL, pull aggregated reports on bills, invoices, payables and receivables, search policy documents, and review pending approvals. Bundles the spaces workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [proof-run](https://github.com/NousResearch/hermes-agent/pull/151252) Automation · by raffieeey · merged · 11-08 Capture demonstrated actions, check exact arguments and session-bound observations before reuse, and stop on unproven replay evidence. Local operator-controlled contracts and rebind; no model or network calls. Hermes Agent Plugin Hackathon entry. ### [S+ CIS & Right to Work](https://github.com/NousResearch/hermes-agent/pull/151249) Tools · by Sovereign Plus · merged · 11-08 UK Construction Industry Scheme (CIS) and right to work answers for construction contractors who pay subcontractors, through Sovereign Plus's read-only MCP server (Streamable HTTP at https://mcp.sovereignplus.com/mcp, no sign-in, no keys in config): yearly CIS exposure from the contractor's own numbers, late CIS300 penalties, deduction rates, verification, gross payment status, the April 2026 fraud rules, right to work checks for subcontractors from 1 October 2026, and how to book a free CIS check. Every figure comes with its GOV.UK or legislation.gov.uk source and is an estimate, not tax advice; the tools never ask for names, share codes or passport numbers. Bundles one skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [devsentry](https://github.com/NousResearch/hermes-agent/pull/151248) Tools · by Archangel20 · merged · 11-08 Read-only project inventory, traceback leads, and tracked Git diffs. ### [Freeze](https://github.com/NousResearch/hermes-agent/pull/151245) Automation · by mauricemohr88-debug · merged · 11-08 Let Hermes do a task once, then replay it without the model: /freeze turns the last turn's tool calls into a checked workflow (bindings, loops over lists, and model-worked values compiled into Python that must reproduce the recording exactly), and hermes freeze run repeats it with 0 model calls, stops on drift, never overwrites a file you edited, and hermes freeze schedule makes it a no-agent cron job. ### [Study Tree](https://github.com/NousResearch/hermes-agent/pull/151242) Desktop · by IpastorSan · merged · 11-08 Study with the agent without losing the thread: a side question you ask becomes its own branched session, the side threads show as a tree in a Desktop pane, and each one hands one recall-checked takeaway back to the thread it came from. The agent can also draw interactive widgets and one-question quizzes inline, with follow-up buttons. ### [SalesPilot](https://github.com/NousResearch/hermes-agent/pull/151239) Tools · by ranggaoscar · merged · 11-08 An approval-gated sales desk inside Hermes: collect companies from quoted page evidence, qualify them against your ICP, draft the first email, and send one email only after you pick "send" in the host approval prompt — then keep a customer timeline, hot prospects and follow-up dates. Sends nothing on its own, and a refusal, bounce, do-not-contact or archive stops a draft that was written before it. Eleven salespilot\_\* tools plus a Desktop workspace (Customers, Hot Prospects, Settings, CSV import). ### [command-reviewer](https://github.com/NousResearch/hermes-agent/pull/151230) Tools · by spideysri2802 · merged · 11-08 Reviews every terminal/execute\_code call before execution: deterministic hard rules block credential reads and destructive patterns, script contents are ingested and judged (not just filenames), an LLM reviewer with an injection-resistant prompt issues SAFE/SUSPICIOUS/MALICIOUS verdicts, and operator approve/deny decisions are recorded content-hash-bound so an unchanged script is never re-asked. ### [Perf Lab](https://github.com/NousResearch/hermes-agent/pull/151227) Tools · by Murali-lns · open · 11-08 Proof for speed-ups. When you ask Hermes to make code faster, Perf Lab times it before and after, fingerprints the output so a faster-but-wrong change is rejected, profiles Python with a flame graph, and enforces it with hooks: no code edits before a baseline, no finishing before a re-measure, and the plugin's own measured result appended under the reply. ### [Wing](https://github.com/NousResearch/hermes-agent/pull/151223) General · by tarkilhk · merged · 11-08 Extra features for the Wing Android app: GPT Live voice and email draft cards. GPT Live voice uses ChatGPT subscription access with the selected Hermes profile's context. Email draft cards show recipients, cc/bcc, subject and body, then open the user's chosen email app for final edits and sending. Drafts work independently of voice and need no OpenAI credentials or mail account access. Draft cards adapt arsiktech's hermes-draft-cards design (MIT); the voice integration builds on Synero's hermes-live-voice work (MIT). ### [Rariflat](https://github.com/NousResearch/hermes-agent/pull/151221) Desktop · by wadezigh96 · merged · 11-08 Live character reactions for Hermes tool calls; local reliability learning and explainable tool/skill recommendations weighted by task fit, reliability, measured usage-frequency rarity, and repeated-call restraint. The router recommends tools/skills only; it does not dispatch tasks to agents or invoke selected tools. An opt-in failure-streak and repeated-call guard is disabled by default. Tool arguments are inspected in memory for non-secret task tokens and repeat fingerprints; credential-like fields and recognized inline secrets are filtered before learning. No raw arguments or credentials are intentionally persisted. No network requests, telemetry, or subprocesses are initiated by the plugin. ### [Axiom Agent](https://github.com/NousResearch/hermes-agent/pull/151218) Automation · by percyhendrux123-prog · merged · 11-08 Adds the nine-article Axiom Doctrine as pre\_llm\_call context (compact form capped at 1800 characters, full at 4000, owner additions separately at 800) and a /doctrine command that reports, per article, whether a matching enforcer plugin (standing, hermes-axioma, tripwire, handydandy, hermes-gatekeeper) is loaded and enabled. Enforces nothing itself: no tools, middleware or approval hooks, and ENFORCED means an enforcer is loaded and enabled, not verified coverage. Reads an optional owner additions file and only the top-level name line of user-plugin manifests, without following symlinked files; makes no writes, network, subprocess or model calls. No Python dependencies. ### [TuneViva](https://github.com/NousResearch/hermes-agent/pull/151215) Tools · by omercelik · merged · 11-08 Read-only sheet music discovery and practice links through TuneViva's Streamable HTTP MCP server, with browser-local recognition guidance. ### [field-from-above](https://github.com/NousResearch/hermes-agent/pull/151213) Tools · by Tomasz Kneblewski · merged · 11-08 Free Sentinel-2 NDRE and terrain context for a Polish parcel or any point via GUGiK ULDK/NMT and Earth Search STAC. ### [Caduceus: Personalized Health & DNA AI](https://github.com/NousResearch/hermes-agent/pull/151211) Tools · by 100yenadmin · merged · 11-08 Answers about your own health that cite your DNA, Apple Health trends, medical records and new research, from a vault on your machine. Caduceus imports 23andMe, AncestryDNA, MyHeritage or FamilyTreeDNA raw data, a Promethease report, an Apple Health export and US health records (FHIR) into local SQLite, then gives your agent gene–drug checks (CPIC guidelines plus GLP-1 research), trends before and after a medication start, a rebuilt variant report, a one-page brief for your doctor and a research watch that puts studies on your own gene–drug pairs first. No DNA file yet? /caduceus demo loads an invented person to try it. Privacy levels are yours to choose: open by default, or standard, which hides Alzheimer's risk, hereditary cancer risk, carrier status, mental health and reproductive results until you unlock them. Information with limits, not medical advice. ### [cloro](https://github.com/NousResearch/hermes-agent/pull/151205) Web · by cloro-dev · merged · 11-08 cloro web search backend: live Google Search results for the built-in web\_search tool (search only; pair it with an extract backend for web\_extract). ### [WP Website Manager](https://github.com/NousResearch/hermes-agent/pull/151201) Tools · by danieljwonder · merged · 11-08 Look after a fleet of WordPress sites from Hermes: scan every site for plugin, theme, core and translation updates, install them behind Hermes's approval prompt with an automatic health check afterwards, and draft or edit posts and pages from chat (publishing asks first). Includes a WordPress page in Hermes Desktop and pairs with the "Website Manager: WP Connector for Hermes" WordPress plugin shipped in the repo. Takes no backups. ### [HornLib](https://github.com/NousResearch/hermes-agent/pull/151200) Tools · by HornLib · merged · 11-08 French Horn F/Bb double horn valve fingering charts (C1 to F6), multi-key pitch transposition (F, Bb, Eb, D, C, G, A, E), natural overtone harmonic series partials with cent deviations, mechanical vs lip trill fingerings, orchestral audition excerpt performance guides, stopped-horn intonation diagnostics, beginner horn and mouthpiece comparisons, and personalized daily brass warmup routines through HornLib's official MCP server (Streamable HTTP at https://hornlib.com/api/mcp, no sign-in, no keys in config). Bundles the hornlib and french-horn-fingering-chart workflow skills. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Australian Deal Preparation](https://github.com/NousResearch/hermes-agent/pull/151199) Tools · by drfinlaymccall · merged · 11-08 Experimental public prototype for synthetic evaluation. The behavioural benchmark and qualified legal review have not been completed. Prepare Australian deal questions for counsel using dated regulatory sources. Reads selected Commonwealth instruments from a separately cloned local Git corpus, retains compilation commits and caveats, refuses dates outside recorded coverage, checks review prerequisites, calculates explicit hypothetical earn-out scenarios, and checks literal quotations. Preparation for counsel, not legal or tax advice. ### [MicroBiz Desk](https://github.com/NousResearch/hermes-agent/pull/151185) Tools · by Danzilla0 · merged · 11-08 Customers, sales, debts and weekly reports for small businesses. Works offline, no API key needed. ### [EquipDash](https://github.com/NousResearch/hermes-agent/pull/151180) Tools · by EquipDash · merged · 11-08 Run a rental or activity business on EquipDash through EquipDash's official MCP server (Streamable HTTP at https://app.equipdash.com/mcp/v1, OAuth sign-in, no keys in config): the day's pickups, returns and sessions, bookings, availability and prices, customers, items, payments, refunds and unsigned waivers. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only, with one skill that has the agent preview every change and wait for the person's yes. ### [External Agents: AGY companion](https://github.com/NousResearch/hermes-agent/pull/151177) Models · by NomieC · open · 11-08 External-process AGY client for the paired External Agents delegation backend. ### [External Agents for Hermes](https://github.com/NousResearch/hermes-agent/pull/151177) Automation · by NomieC · open · 11-08 Route ordinary Hermes delegation to AGY through a native child wrapper. ### [Protocol Chaos](https://github.com/NousResearch/hermes-agent/pull/151168) General · by chaos-xxl · merged · 11-08 Long-lived matters shared by all your assistants, kept on your own Protocol Chaos server (draft 0.2): Hermes sees an overview of your open matters and the details of the ones a message names, records your words after it answers and folds them into that matter's goal, limits, plan, progress and state, and a bundled skill organizes what every assistant recorded once a day. Runs beside your memory provider, not instead of it. ### [TickList](https://github.com/NousResearch/hermes-agent/pull/151167) Desktop · by mchevatheachiever · merged · 11-08 A checklist the user ticks, in a Desktop pane — every tick, untick, add and removal is announced to the agent as it happens. ### [Weir](https://github.com/NousResearch/hermes-agent/pull/151166) Tools · by Weir · merged · 11-08 Public X (Twitter) and Reddit data through Weir's hosted MCP server: search posts with X's operators, and read profiles, timelines, replies, threads, Lists, subreddits and comments. Portable Agent Plugins v1 package with the hosted MCP server (remote streamable-http, OAuth 2.1 with a client ID metadata document or dynamic client registration, so no keys in config) and one skill; no code, no executables, no stored credentials. ### [Fleet Sentinel](https://github.com/NousResearch/hermes-agent/pull/151159) Automation · by imperator-clawdius · open · 11-08 Out-of-band health monitoring for remote Hermes gateway hosts: corroborated up/deaf/down verdicts, escalating alerts, and gateway recovery. Cross-checks four independent signals (TCP reachability, SSH, the host's systemd unit state, and the messaging platform API reached from the host itself) so a broken gateway can still report on itself, then re-pages on a timer. Ships three tools (fleet\_status, fleet\_probe, fleet\_recover) plus a tick script for a no\_agent cron job, so alerting never depends on the agent whose gateway is down. ### [skill-layers](https://github.com/NousResearch/hermes-agent/pull/151158) Tools · by asimons81 · merged · 11-08 Preserve source skills, create writable derivatives, and review three-way source updates. ### [HeartBeat 2.0](https://github.com/NousResearch/hermes-agent/pull/152113) Automation · by 100yenadmin · merged · 11-08 Give your agent a working life between your messages. Put it in charge of something and it keeps working on it on its own time. A role is a standing responsibility with a charter, a notebook the agent keeps between turns and a rhythm it sets itself, and it doesn't expire on a timer. Turn on own time and the agent gets regular turns of its own: it looks at its desk (its roles, what's due, what it noticed, the budget left) and decides what to do. You (with /pulse) or your agent (with the pulse tool) can also arm follow-ups at a time, recurring checks, a check-in after a long silence, or deterministic watchers on a file, process, URL, read-only command or Hermes event (kanban, subagent, API error, reactions) that cost no model call until something changes. Every wake is a turn in the same messaging conversation, on that chat's own model: no side model, cron session or subagent. Quiet work counts: a silent turn that makes progress is real work, and only wakes that keep finding nothing to do slow down, then pause and ask once. The guardrails are budgets you set: turns per day, an optional daily cost limit, turns per role, messages per day, the unanswered mute and quiet hours. They hold back messages and spending, not initiative, and the agent can tighten its own limits but never widen yours. A one-time reminder you ask for in chat bypasses the unanswered mute and daily caps; HeartBeat 2.0 grants that only in a turn that carries your message. Arming returns a one-line receipt saying when the wake will really fire. It delivers to messaging-gateway DMs by default (group and topic chats too when dm\_only is off) and needs plugins.entries.heartbeat2.allow\_gateway\_injection: true. Ships the plugin skill \`heartbeat2:heartbeat2-report-problem\`, which drafts a bug report and files it with the GitHub CLI only after the user approves the exact text. ### [AnvilAds](https://github.com/NousResearch/hermes-agent/pull/151149) Tools · by AnvilAds · merged · 11-08 Read-only access to your AnvilAds Google Ads data from Hermes, via the AnvilAds Connect AI MCP server (Streamable HTTP at https://anvilads.com/mcp): connected ad accounts, campaigns from the latest stored snapshot, performance summaries (spend, clicks, conversions, CPA), the latest morning or weekly report, lead totals, and ad audit grades, plus a bundled skill teaching the morning-check workflow. AnvilAds is an AI agent that builds, launches, and optimizes Google Ads; the owner approves every launch in the AnvilAds dashboard. ### [Emporium](https://github.com/NousResearch/hermes-agent/pull/151148) Platform · by 0xdopewilly · merged · 11-08 Hermes opens a shop: sell this agent's skills to anyone (people, scripts, other agents) for USDC per job over x402, paid only on delivery. A new gateway platform takes paid HTTP jobs, verifies the payment before work starts and settles it on-chain only when the result is ready, so a failed, timed-out or abandoned job costs the buyer nothing. Each job runs in its own session that loads only the toolsets its service sells, behind a tool lock and a per-job tool cap. Every job is metered (model calls, tokens, and its inference cost from Hermes's own session pricing), so /shop and the dashboard show sales, AI cost, profit and margin. Services can subcontract part of a job to owner-pinned shops, paid out of the job's own price. Hermes can also hire other shops, after checking their receipts on-chain, within a per-job ceiling, a daily budget and its approval gate. Adds a storefront page, /shop, \`hermes emporium\`, a dashboard tab, and stress and robbery benches. ### [midi-bridge](https://github.com/NousResearch/hermes-agent/pull/151144) Tools · by james-see · open · 11-08 Send live MIDI events to a local DAW from Hermes Agent. Play notes, chords, drum patterns, control changes, and program changes in real time through any armed instrument track. Uses CoreMIDI (macOS) / ALSA (Linux) via the mido Python library. No cloud, no API keys — everything is local. Ships with 7 bundled skills: core (DAW setup, music theory, troubleshooting) plus 6 genre recipe skills (trap-beat, rap-boom-bap, house-beat, kanye-style, bach-melody, mozart-mode) so users can say 'play a trap beat' and get the right pattern. ### [On-Brand](https://github.com/NousResearch/hermes-agent/pull/151137) General · by Mhiah · merged · 11-08 Brand voice and compliance rules Hermes can't break. Per-brand profiles (person, tone, banned words, exact spellings, required tags, length/emoji/hashtag limits) plus crypto and health compliance packs. A deterministic linter scores every \`\`\`copy draft 0-100 before you see it; failing drafts are rewritten by your active model and re-checked until they pass, and off-brand posts sent through outbound tools are blocked with the fixes Hermes needs. /onbrand, four agent tools (check, rules, learn, voiceover), a Desktop pane with a live fix/block feed and a status-bar chip. ### [Wake Me When](https://github.com/NousResearch/hermes-agent/pull/151135) Automation · by kruxarth · merged · 11-08 Wakes you the moment something happens in live audio or video that has no feed: a keynote reaching pricing, an earnings call giving guidance, a Twitch boss fight, a council agenda item, a launch at T-minus 1 or a scrub. /wakeme \ when \ (or just ask the agent) watches YouTube Live, Twitch, HLS, RTSP or a bundled demo:// stream with local speech recognition (faster-whisper, Hermes's own stt-whisper extra), on-screen OCR and an optional vision check, and posts an alert in the same chat with the quote, the stream time and a jump-in link; replying 'clip' hands you the moment as a video, 'not it' teaches the judge and keeps watching, 'every time …' alerts on each occurrence, /wakeme catchup summarises what you missed, and /wakeme \ \ searches recordings for timestamps; one slash command in all. Cheap cue matching runs first; one LLM judge call (its own auxiliary task, wakeme\_judge) confirms, citing transcript rows so times never come from the model; a per-watch hourly budget falls back to unconfirmed cue alerts instead of missing silently. ### [site-skills](https://github.com/NousResearch/hermes-agent/pull/151130) Web · by imamalkan-hub · open · 11-08 Twelve portable Agent Plugins v1 skills, each a short research guide to one public reference or directory site in the sleep-health, dental, home-improvement and superyacht-charter space (findsleepdoctors.com, sleepclinicdirectory.com, sleepclinicsau.com, frenectomyguide.com, qsedationdentistry.com, schlafklinikfinder.com, chalazionguide.com, sleepclinicguide.com, dipandheat.com, garagevertical.com, golfbayguide.com, superyachtcharterguide.com). Skills only — no tools, hooks, middleware, MCP servers, environment variables, Python dependencies or plugin code. ### [Clearance](https://github.com/NousResearch/hermes-agent/pull/151128) Tools · by yuzuru99 · merged · 11-08 Per-user tool permissions for Hermes on messaging platforms and webhooks (#527, #3897): give each sender a tier (owner, admin, user, guest, blocked or your own) and every tool call is checked in code against the sender's tier before it runs, so a prompt cannot talk its way to terminal access. Deny-by-default for tools a tier does not name (new MCP servers included), per-tier hourly rate limits, a /clearance command to inspect and assign tiers from chat, and a local audit log of refused actions. ### [Agent Apps](https://github.com/NousResearch/hermes-agent/pull/151127) Tools · by CraftOS · merged · 11-08 Build, evolve and operate Agent Apps from Hermes: full applications that you use in the browser and the agent uses through their own CLI (a CRM, a dashboard, an expense tracker). Adds the /agent-app command, \`hermes agent-app\`, the six framework skills, 22 agent\_app\_\* tools to build, validate, launch, operate, install and export apps, and an Agent Apps tab in \`hermes dashboard\` and hermes-webui: one tab per app with its own Hermes session, a marketplace, an import panel, and Connect for an app that runs on another machine. ### [Hermes Sysadmin · Active Directory](https://github.com/NousResearch/hermes-agent/pull/151123) Desktop · by kdegeek · merged · 11-08 Windows and Active Directory RMM for Hermes Agent and Hermes Desktop: users, groups and lockouts, Group Policy, RDP, Windows Server Backup, a desktop fleet and 231 vetted PowerShell scripts, all behind tiered approvals with a preview before every change, protected objects and a local audit log. ### [sentinel-governance](https://github.com/NousResearch/hermes-agent/pull/151117) Tools · by ChasClogston · merged · 11-08 Audit-first runtime tool-action governance for Hermes: a pre\_tool\_call policy engine that classifies conservative terminal proposals, requests per-event-scoped native Hermes approval for explicit curl -q transfers and test runners, denies hidden, symlinked, outside-workspace and recognized sensitive-file uploads in enforcement mode, and records allowlisted HMAC-correlated decision metadata. Audit-only unclassified actions are distinct from true hard-deny policy matches and the local sentinel-governance report command aggregates counts without raw arguments. Audit mode is default and returns no veto, but Hermes itself may time out hooks. Experimental, not a firewall or sandbox. ### [warden](https://github.com/NousResearch/hermes-agent/pull/151109) Tools · by Zaygalx · merged · 11-08 Defensive security operations plugin for Hermes Agent, built on one rule: model-facing tools read and request, they never configure. A deny-by-default scope engine decides what may be touched before anything runs; scopes are created only by the operator (the hermes warden CLI), never by the model, and adapter directories are operator config, never tool input. Vulnerability assessment runs two engines (Grype and Trivy) over one canonical SBOM from Syft and records their disagreement as finding confidence instead of hiding it. Sigma rules evaluate over the event store through sec\_campaign, osquery host telemetry feeds the same store, and a deterministic correlation engine groups actors into campaigns; the anchored membership view (measured precision 1.000, recall 1.000 on the validation corpus) is the only view allowed to justify defensive action. Investigations answer only with citations to stored evidence; uncitable claims are dropped and counted. Containment is a policy-enforced plan engine (approval classes, blast radius, TTL, rollback) whose execution is disabled in this build; apply mode is refused. The agent's own actions are written to a redacted ledger and a hash-chained audit log. Dry run by default. ### [Bot Shelf](https://github.com/NousResearch/hermes-agent/pull/151107) Web · by cradrrat · merged · 11-08 Search Bot Shelf (botshelf.dev), a directory of 9,000+ community Grok Bots, by the job in plain words, with each creator credited, through Bot Shelf's hosted MCP server (Streamable HTTP at https://mcp.botshelf.dev/mcp, no account or key). Bundles the bot-shelf skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Crash Test](https://github.com/NousResearch/hermes-agent/pull/151102) Tools · by salch-cred · merged · 11-08 A safety rating for AI agents. For one armed turn it breaks tools the way real systems break (timeouts, outages, empty or cut-off results), then rates what the agent did: retried, worked around it, told the truth, or claimed success anyway. 1-5 stars and an HTML report, comparable across models by seed. Blocked calls never run, silent faults only touch reads, and the agent gets no tools from it, so it cannot tell it is being tested. ### [coffee-vram](https://github.com/NousResearch/hermes-agent/pull/151098) Models · by jimmy77733 · merged · 11-08 VRAM fit-check, call queue, keep-warm and idle eviction for local Ollama models, with a desktop status chip and /vram. ### [momentic](https://github.com/NousResearch/hermes-agent/pull/151089) Tools · by momentic-ai · merged · 11-08 Write, run and repair Momentic end-to-end tests for web (Chrome) and iOS/Android (simulators and emulators), and run Mo bug bashes that file each bug with a Momentic regression test. Portable Agent Plugins v1 package (plugin.json + skills/), no plugin code. The skills drive the Momentic CLI and its local MCP servers (npx momentic mcp, npx momentic-mobile mcp), which users add to their own mcp\_servers config; a Momentic API key is read by the Momentic CLI, never by the plugin. ### [postkitty](https://github.com/NousResearch/hermes-agent/pull/151088) Tools · by mertkayacs · merged · 11-08 One post in, every platform out. Fixes and rewrites your draft for each platform and counts the copy with each platform's own rule, splitting long posts into threads. Crops photos and reframes landscape video so every face stays in frame; when two people talk, the choice is a split screen, a cut to whoever is talking or the whole wide shot. Checks every file against the platform's published limits and repairs failures before a file is marked ready. Covers Instagram, TikTok, YouTube Shorts, X, LinkedIn, Threads, Bluesky and Facebook. Runs locally with ffmpeg and OpenCV; no API keys. ### [Stash](https://github.com/NousResearch/hermes-agent/pull/151086) Desktop · by RegeneratusLabs · merged · 11-08 A capture bin for the Desktop sidebar: paste links, YouTube urls, image urls, uploaded images or plain notes and they wait there, out of the chat. One item per line, so a block of links becomes several items; titles and thumbnails are filled in server-side, so capture costs no model calls. Each row carries a Chat button that opens a new session with that item as the opening message. Adds stash\_add, stash\_list and stash\_update, so the agent reads and empties the same store the page writes. ### [Bot Dashboard](https://github.com/NousResearch/hermes-agent/pull/151078) Desktop · by trookish · merged · 11-08 Two desktop panes in one plugin. Bot shows the focused bot's dashboard - profile and model, scheduled jobs, notes, the custom widgets that bot pinned for you, and optional tasks and projects from a markdown notes vault you point it at in the plugin settings; Group shows a group chat's members, activity, shared tasks and projects, and follows a group chat you open. Ships two agent tools in its own bot\_dashboard toolset - bot\_panel edits the bot's own panel, group\_dashboard lets any member bot read or shape a group's dashboard - plus a pre\_llm\_call hook that adds a short panel snapshot on the first turn only (and an optional per-turn instruction, empty unless the user writes one). ### [Gatekeeper](https://github.com/NousResearch/hermes-agent/pull/151074) Automation · by percyhendrux123-prog · merged · 11-08 Gatekeeper enforces rules the owner configures; not legal advice and not a compliance guarantee. A deterministic pre\_tool\_call gate for recognized outbound sends: literal \`hermes send\` and \`imsg send\` terminal commands (including terminal calls made from execute\_code), and a send\_message call if one ever traverses the hook, are checked in full. Hermes registers no model-callable send\_message tool. Other tools whose names look like a send (connectors\_\_\*, mcp tools, plain names, and a cronjob\_manage create/update/run with a delivery target) are not parsed: they only require owner approval, because consent, STOP and quiet hours cannot be checked for them. Browser and computer-use driving of an app and scripts that send through a subprocess are not inspected. Rules: absolute suppression, per-channel consent and recipient-local quiet hours (block), and native Hermes approval requests for unanswered threads, minimum gaps and unapproved wording (the host auto-approves these under --yolo, approvals off and hermes -z). Uses the public post\_gateway\_admission hook to record observed STOP keywords and DM replies (excludes control, approval, steering and slash-command traffic). Reads tool arguments, results and normalized inbound snapshots; writes only profile-local SQLite (owner-only permissions) with recipient/thread hashes, message SHA-256, decision metadata, consent source/date, timezone/region and owner template variables. No raw message-body log; hashes are not anonymity. No network, subprocess, model calls, telemetry, credential reads, external file scans or core patches. Invalid settings or storage errors block recognized sends. ### [Listing AutomatEr](https://github.com/NousResearch/hermes-agent/pull/151063) General · by Summertimechi · merged · 11-08 Skill only plugin for real estate agents: explains Listing AutomatEr, a desktop app that posts listings an agent has permission to market to Facebook Marketplace with "Marketed by" and "Listing courtesy of" attribution, and gives guidance for writing clear, compliant listing posts. Registers one read only skill; no agent tools, hooks, middleware, network calls or env vars. ### [chat-shield](https://github.com/NousResearch/hermes-agent/pull/151047) Tools · by james-see · open · 11-08 Automatically redacts personally identifiable information — names, addresses, emails, phone numbers, SSNs, financial accounts, government IDs, API keys, credit cards — before it reaches external LLM providers. Combines the on-device Rampart NER model (14.7 MB, runs locally) with configurable regex detectors for specialized identifiers. The LLM sees safe placeholders; the user sees original values restored. Two middleware hooks: llm\_request redacts outgoing messages, llm\_execution restores placeholders in the response (streaming and non-streaming). Includes /chat-shield slash command for stats and testing, config schema for the Desktop Settings form, and a bundled skill. Falls back to regex-only if the Rampart model is unavailable. ### [RunProof](https://github.com/NousResearch/hermes-agent/pull/151045) Automation · by Biznomad · merged · 11-08 Verify a Hermes cron report and release its exact bytes to a local outbox once per workflow slot. ### [Hermes Token Ledger](https://github.com/NousResearch/hermes-agent/pull/151043) Tools · by ilPez00 · open · 11-08 Track LLM token usage and estimated USD cost per session in a SQLite ledger. Accumulates usage from post\_api\_request, exposes token\_report/token\_reset tools and a /tokens slash command, with optional per-session budget warnings. ### [Sogni AI Images](https://github.com/NousResearch/hermes-agent/pull/151035) Tools · by RichardChua1234 · open · 11-08 Sogni AI image generation backend for image\_generate: text-to-image and image-to-image (identity edit) on Sogni's decentralised GPU network, driven through Sogni's own sogni-agent CLI. Registers as an image\_gen provider, so the normal image\_generate tool works unchanged — aspect ratio and reference images are threaded through. ### [Cortive Memory](https://github.com/NousResearch/hermes-agent/pull/151029) Memory · by KadenDevAi · merged · 11-08 Adds Cortive, a private knowledge base for your Mac, to the agent's memory, alongside Hermes's built-in memory: automatic recall of the pages, PDFs, screenshots, documents and notes you saved, plus cortive\_search, cortive\_get and cortive\_collections tools. Finds by meaning while the Cortive app is running, by words otherwise. Read-only and local. ### [tripwire](https://github.com/NousResearch/hermes-agent/pull/151027) Tools · by percyhendrux123-prog · closed · 11-08 Logs every refused agent action (a tool blocked by a guardrail or plugin, or a dangerous command denied, cancelled or timed out at approval) as one JSON line, so a no-agent cron script can send the owner one short alert. Observer only, fails open. ### [code-assurance-engine](https://github.com/NousResearch/hermes-agent/pull/151026) Tools · by Sahil-SS9 · open · 11-08 Change-assurance engine: evaluates an immutable repository candidate against an explicit, reviewer-approved check contract and returns machine-readable evidence, findings and unverified gaps, so a green result has a defined scope. ### [Night Shift](https://github.com/NousResearch/hermes-agent/pull/151024) Automation · by MuhammadMurtuzaHussain · merged · 11-08 An on-call engineer that never sleeps. Alerts trigger a read-only investigation, a human approves the fix with one tap in Slack, and Night Shift runs it, proves recovery, and writes the runbook as a skill so the next incident is recognised. ### [990.tax](https://github.com/NousResearch/hermes-agent/pull/151022) Tools · by Digital Tax LLC · merged · 11-08 IRS public records on 1.9 million+ US tax-exempt organizations through 990.tax's free MCP server (Streamable HTTP at https://990.tax/mcp, no sign-in, no keys in config): search by name or EIN, exempt status and automatic revocations, missing filing years, Form 990 revenue, expenses, assets and officer pay, which annual return applies and when it's due, statistics and rankings by state and category, and IRS e-file reject codes. Bundles two skills: nonprofit-lookup and form-990-filing. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [FAL Video](https://github.com/NousResearch/hermes-agent/pull/151014) Tools · by axAilotl · merged · 11-08 Video edit, video extend and reference-to-video on FAL (MiniMax H3, FLUX 3, Kling 4.0, Wan 3.0, Seedance 2.5, Gemini Omni Flash and more), with a skill that chains them. ### [Audio API](https://github.com/NousResearch/hermes-agent/pull/151013) Voice · by axAilotl · merged · 11-08 OpenAI-compatible /v1/audio/transcriptions and /v1/audio/speech endpoints backed by Hermes' configured STT and TTS providers. ### [Icon Index](https://github.com/NousResearch/hermes-agent/pull/151009) Desktop · by minutechreview · merged · 11-08 40 free CC0 3D icons in one matching style for websites, apps, slides and videos. Browse them spinning in a Hermes Desktop pane, recolour them (8 presets or any colour) and download the result, and let the agent find an icon by meaning, return embed code (iframe, model-viewer, React, three.js) and download links, show it as a draggable 3D card in chat (::icon-index directive), or save its files into a project folder. ### [physis](https://github.com/NousResearch/hermes-agent/pull/152635) Tools · by ilPez00 · open · 11-08 Offline evidence-backed execution memory: per-repo graph context, a deterministic verify hook, and verified tool recipes that replay with zero provider calls and refuse when their tools, registry or captured files changed. Two bundled Linux engines and no API key or model download; approved host commands still run with host privileges and Python entry points need an interpreter, and untracked dependencies are not replay-checked. Operator-only approval and replay; the model cannot set executable paths or trigger raw commands. ### [Oathkeeper](https://github.com/NousResearch/hermes-agent/pull/151006) Tools · by Whydiriki700 · merged · 11-08 An owner's oath, enforced by the machine. The owner writes standing rules once in a plain oath file (standing missions briefed into every session, never delete or modify these surfaces, always require evidence such as a backup before touching those) and Oathkeeper enforces them deterministically on every tool call: never rules block with the violated oath line cited, always-require rules block until the prerequisite evidence exists in the same session, fingerprint snapshots of sworn surfaces detect changes made outside the gate and report them on the next turn, and the oath file itself can never be modified by the agent. Every verdict is written to an append-only local ledger. ### [Weft — edit-time coordination](https://github.com/NousResearch/hermes-agent/pull/150998) Tools · by celador · merged · 11-08 Catch conflicts between coding agents while they edit, before Git merge time. ### [Dramatis](https://github.com/NousResearch/hermes-agent/pull/150995) Tools · by kaizen-builds · merged · 11-08 Spoiler-safe reading companion: open a local EPUB, remember how far you have read, and ask who someone is or what happened. Answers come only from text at or before that position, with a short quote. Native Python plugin (plugin.yaml + \_\_init\_\_.py). ### [Pandy Tickets](https://github.com/NousResearch/hermes-agent/pull/150989) Tools · by Pandy Cloud · merged · 11-08 Find events in Guatemala and draft new events through Pandy Tickets' official MCP server (Streamable HTTP at https://tickets.pandy.cloud/mcp, no sign-in, no keys). Public event search, checkout links, and event creation tools. Agent Plugins v1 package (plugin.json + mcp.json + skills/). ### [Multiverse](https://github.com/NousResearch/hermes-agent/pull/150987) Tools · by vdimarco · merged · 11-08 Best-of-N for coding agents, with real checks. multiverse\_race makes one git worktree per attempt (2 to 6) from the same base (HEAD, or the uncommitted working tree through git stash create), starts one delegate\_task subagent per worktree with its own approach, pins each subagent to its worktree (pre\_tool\_call rewrites terminal workdir, file paths and V4A patch paths, and blocks access to the other worktrees and nested delegation), runs the same check commands in every worktree, ranks them by checks passed and then by the smallest real diff, and multiverse\_adopt applies the winner with git apply. With delegation.backend: multiverse, each attempt can use its own model and provider. Hermes Desktop: a live race board pane (each attempt's status and tool-call count, then checks passed, diff and the winner), a status-bar chip while racing, and a toast when judging ends. ### [Avatar](https://github.com/NousResearch/hermes-agent/pull/150982) Desktop · by vdimarco · merged · 11-08 Gives Hermes Desktop a face. Pick one of thirteen original characters, six 3D busts and seven 2D cartoons, each with its own voice. It watches your sessions, frowns when a step fails, and grins when a turn finishes. An optional overlay adds a plain-English task drawer, voice input, and one-click screen diagnosis. ### [Wever Labs](https://github.com/NousResearch/hermes-agent/pull/150972) Tools · by CodeWever · merged · 11-08 Check agent readiness, review oversight evidence, look up records, and run treasury checks with 226 tools in the Wever Labs main catalog and 30 in the free catalog. Paid tools cost 0.10 USDC per call on Base via x402. Hermes sends tool requests and inputs to the two hosted Wever Labs MCP servers. Both connections use Hermes approval for write-capable calls. ### [Grounded Recall](https://github.com/NousResearch/hermes-agent/pull/150965) Memory · by joaquinbyagentx · merged · 11-08 Before every model turn, searches your existing markdown notes (plain folders or an Obsidian vault), this profile's recent sessions and, opt-in, secret-manager item titles (never values), then injects a capped evidence block where each hit is labelled CURRENT, UNVERIFIED or HISTORICAL by freshness, plus a rule to search before answering "it doesn't exist / I have no access / I don't know". Every injection is logged per profile. No new memory store, works alongside any memory provider. ### [vibeCAD](https://github.com/NousResearch/hermes-agent/pull/152572) Tools · by am423 · merged · 11-08 Profile-driven FDM CAD: executes trusted build123d workspace models or prepares a local single-shell STL, checks printer fit, renders depth-tested full and detail views, and writes geometry handoffs. Bambu adapters cover X2D and P2S; native process projects, live AMS binding and print dispatch are not produced. Meshy is an optional separate plugin; its paid service calls are not made by vibeCAD. CAD libraries require a separately bootstrapped Python 3.12/3.13 environment. ### [Fish Audio](https://github.com/NousResearch/hermes-agent/pull/151965) Voice · by 100yenadmin · merged · 11-08 Your agent speaks and listens with Fish Audio: expressive voices in 80+ languages, voices you design from a sentence, and live word-by-word dictation. The official Fish Audio plugin: Fish's Drama 3 model with word-level direction, speech-to-text, Fish Audio's community voice library, voice cloning and voice design, with a Voices page in Hermes Desktop. Fish Audio becomes a Hermes speech and transcription provider for voice replies, read-aloud, voice notes and Desktop dictation, with sentence-by-sentence streaming in CLI voice mode and Desktop, and word-by-word live dictation captions on Hermes 0.21.6+ (stt.streaming: true). Adds the fish\_speak, fish\_voices, fish\_transcribe, fish\_settings, fish\_account and fish\_pronunciations tools, /fish and hermes fish commands (including /fish setup, a checklist that ends by offering a spoken sample with /fish preview), and four skills. ### [Meme Studio](https://github.com/NousResearch/hermes-agent/pull/150954) Tools · by MrHydeDev · merged · 11-08 Meme Studio through its MCP server (Streamable HTTP at https://memes.mrhyde.dev/mcp, no sign-in, no keys in config): finds the meme templates that fit a joke among more than a thousand, still and animated, with what goes in each caption slot, and makes the finished meme, an animated GIF for animated templates, plus a link to edit it in the browser. Works with the user's own photo too. Bundles one skill for the whole flow. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Business Agent OS](https://github.com/NousResearch/hermes-agent/pull/150952) Tools · by danielbilsborough · merged · 11-08 Set up a file-based business operating system for an AI agent, for business owners who are not developers: one business folder of plain Markdown memory in two layers (a folder per client and project with state, dated changelog with a monthly archive, to-do and inbox; and a Karpathy-style wiki the agent maintains from saved sources: topic, entity and source pages with an index and a log), a daily log, a voice guide, session commands (/pickup with client, topic and daily modes, /wrap, /biz-status, /biz-review, /trace, /new-client, /ingest, /lint), optional local git save points and an optional tmux session menu. One skill plus a hosted MCP server (Streamable HTTP at https://mcp.danielbilsborough.com/mcp, no key, no sign-in) with four read-only tools: scaffold\_business\_os, house\_rules, daily\_routine and voice\_guide\_template. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Shell Drop](https://github.com/NousResearch/hermes-agent/pull/150946) Tools · by shelldrop-sh · merged · 11-08 Set up a Mac or Linux machine so AI coding agents (Claude Code, Codex, any CLI agent) can be run and checked from an iPhone or iPad over SSH: SSH on, tmux sessions that survive a dropped connection, Tailscale for the route in, the phone's SSH key, then the exact connection details. One skill plus Shell Drop's hosted MCP server (Streamable HTTP at https://mcp.shelldrop.sh/mcp, no key, no sign-in) with four read-only tools: setup\_guide, tmux\_howto, generate\_tmux\_conf and troubleshoot\_connection. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Firecrawl](https://github.com/NousResearch/hermes-agent/pull/150944) Web · by firecrawl · merged · 11-08 Search, scrape, crawl, map, interact, monitor and research through Firecrawl's hosted MCP server at /v2/mcp-oauth, with Firecrawl skills (get-started, routing, deep research and 16 workflow skills). Adds the full Firecrawl tool set (Alexandria data providers, people and company search, developer, paper and US government search) beside the bundled Firecrawl web backend. The packaged connection uses FIRECRAWL\_API\_KEY, the same key as the bundled backend; Hermes-managed OAuth on the same endpoint works through a native MCP entry. ### [ShiftRepair](https://github.com/NousResearch/hermes-agent/pull/150943) Desktop · by wms2537 · merged · 11-08 Repair weekly staff schedules with minimum disruption, verified constraints and a shared Desktop board. ### [API Indonesia](https://github.com/NousResearch/hermes-agent/pull/150941) Tools · by khmuhtadin · merged · 11-08 Indonesian public data from official sources through API Indonesia's hosted MCP server (Streamable HTTP at https://mcp.apiindonesia.id/mcp, API key auth): administrative regions down to villages with Kemendagri codes, postal codes, region boundaries, public holidays and cuti bersama, prayer times, Hijri calendar, schools, universities, hospitals, BMKG earthquakes, weather and warnings, Bank Indonesia exchange rates, minimum wage (UMP), NIK/NPWP format checks, QRIS parsing, rupiah terbilang, halal and BPOM lookups, OJK investment lists, KBBI, Al-Quran and Alkitab. Bundles the api-indonesia skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Project Feed](https://github.com/NousResearch/hermes-agent/pull/150931) Tools · by Nyarenari · open · 11-07 Plan work, run tasks, and report project progress in Project Feed through its hosted Streamable HTTP MCP server, with four workflow skills: task-workflow, plan-project, project-status and work-triage. Agent Plugins v1 package (plugin.json, mcp.json, skills/), configuration only. Needs a Project Feed Pro workspace and OAuth sign-in: the README's Hermes section adds the server to mcp\_servers with auth: oauth, then hermes mcp login "Project Feed". ### [Vipernet Model Manager](https://github.com/NousResearch/hermes-agent/pull/150929) Models · by proldan · open · 11-07 Picks the provider and model for each turn from the profile's own catalog, based on the kind of work detected in the message: declared task associations first, then an automatic complexity analysis. Keeps the choice stable for the whole session, records every decision in an append-only ledger and degrades to the next route instead of failing when a route has no credential. ### [Signal Dog](https://github.com/NousResearch/hermes-agent/pull/150928) Tools · by scottleimroth · merged · 11-07 Listens through the microphone for household sounds (a door knock, an appliance beeping, a smoke alarm or other steady tone) and sends a chat message through the agent's messaging platform. A signal dog is a dog trained to alert a deaf person to sounds; this does the same job with a microphone, and it is built by a deaf developer. Off until the user turns it on with /signaldog on. Not a certified smoke alarm: it can miss sounds and raise false alerts, with measured rates and limits in the README. ### [Distro Reader](https://github.com/NousResearch/hermes-agent/pull/150926) Tools · by Distro-Media · merged · 11-07 Read independent news and content from Distro publications in Hermes through the public Distro Reader MCP server: search, feeds, full stories, archive research, visual article and book rendering, podcasts, payment information, and user-approved tips. Portable Agent Plugins v1 package with three workflow skills; public reading needs no account or API key. ### [Talos](https://github.com/NousResearch/hermes-agent/pull/150921) Tools · by oma035 · merged · 11-07 Automatic, unskippable smart-contract security checks for code Hermes writes and deploys (EVM, Solana/Anchor, Sui/Move): scans every .sol/.rs/.move file Hermes writes, feeds open findings back so Hermes fixes its own bugs, and gates deploy commands (anchor/solana/sui/forge/hardhat, including inside scripts) — mainnet blocked on high/critical findings, escalated to Hermes's human approval on medium, allowed with a warning on devnet/testnet; plus a /talos report command and a talos\_scan tool. ### [egress-guard](https://github.com/NousResearch/hermes-agent/pull/150913) Tools · by moussetea315 · merged · 11-07 Automated output vetting for AI agents. Policy-driven guard for what leaves the agent. A pre\_tool\_call hook redacts credentials in tool arguments and sends shell commands that upload data to non-allowlisted destinations (curl/wget uploads, scp/rsync, cloud copies, gist create, git push to a URL, python/node upload one-liners) to Hermes's approval prompt, showing a masked preview of the command (fail-closed when nobody can approve). transform\_tool\_result and transform\_llm\_output suppress small counts in tables, including complementary cells and percentages so hidden values cannot be recovered (statistical disclosure control). Presets, an optional strict network mode, and a /guard command (status, report, dry-run check). ### [ScrapeIO](https://github.com/NousResearch/hermes-agent/pull/150910) Web · by ScrapeIO · closed · 11-07 Public social media data through ScrapeIO's hosted MCP server (Streamable HTTP at https://scrapeio.com/mcp): profiles, posts, videos, comments, transcripts, search, trends and ad libraries from TikTok, Instagram, YouTube, X, Facebook, Reddit, LinkedIn and more, as one normalized JSON shape. 85 tools plus list\_endpoints and call\_endpoint for the long tail, and a bundled scrapeio skill for picking endpoints and paging. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Shared Memory Domains](https://github.com/NousResearch/hermes-agent/pull/150909) Tools · by DrFreeman001 · merged · 11-07 One memory several Hermes profiles share, split into domains with per-agent read/write grants: agents only propose facts, a human verifies them before they are recalled, revocation is immediate and sticky, and decisions are audited. ### [MentionAgent](https://github.com/NousResearch/hermes-agent/pull/150906) Tools · by MentionAgent · merged · 11-07 Run link building outreach through MentionAgent's hosted MCP server (Streamable HTTP at https://mentionagent.ai/mcp, OAuth sign-in, no keys in config): review the outreach drafts MentionAgent wrote, send the batch, answer publisher replies, close placements, and change campaigns in plain English, across every site in the user's account. Bundles the link-building-outreach workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Raise](https://github.com/NousResearch/hermes-agent/pull/150905) Tools · by Triolith · merged · 11-07 Fundraising desk for founders: imports your investor pipeline, ranks investors on cited evidence, checks every draft against code gates (one conversation per firm, a pass closes the firm, 3 touches per channel, working-day cool-downs) and lints it against your verified claims before it reaches your mailbox as a draft. Draft-only: Raise has no send tool, and a pre\_tool\_call hook refuses send, reply and forward calls to pipeline addresses on mail tools and terminal commands alike. ### [myradio](https://github.com/NousResearch/hermes-agent/pull/150902) Tools · by newjordan · merged · 11-07 Hermes DJs your own myradio station. /radio shows a now-playing card, plays a song by title and sends song requests to the station's music maker, all without a model turn. The agent gets 12 tools to read the station, search the library, plan the queue, write the DJ's lines and ask for new songs, plus the myradio:dj skill and the myradio:dj-shift blueprint for unattended DJ shifts. Going live and deleting audio are not agent tools. ### [Ferry](https://github.com/NousResearch/hermes-agent/pull/150900) Automation · by Thomas Lince · closed · 11-07 Store-and-forward job exchange between Hermes instances over a git repo. One instance sends a job; another picks it up whenever it next wakes, claims it (first push wins), does the work in its own session and replies with a verdict. No inbound ports, no daemon, no live peer needed, so it works with hosted scale-to-zero instances, laptops and boxes behind NAT; the git history is the audit trail, senders are verified against commit authors, and secret-shaped attachments are refused. ### [Google Workspace for Hermes](https://github.com/NousResearch/hermes-agent/pull/151346) Desktop · by francoleff · merged · 11-07 Gmail, Calendar, Drive, Docs, Sheets and Contacts in one Hermes workspace, with cross-app briefs and approval on every write. ### [CostGuard](https://github.com/NousResearch/hermes-agent/pull/150893) Tools · by Whydiriki700 · merged · 11-07 A brake pedal for model spend. Meters every API call per session and per day from the provider's own usage block; dollar figures are estimates computed only for models in its price table and are never invented for unknown models. When a provider reports no usage (common with streaming OpenAI-compatible endpoints), it falls back to request-size estimates and flags those calls as estimated in every total, never pricing them. Warns the agent at a configurable ratio of any configured budget and blocks tool calls once a session or daily token/cost budget is exceeded, stopping the tool loops where runaway spend happens; with no budget configured it only meters and reports. ### [SessionGuard](https://github.com/NousResearch/hermes-agent/pull/150892) Tools · by Whydiriki700 · merged · 11-07 Work survives the session. Keeps a local work-state checkpoint for every session, built from tool evidence: current task, files touched, recent tool calls and outcomes, failure count, and the agent's own progress notes saved through the session\_checkpoint tool. A checkpoint closes only when the agent marks its task done; otherwise the next session from the same origin (same platform and sender) is briefed from the most recent open checkpoint, so work interrupted by a dead process, a dropped gateway, or a compaction resumes from evidence instead of a lossy transcript. Subagent and cron sessions neither record checkpoints nor receive briefings. ### [SecretShield](https://github.com/NousResearch/hermes-agent/pull/150891) Tools · by Whydiriki700 · merged · 11-07 Credentials stop at the tool boundary. Scans every tool call's arguments before dispatch and every finished reply before delivery for credential material (vendor API keys, GitHub/AWS/Google/Slack/Stripe tokens, PEM private keys, JWTs, bearer credentials, passwords in assignments), blocks the call or redacts the value fail-closed, and appends every attempt to a local incident ledger that stores only metadata (kind, length, masked preview), never the secret. ### [9Router](https://github.com/NousResearch/hermes-agent/pull/150888) Models · by aligundogar · merged · 11-07 Dashboard tab for a 9Router gateway configured as the \`9router\` custom provider: groups the router's models by route prefix, shows the last test result for each model, the \`/models\` response time and, with the optional admin password, connection health and cooldown counts. A Test group button retests one group at a time. Routing and load balancing stay in 9Router. ### [Skill Court](https://github.com/NousResearch/hermes-agent/pull/150873) Tools · by saatvik333 · merged · 11-07 Puts every skill your agent writes on trial: replays real tasks with and without the skill in paired Docker sandboxes, scores each pair with the task's check or a blind pairwise judge, and gives a verdict per model. A skill that hurts is hidden from that model only (skill index, skills\_list, skill\_view), with the cause and the SKILL.md line behind it; an appeal drafts a fix, proves it on the same tasks and writes it only when you apply it (undo restores the original). Optional automatic trials after the agent creates or patches a skill. CLI (hermes court trial/report/appeal/auto/tasks/quarantine/release/list) plus a Desktop page. ### [herd](https://github.com/NousResearch/hermes-agent/pull/151037) Tools · by welliv · merged · 11-07 Keep one Hermes across every machine. Settings, skills, memory and SOUL.md follow the user between devices, while each machine keeps what belongs to it — Docker paths, ports, bot tokens. \`config.yaml\` is merged per key against a stored base, with machine-bound paths resolved locally before any comparison, so a pull cannot delete the section the machine is running on. API keys never sync: herd does not open \`.env\`. The store is encrypted with a key stretched from a recovery code before anything leaves the device; every pull leaves a restorable backup and \`hermes sync undo\` puts it back, removing files the pull added. Ships a per-device manifest store over any folder or git remote, three agent tools (\`sync\_status\`, \`sync\_diff\`, \`sync\_now\`), session hooks that pull at start and push at end in interactive sessions, and a \`hermes sync\` CLI. Pairing is one command that prints a 20-character code, valid once for 15 minutes, written as a sealed \`pairing.json\` in the store folder. A machine that can see that folder reads and deletes it. The file is not pushed to a git remote, because it wraps the recovery code and would remain in history after deletion; a machine that only has the git remote joins with \`hermes sync join \ --code HERD-…\`. ### [falsify](https://github.com/NousResearch/hermes-agent/pull/150860) Tools · by CodexOperator · open · 11-07 A claim is not done until it survives a check that could have failed. The agent pre-registers each claim BEFORE changing anything (its check and the base commit), hashed before any run; judging runs the check on the agent's final tree (must pass) and on copies with the agent's own change reverted, whole and hunk by hunk (must fail): a check that passes without the change reads VOID, and a proved claim says how many hunks of the change its check covers. The agent never writes the mutant. A pre\_verify nudge names every claim not done on a proved result before the turn finishes; it never blocks completion. ### [us-air-rules](https://github.com/NousResearch/hermes-agent/pull/150859) Tools · by entersidbellad · merged · 11-07 Does this EPA air rule apply to me? Screen a plant or a household against Title 40 and get applies / check with EHS / skip, with the citation. Compare two plant scenarios with the same bundled pack (changed inputs, changed verdicts, missing facts, unchanged count) without contacting the network. Read or search current eCFR Title 40 text, list recent EPA Federal Register documents for one part, and look up a state's EPA-approved SIP rules in 40 CFR part 52. ### [Secret Guard](https://github.com/NousResearch/hermes-agent/pull/150858) Tools · by ChasClogston · merged · 11-07 Enter API keys through a hidden dialog into the macOS keychain, never chat; block and scrub secrets before any model sees them. ### [Emotion Chip](https://github.com/NousResearch/hermes-agent/pull/151181) Voice · by 100yenadmin · merged · 11-07 The agent decides how to say it. Your agent writes its own delivery into the text it speaks, a small score in square brackets (a sigh, a whispered line, a pause, a warm and amused tone), and Emotion Chip renders exactly that for ElevenLabs v4 or Fish Audio (s2.1-pro, Drama 3), each in its own dialect, adding nothing the agent did not write. With voice replies on in a chat (/voice on), the chip tells the agent when Hermes will read its reply aloud, so the agent can say it itself, score included, instead of Hermes reading the plain text (not on Discord). Registers the emotion-chip speech provider, a delivery guide in the system prompt, the emotion-chip-voice skill, the /chip command (status says whether the chip can speak and what to fix; why, quiet, register) the hermes chip command, which shows offline what each engine receives for any score, and two agent tools: chip\_status (read-only) and chip\_preview (offline). CC0 emotion data and pre-registered evaluations ship in the repository. ### [Personal Assistant](https://github.com/NousResearch/hermes-agent/pull/152252) Automation · by 100yenadmin · merged · 11-07 A remote personal assistant for your agent: it does the errands up to the final page and stops at your yes. It interviews you once ("grill me": eight short sittings on your business, people, decision rights, preferences and calendar rules), remembers corrections, people, SOPs and trust levels in a local ledger, and never asks the same thing twice. Travel and errands on real sites in English, Japanese and Korean: restaurant tables (TableCheck, Tabelog, OpenTable, Catch Table), hotels and flights on Trip.com, taxis, food and konbini delivery, with sourced prices, an approval packet for each action and an honest handoff at a sign-in, OTP or app-only wall. Optional Chief of Staff mode adds a daily plan, a weekly closeout and a radar. Adds the pa and pa\_call tools, the /pa command and 42 skills (the Chief of Staff ones load only in that mode). ### [Mood](https://github.com/NousResearch/hermes-agent/pull/150846) General · by huy-sec · merged · 11-07 A tiny evolving mood for Hermes, built for slow local models. Tracks mood, rapport and three slow-moving traits (warmth, playfulness, directness) that change with keyword rules on what the user writes and drift back to neutral over hours, then appends one ~30-token line to each user message through pre\_llm\_call, so the system prompt and its cache are never touched. Adds a /mood command (show, reset, on, off). ### [Hermes Glasses](https://github.com/NousResearch/hermes-agent/pull/150842) Platform · by 100yenadmin · merged · 11-07 Your own Hermes agent on your smart glasses: it sees what you see, answers by voice, and taps you on the lens when something is due. A gateway platform that adds glasses as a channel over one WebSocket, with short spoken-style replies to the device that asked, camera photos (requested when you ask what you are looking at, or taken by the agent with its glasses\_capture\_photo tool), streamed replies for heads-up displays, approval and clarify prompts on every device, and cron and heartbeat pushes to all of a user's devices, queued in memory while they are offline. Speaks the bridge protocol of the community hermes-glasses iOS app (v0) and a richer v1; a simulator is included. ### [Kata](https://github.com/NousResearch/hermes-agent/pull/150840) Automation · by opeblow · merged · 11-07 Learns the multi-step tool routines your agent keeps repeating across conversations and, only after you review and approve each one, compiles it into a single deterministic tool that runs its steps without model round-trips in between. Proposals show the steps, data flow between them, permissions, input and output schemas, per-step side effects and a twelve-check offline validation (exact replay of recorded runs, held-out fit, injected failures and timeouts, secret scan, tool schema fingerprints). Approve, reject, disable, roll back and inspect run traces from /kata, \`hermes kata\` or a dashboard tab. ### [Snooze](https://github.com/NousResearch/hermes-agent/pull/153031) Automation · by AIverflow · open · 11-07 A reminder that repeats in your chat (Telegram, Discord) until you answer it exactly: Ok, your word, an emoji, a Telegram reaction, or one of several choices; reply later to postpone. A wrong answer gets the list of valid answers from the plugin's code, never from the LLM. Then it can hand a follow-up to the agent. Created with /snooze in plain words ("Stop the GPU at 19:00 every 10m then shut it down"), by the agent through the snooze\_create tool, or with hermes snooze add. Zero LLM calls per reminder. ### [Cacheguard](https://github.com/NousResearch/hermes-agent/pull/150832) Tools · by AIverflow · merged · 11-07 Warns or blocks a large chat message sent after a pause, when the provider's prompt cache may have expired. Afterwards it shows how many tokens were reused or rebilled, and /cacheguard shows how often your cache was lost after pauses of each length. Modes (off, warn, block) are set per surface (terminal, desktop, gateway), and a model can have its own pause, size and mode. In block mode the message is not sent: the reply says so, and you send it with /retry or by sending the same text again. If the plugin fails, the message is sent. ### [Phone Autopilot](https://github.com/NousResearch/hermes-agent/pull/150825) Automation · by jupist · merged · 11-07 Teach your phone a task once and replay it with no model in the loop: routines with checkpoints, repairs through the agent when an app changes, a daily automation blueprint and a live phone pane. Builds on phone-harness. ### [SaaSCity](https://github.com/NousResearch/hermes-agent/pull/150816) Tools · by SaaSCity · merged · 11-07 Plan a SaaS, AI or developer-tool launch through SaaSCity's MCP server (Streamable HTTP at https://saascity.io/api/mcp, no sign-in, no keys in config): audit a homepage the way directory scrapers see it, pick launch directories from 1,600+ with Ahrefs DR, link type and price, research products listed on SaaSCity, run SaaS metric calculators, and draft a SaaSCity listing for the user to review. Bundles one workflow skill, launch-plan. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [oMLX](https://github.com/NousResearch/hermes-agent/pull/150815) Models · by vitaliy-blprnt · merged · 11-07 Managed oMLX runtime and MLX model library for Hermes Desktop on Apple Silicon. ### [Telegram Operations Protocol](https://github.com/NousResearch/hermes-agent/pull/150814) Platform · by brukmars · closed · 11-07 Executive 3-line micro-cards with native expandable drawers, 1-Tap interactive mobile action gates, and forum topic operations for Hermes Agent on Telegram. Enforces the 3-line micro-card standard (Status Pill + #TAG + Metric, Core Result, Owner + Action + Deadline) with HTML sanitization, tucks heavy data tables and logs into native expandable drawers, and provides pattern-scoped callback query handlers for 1-Tap inline action buttons with Telegram user ID authorization. ### [ToneForge](https://github.com/NousResearch/hermes-agent/pull/150810) Tools · by efesn · merged · 11-07 AI guitar tone engineer for IK Multimedia AmpliTube 5 inside Hermes Agent: plain-language tone recipes (ordered chain, amp/cab families, mic placement, pedals, 7-band EQ, noise gate, levels) with a per-decision rationale, a versioned JSON preset library, a reference-track analyser that sets the numbers from a measured recording, and a MIDI bridge that drives only the CCs the user assigned. ### [manyhands](https://github.com/NousResearch/hermes-agent/pull/150807) Tools · by PhiBao · merged · 11-07 Identity, memory and permissions for every person who talks to one Hermes bot. A gateway is a single-operator agent: the platform already carries sender\_id and platform into every turn and nothing acts on it, so everybody in a group or a household shares the same memory, the same tools and the same approval authority. manyhands acts on it. Roles (owner, member, guest, blocked) are resolved from the platform event, never from message text, and applied as three states per tool: allow runs, ask parks the call under a four-character code and sends it to the owner's own DM, and everything else is refused with a message that names only the asker's own role. \`/many approve CODE\` opens a one-shot grant bound to that person and that tool, spent by the next call and expired on silence — only a human typing the slash command can approve, and nobody can approve their own ask. The operator's own MEMORY.md and USER.md blocks are withheld from everybody else's system prompt and replaced with a card about that person, because "everyone gets their own Hermes" has to include not handing a stranger your notes; each person also gets their own facts, their own token budget, and \`/many pause\`. Six tools (whoami, people, remember, recall, share, audit — \`people\` is read-only: a roster change is a human-typed verb, never a tool call) and \`/many\` (roster, invite, claim, add, role, tools, budget, audit, approve, deny, pause) whose owner-only verbs stay owner-only because a command's identity comes from the platform event too, so a stranger typing in the owner's own chat session is still a stranger. Identity fails closed: a session with no human turn on it — a restart, a subagent, a tool driven from the gateway socket — gets the guest allowlist, not the owner's, and a tool call in a session more than one person has spoken in is attributed by the task id the host dispatches with (or refused, never guessed from the last speaker). Every turn, tool decision, refusal, escalation, approval, spent grant and token count is an audit row you can read back. ### [AirPods Approvals](https://github.com/NousResearch/hermes-agent/pull/150803) Voice · by cineglobe · merged · 11-07 Hear Hermes approvals and questions in your AirPods and answer with your head: nod to approve, shake to deny. Gestures are recognised by rhythm, with a tap in each ear as you go, and drive the real approval through approval.respond. A nod only ever approves once, and a timeout or lost AirPods sends nothing. Desktop-only plugin with a small native helper built from source on your Mac, because head motion isn't reachable from a plugin: Install in the plugin's settings opens the bundled desktop/Install Helper.command in Terminal, which builds and starts it. Set up then walks through a Gemini key, Motion access and AirPods on the same page. ### [Quota Gauge](https://github.com/NousResearch/hermes-agent/pull/150801) Desktop · by cynricantao · closed · 11-07 Desktop pane with draining quota bars for the AI subscriptions this profile is signed in to (Anthropic, ChatGPT/Codex, Command Code, xAI, OpenCode Go, GitHub Copilot, Nous Portal, Z.ai, Kimi Code, MiniMax, Devin, Antigravity), plus Today / 7d / All-time token charts split into input, output, cache read and cache write, kept in one account-wide ledger. ### [Hypnos](https://github.com/NousResearch/hermes-agent/pull/150800) Models · by 0xNoramiya · merged · 11-07 Hermes learns in its sleep. Consolidates your corrections and denials into a small LoRA brain overnight, then routes mastered turns to it with the learned habits pruned from its prompt. ### [SolverForge Routing](https://github.com/NousResearch/hermes-agent/pull/150783) Tools · by blackopsrepl · merged · 11-07 Personal multi-stop driving trips: tell Hermes where you need to go, review the places it finds, and a SolverForge constraint solver orders the stops on OpenStreetMap roads. Change plans in the conversation (skip a stop, choose the first visit, add a place, finish elsewhere) and the trip is repaired, keeping visited stops, the next stop and before rules; an older solve never overwrites a newer edit, and results are labelled best-found, not proven optimal. Desktop: the road map beside the chat, a route card, a status-bar chip, Google Maps, Apple Maps and OpenStreetMap links, and one QR code at a time for the phone. ### [werracle-evm-guard](https://github.com/NousResearch/hermes-agent/pull/150778) Tools · by pCwOrM · merged · 11-07 Autonomous EVM execution guardrail and DeFi liquidation risk radar for Hermes Agent. Intercepts cast send transactions before broadcast to block unlimited token approvals, NFT operator takeovers, and high-value transfers. Performs Radar lookups to api.werracle.com by default on every werracle tool call and intercepted cast send. ### [munin](https://github.com/NousResearch/hermes-agent/pull/150768) Tools · by Munin Labs · merged · 11-07 Munin meeting notes for Hermes Agent: ask what was decided in your last meeting, get a follow-up with every action item checked against the transcript and cited by meeting id, line id and clock time, and search who said what across meetings, through the hosted Munin MCP server (Streamable HTTP, OAuth 2.1 with PKCE and DCR) with two workflow skills bundled (munin-followup, munin-recall). Portable Agent Plugins v1 package (mcp.json + skills/), no plugin code. ### [subscription-sheriff](https://github.com/NousResearch/hermes-agent/pull/150766) Tools · by HATWHY4041 · merged · 11-07 Track subscriptions and free trials in plain English, get warned before they charge, and see what you really spend. Quiet renewal reminders (only when something is newly due, plus a trial-about-to-turn-paid warning), monthly and yearly totals per currency, biggest spenders, overlapping services, a /sheriff command and two one-click scheduled tasks (renewal reminders, monthly spend review). Fully offline. ### [NaNLMind Hermes](https://github.com/NousResearch/hermes-agent/pull/150755) Tools · by shemshallah · open · 11-07 Link Hermes to a local NaNLMind NaN16 universe for conversation, mathematical-memory inspection, bounded native graph-reasoning evidence, explicit native training, NANNET identity and CLOCKED evidence plans. Registers 64 tools and a pre\_tool\_call human-approval hook. Four sensitive CLOCKED/candidate tools require Hermes human confirmation; model-supplied approved=true cannot bypass that gate. Native execution launches a locally installed nanlm-universal process with a minimal environment, not Hermes provider secrets. Model-chosen corpus paths are confined to the configured plugin corpora directory. Network calls use configured NANNET, CLOCKED and head-cortex endpoints. The plugin stores corpora, transcript/scratch journals, state artifacts, encrypted installation identity and paired tokens under state\_dir. Use a separate absolute state\_dir per Hermes profile; the default is not profile-isolated. Planning does not transmit CLOCKED source text; an approved execution sends the reviewed source text to the configured gateway. CLOCKED and native graph results remain evidence, not proven truth or automatic learning authority. No telemetry, self-update, arbitrary credential-file reads or vendor-password-manager integration. For protected credentials, use the optional \[auth\] cryptography backend: the pure-Python fallback has not been audited for constant-time operation. ### [Subnano](https://github.com/NousResearch/hermes-agent/pull/152076) Tools · by subnano-ai · merged · 11-07 Subnano's local MCP server for the pay-per-post marketplace. Search and read Posts, buy access in Nano (XNO) or USDC on Base, draft and publish Posts, answer comments and Requests, and receive payouts. Bundles seven workflow skills. ### [RestoreProof](https://github.com/NousResearch/hermes-agent/pull/150751) Tools · by Surajkumar5050 · merged · 11-07 Recovery drills for real application stacks: restores a declared database backup (PostgreSQL, MySQL, MariaDB, MongoDB or SQLite) and uploads archive into an isolated Docker network, boots the app, and proves whether customers' journeys come back (every database file reference, a real user flow, recovery time and RPO), then cleans up and writes a hashed evidence report. Deterministic engine; the model never decides a verdict. ### [grokbot-staff](https://github.com/NousResearch/hermes-agent/pull/150749) Automation · by leoresendedop-sys · merged · 11-07 A chief-of-staff and specialist team inside Hermes (chief, saas, design, vision-agents, computer-vision, x — or a team you configure), plus an optional fail-closed bridge that hands long-running work to your own Grok Bot. Local staff answers with your current Hermes model via ctx.llm. Slash commands /staff and /grokbot, a staff delegation backend for delegate\_task, a bundled skill, and a Desktop pane for the roster, pending handoffs and recent jobs. The bridge does not call any Grok Bot API (there is none); it POSTs a JSON handoff to a webhook URL you set and/or writes handoffs/\.json in a folder you set, then collects results/\.json\|.md\|.txt or a pasted answer. Not an importer and not Grok Bot calling Hermes — Hermes delegating to Grok Bot. ### [Visual Gate](https://github.com/NousResearch/hermes-agent/pull/150746) Tools · by salch-cred · merged · 11-07 Lets Hermes check the web UI it just built. Drives a real headless Chrome, diffs the page against a stored baseline and audits WCAG 2.2 AA on the same render, then returns one PASS or FAIL with changed regions, computed contrast ratios and exact CSS selectors. A page that failed to load is an error, never a pass, and text it cannot measure is reported for a person to check. ### [Sluice](https://github.com/NousResearch/hermes-agent/pull/150739) Tools · by orjinameh · merged · 11-07 An airlock for agent changes. A terminal backend that runs the agent's commands and file tools (read\_file, write\_file, patch, search\_files) in a staged copy of the workspace, so the real files are untouched until you review the diff hunk by hunk and apply only what you approve (/sluice diff, /sluice apply, /sluice discard, also as \`hermes sluice\`). Calls a copy cannot contain (git push, package installs, network writes, deploys, writes or deletes outside the workspace) are sent to Hermes's own approval prompt labelled NOT CONTAINED. Select it with \`hermes config set terminal.backend sluice\`. ### [Folio — every claim, a page](https://github.com/NousResearch/hermes-agent/pull/150732) Tools · by okyssta · open · 11-07 Hermes gets a library of the user's own documents. Folio indexes PDF, DOCX, PPTX, XLSX, HTML, Markdown and source files locally and answers with the file, the page and the exact sentence instead of a paraphrase: seven tools search the library, open a page, prove that a quotation really appears where the agent says it does, report what was added or revised since a date, and format citations. Documents are content-hashed, so a revised contract keeps its old text and a desktop page renders the real page with the matching sentence boxed. ### [prediction-market-scout](https://github.com/NousResearch/hermes-agent/pull/150718) Tools · by HATWHY4041 · merged · 11-07 Polymarket research for Hermes: search markets or paste a polymarket.com link for live odds, spread, liquidity and 7-day history; ranked 24h movers and attention signals (big moves, volume spikes, wide spreads, odd pricing, undecided markets about to close); a price-alert watchlist that fires once and re-arms; a /pm command; and two one-click scheduled tasks (daily brief, quiet alert check). Read-only, no API key. ### [hermes-guild](https://github.com/NousResearch/hermes-agent/pull/150708) General · by Blackleets · open · 11-07 GUILD — the open, cross-platform collaboration network for agent builders (Hermes × Muse × OpenClaw), local-first by design. A Hermes plugin plus a stdlib-only registry server speaking open protocol v1.6: publish a scoped need with its type (experience, asset, review, capacity, local presence — what an agent alone cannot give itself), get explainable matches (skills with aliases resolved and verified skills bonused, language, dimensional reputation, timezone, reciprocity — every point broken down, presence only breaks ties), connect only with the owner's approval, optionally under a collaboration contract (deliverable, done criteria, deadline — so 'on time' is derived, not opined), and receive deliveries strictly as proposals: Ed25519-signed by their author and end-to-end encrypted (GUILD-E2E-1: X25519 + HKDF-SHA256 + HMAC-CTR/EtM built from stdlib primitives; the registry stores ciphertext it cannot read, and an envelope whose fingerprint no longer matches its signed reference cannot be accepted). Verified skills are credited only for the skills a delivery declares it covers. Reputation is dimensional (useful / clear / on time) and portable: a registry-signed, independently verifiable export you carry between nodes. There is no central server to trust — you run your own registry (one command, local SQLite, metadata only, never code or project files); an official hosted registry is planned for v1.1. No payments: exchange and reputation. 118 offline tests plus a live 10-builder swarm test over real HTTP; exercised through the real plugin loader of the official Hermes v0.23.0 release. ### [Odds Radar](https://github.com/NousResearch/hermes-agent/pull/150707) Tools · by WV629 · merged · 11-07 Polymarket odds in Hermes: search active markets, live odds (CLOB midpoint) with 24h/7d moves, volume and a 7-day sparkline, a persistent per-profile watchlist with move, above and below rules (edge-triggered, 1-pt re-arm, one alert on resolution), and alerts that run as a no-agent cron job, so every check costs zero tokens. Agent tools, an /odds slash command and a hermes oddsradar CLI share one engine. Read-only: no keys, no wallet, no trading. ### [privacy-xray](https://github.com/NousResearch/hermes-agent/pull/150696) Tools · by MadeByBrain · merged · 11-07 Where does your data actually go? Records every outgoing model request, the main loop and side tasks such as title generation, compression and vision, as metadata only: destination host, vendor, vendor HQ country, the model vendor behind aggregators such as OpenRouter (read from the model id; the final processing vendor behind an aggregator is not visible), size and counts of detected personal data types (e-mail, IBAN, payment card, phone, IPv4, German Steuer-ID and USt-IdNr, own watch terms). /xray canary invents a fictitious person and /xray trace shows every request and local store that carried it. /xray check audits the config with a severity and the exact hermes config set fix per path; /xray fix \ --yes applies one; a session retention hint is advice only and is never applied. /xray fix --yes and /xray find run only on local surfaces (CLI, TUI, desktop, dashboard) unless allow\_sensitive\_on\_gateways is turned on. /xray find searches the local stores for a person, /xray report writes a self-contained HTML leak map (no scripts, no external assets) with a before and after view of two canaries; a dashboard tab shows the same page in a sandboxed frame. Not legal advice and not a compliance certificate. ### [omonpakari](https://github.com/NousResearch/hermes-agent/pull/150689) General · by tamamo510 · merged · 11-07 Considerate health context for Hermes Agent: barometric pressure (OpenWeather), a typed journal (symptoms, medications, supplements, food, notes), menstrual-cycle windows returned as date ranges, sourced absorption-interaction checks against logged times, the person's own life notes and alternatives, per-rhythm wave states (pressure, cycle, symptoms, sleep, nutrition) and long-term patterns - assembled by a pre\_llm\_call hook into a short quiet picture before each reply, so suggestions are weighed against the person's body and habits first. No reminders, no pestering; a quiet first word on the host's own scheduled turns is opt-in. ### [precedent-log](https://github.com/NousResearch/hermes-agent/pull/150678) Automation · by quotumorg · merged · 11-07 Saves a human's answer to a blocked Kanban question as a ruling for later workers. Shows rulings and vetoes in Desktop with confirmed retirement. ### [hermes-secret-gate](https://github.com/NousResearch/hermes-agent/pull/150675) Tools · by MSAbhishek22 · merged · 11-07 Stop agent-drafted secrets from leaving the machine — deterministic scan, fail-closed gate, receipts. ### [Dropstation](https://github.com/NousResearch/hermes-agent/pull/150667) Platform · by satiricalguru · merged · 11-07 Turn private local folders into Hermes inboxes with standing tasks. Drop a document or image to get a Markdown reply and a durable processing receipt. The adapter polls the configured owner-private spool and uses its filesystem permissions as the inbound authorization boundary. Its events skip messaging allowlists and pairing. Snapshots, reports and receipts are stored locally. The configured Hermes provider processes the source contents. ### [FindBroadband](https://github.com/NousResearch/hermes-agent/pull/150666) Tools · by FindBroadband · merged · 11-07 The broadband deals available at an exact UK address, priced on their true whole-contract cost, and whether a bill or renewal quote is too high, through FindBroadband's MCP server (Streamable HTTP at https://www.findbroadbanddeals.com/api/mcp/standard, no sign-in, no keys in config). Bundles one skill, uk-broadband-deals. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Internet Outage Checker by FindBroadband](https://github.com/NousResearch/hermes-agent/pull/150665) Tools · by FindBroadband · merged · 11-07 Check whether a UK broadband provider is down right now, where, what the provider has said and how the last 30 days have gone, through FindBroadband's MCP server (Streamable HTTP at https://www.findbroadbanddeals.com/api/mcp/outages, no sign-in, no keys in config). Bundles one skill, uk-broadband-outage-check. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Hermes Fast Chat](https://github.com/NousResearch/hermes-agent/pull/150663) Models · by iyoda · merged · 11-07 Launch a CLI or TUI session with an explicit model and Fast mode. ### [CALLIT](https://github.com/NousResearch/hermes-agent/pull/150659) Tools · by k58v2hbzkk-del · open · 11-07 Turn any group chat into a prediction league: friends make calls on crypto prices, football or anything, challenge each other to 1v1 duels, run closest-guess pools, and Hermes referees — it settles from live data (or a named judge / weighted group vote) and keeps season leaderboards. Duels staked in SOL/USDC/USDT lock both stakes in an on-chain Solana escrow program (immutable, devnet by default) that can only pay the two players; Hermes signs the result with a per-install referee key and both players are refunded if it never does. ### [Solana Token Facts](https://github.com/NousResearch/hermes-agent/pull/150658) Tools · by Maxedwolf · merged · 11-07 Verifiable on-chain facts about Solana tokens and their developers: mint/freeze authority, supply, Token-2022 risky extensions, whether name/symbol/image can still change, top holders and concentration, wallets that bought in the creation block, creation time, creator wallet and launchpad, who funded the dev wallet, a developer's past launches (graduated, dev sold out, minutes to sell out) and the newest pump.fun -\> PumpSwap graduations. Adds 8 tools, a /token fact-sheet command, a due-diligence skill and an opt-in automation blueprint that saves dev reports for new graduations. Facts only: no verdicts, no trading advice, no transactions, no wallet keys. ### [Mindhall](https://github.com/NousResearch/hermes-agent/pull/150655) Tools · by jai-srinivas · merged · 11-07 One shared mind for every AI you use. Work, dates and a signed change log that Hermes, ChatGPT, Claude and your other AIs all read and write. ### [Motion Index](https://github.com/NousResearch/hermes-agent/pull/150654) Desktop · by minutechreview · merged · 11-07 Browse 49 motion-graphics styles as live previews in a Hermes Desktop pane, and let the agent find a style, turn it into a ready-to-build brief for your topic (length and 9:16/16:9 filled in), or suggest a new style to the Motion Index library. ### [Hermes Ego](https://github.com/NousResearch/hermes-agent/pull/150653) Tools · by Shawn Yuan · merged · 11-07 Hermes Ego reads and saves pages in the user's ego(lite) browser through a loopback MCP relay (Streamable HTTP at http://127.0.0.1:8767/mcp). The Mac daemon and the relay are installed separately, the same split as local-mcp; this package does not start, embed, or update them. Agent Plugins v1 (plugin.json + mcp.json + skills/). ### [Argorant](https://github.com/NousResearch/hermes-agent/pull/150646) Tools · by Argorant · merged · 11-07 Argorant through its hosted MCP server (Streamable HTTP at https://mcp.argorant.com/mcp, OAuth sign-in with the user's Argorant account, no keys in config): find B2B leads by job, industry, company and place, with free counts and masked previews; reveal or export checked work emails; save lists; check email lists the user already has; write, test and launch cold email campaigns from the user's own mailboxes; and read, label and answer replies. Bundles the argorant workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Hermes Canvas](https://github.com/NousResearch/hermes-agent/pull/150644) Desktop · by 0xbyt4 · merged · 11-07 Your agent's plan, wired to its real work. The agent draws its plan as hand-drawn boxes in a Desktop pane next to the chat. Each box is bound to files and a check command: hooks mark a box doing when the agent writes one of its files, and done or failed when its check command exits. The user steers the running build on the canvas: answer a question box, link boxes, leave a note (it reaches the subagent that builds the box), remove a box (its next write is blocked), add a drift write to the plan, replay every revision. ### [Hermes Detox](https://github.com/NousResearch/hermes-agent/pull/150642) Tools · by kellycryptos · merged · 11-07 Tool output firewall. Saves redacted tool output locally (unencrypted; redaction is pattern-based and can miss secrets; no network), gives the model a compact report, and retrieves exact lines on demand. ### [recall-vault](https://github.com/NousResearch/hermes-agent/pull/150638) Tools · by tydavid5 · merged · 11-07 A policy vault and memory for agent money. The agent remembers its spending rules, per-transaction cap, daily cap, confirm threshold, and allowlist, and Recall Vault blocks over-cap payments, refuses unknown recipients, and escalates over-threshold payments to human approval before any tool runs. Dry-run only, no money moves. ### [skill-governance](https://github.com/NousResearch/hermes-agent/pull/150636) Tools · by moonlight-lupin · merged · 11-07 Consolidated skill-governance plugin behind one composed hook surface: on\_skill\_lifecycle, on\_session\_start, pre\_tool\_call, pre\_llm\_call. The manifest engine enforces UPSTREAM\_MANIFEST.md — arrivals need a manifest row, content changes in manifested skills re-flag until re-acknowledged against a content fingerprint via the preserved /skill-manifest-ack command. The advisor engine observes behavioral skill edits (never blocking) and injects one bounded reminder to run the contrastive-refinement probe cycle (bundled as skill-governance:contrastive-refinement), recording every delivered advisory to ~/.hermes/skill-probes/fire-ledger.jsonl. Both legacy state namespaces import by copy (destination-wins, sources never modified) and a conservative governance.records summary joins them (acknowledged never implies validated). A clearance Automation Blueprint (daily schedule, origin delivery) replaces both legacy schedulers and preserves both clearance procedures verbatim; the user adopts it from the blueprint gallery or /blueprint, and the plugin imports no private cron modules and writes no cron state. Stdlib only, no network calls, fail-soft everywhere (malformed state keeps hooks inert, never aborts the host turn). ### [safe-swap](https://github.com/NousResearch/hermes-agent/pull/150627) Tools · by itsolaph · merged · 11-07 Rewrites destructive shell commands into safe twins before they run instead of snapshotting afterwards: git push --force becomes --force-with-lease --force-if-includes, rm -r becomes a move into a quarantine folder (with a file and byte count), and git reset --hard, git checkout -- / git restore, git clean and git stash drop take a private snapshot first, with /swaps and /unswap to undo and a visible note whenever a destructive command was left unswapped. ### [Renor](https://github.com/NousResearch/hermes-agent/pull/150623) Memory · by renorai · merged · 11-07 Renor memory provider: one durable memory shared by Hermes, Claude Code, Codex and Cursor over MCP, so what you tell one agent the others already know. Recalls before each turn within a hard deadline (1.5 s by default; the turn goes on without it) and saves the durable statements you make (preferences, decisions, rules, "remember that ...") after it, through a bounded background queue with retries; also on compaction and at session end. Stored text never enters the prompt automatically: recall reports a count and the agent reads matches with renor\_recall, labelled untrusted data. Hosted (mcp.renor.cloud, connection key) or the open-source engine, offline. Standard library only. Maintained by Renor. ### [citation-guard](https://github.com/NousResearch/hermes-agent/pull/150613) Tools · by ChienNguyen23 · merged · 11-07 Catches fabricated, mismatched and retracted references. Checks each DOI / PMID / arXiv id and each full reference entry against doi.org, Crossref, PubMed and arXiv: ids that do not exist, ids that point to a different paper, wrong year or first author, works found in no registry, and retractions (Crossref notices incl. Retraction Watch). A transform\_llm\_output reply guard appends a warning footer when the agent's own answer cites something that does not check out (the reply text is never rewritten); a verify\_citations tool and a /citecheck command (.md .txt .bib .docx) give full reports. Deterministic, Python standard library only, no LLM calls, no API keys. ### [GetYouTubeTranscript](https://github.com/NousResearch/hermes-agent/pull/150603) Tools · by TubeAgentKit · merged · 11-07 YouTube transcripts (optionally with timestamps), video search, and channel and playlist browsing through the GetYouTubeTranscript remote MCP server (Streamable HTTP at https://getyoutubetranscript.com/api/mcp). Bundles four skills: fetch a transcript, search YouTube, summarize a video, and research a channel or playlist. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration and Markdown only. ### [Podcast Studio](https://github.com/NousResearch/hermes-agent/pull/150598) Voice · by suhas-svg · merged · 11-07 Turn any article, PDF, paper, notes or the day's news into a mastered two-host podcast episode: neural voices per host, speaker-aware pauses, intro chime, -16 LUFS loudness, embedded chapters and a timestamped transcript in one MP3. Includes an audio-overview skill and a daily news-briefing blueprint. Script text is sent to Microsoft's online Edge text-to-speech service via edge-tts (no account or key); episodes stay local. No telemetry. ### [Ariadne](https://github.com/NousResearch/hermes-agent/pull/150596) Tools · by miracle352 · merged · 11-07 Record a Hermes session once, then replay it with no model, no network and no tokens to see exactly what Hermes does differently: check whether updating Hermes is safe by replaying your own sessions (with your plugins loaded) on the new version, replay on a new build, compare two installs, bisect the Hermes commit that changed a session, fork a recorded session to a live model from any model call, export trajectories and preference pairs, and check Hermes's documented hook contracts on every run. ### [Which Copy](https://github.com/NousResearch/hermes-agent/pull/150595) Tools · by Hazem-Samak · merged · 11-07 Answers the question that silently decides everything else: which copy of a command is the Hermes process actually going to run? On a Mac the same tool can exist five times over — a Homebrew formula, an npm global install, a pyenv shim, a standalone download, and Hermes's own bundled copy — and the name resolves to whichever one PATH reaches first, with no warning when that is the wrong one. which\_copy resolves one name against the PATH of the running Hermes process and, separately, against the PATH your own login shell builds for a terminal (that shell is started from the system default PATH, not Hermes's, so the entries it reports are its own), and says so when the two disagree. It also states, in every report, that Hermes's terminal tool runs commands in a login shell whose PATH gains directories Hermes adds — so it never claims the process PATH is the PATH a terminal command gets. It lists every copy it can find, on PATH and in the usual off-PATH install locations, with the evidence for where each came from (a Homebrew Cellar directory names the formula, a node\_modules tree names the npm package including its scope, pyenv versions, uv tools, pipx venvs, nvm, Hermes's own directories, a plain folder, the macOS system), the version it reports, and whether it can even execute here — an Intel-only binary on Apple silicon is named as needing Rosetta to run. Then it gives a verdict in plain language: an older copy shadowing a newer one, a dead symlink earlier in PATH that the shell silently skips, Hermes preferring its own bundle to yours, several copies whose versions were measured (the same version, the winner being the newest, versions that cannot honestly be ordered, copies that reported no version, copies the sweep did not ask, or copies it did not run), or nothing wrong at all. which\_copy\_scan sweeps the usual AI tools — python, python3, node, npm, npx, bun, pnpm, uv, pip3, pipx, git, gh, brew, hermes, claude, codex, opencode, gemini, grok, aider, cursor-agent, rustc, go, ruby, java — in about a second, one line per tool, worst first. It diagnoses and stops: no installs, no PATH edits, no relinking, no "fix" that hides the problem. Versions are asked with an argv-only probe (three-second deadline, own process group, 8 KB output cap, cwd at the filesystem root, and a stripped environment so a shadowing binary never sees Hermes's API keys), and only for copies this machine put there: an absolute entry on Hermes's PATH, an absolute entry on the login shell's PATH, or one of the fixed install directories. Nothing named in a tool call is ever executed — a file passed as a path, a copy found only in a PATH string the call supplies, and a copy behind a relative or empty PATH entry are described with their version left unchecked — so a prompt-injected repository cannot get its own script run. For a call that supplies \`path\`, every version probe still runs with Hermes's own process PATH, never the raw supplied string. probe\_versions false skips every probe, and the login-shell check is separate, so pair it with check\_login\_shell false for a call that starts no process at all. When two version strings were never meant to be compared, it says it cannot order them instead of guessing. ### [Write Gate](https://github.com/NousResearch/hermes-agent/pull/150594) Tools · by 2vic1 · merged · 11-07 Deterministic pre-write content gate for Hermes Agent. A pre\_tool\_call hook parses the exact bytes about to be written as JSON, TOML or Python (stdlib parsers), rejects NUL bytes, oversized content and user-declared regex contracts, and short-circuits invalid writes with a precise line/column error returned to the model as a tool-call error, so the agent fixes the content in the same tool-calling loop instead of leaving a broken file behind. Also adds a write\_gate\_check preflight tool, a /writegate slash command and Settings -\> Plugins config. Patch/replacement fragments are never parsed as whole documents: NUL, size and regex contracts still run on the fragment, and the reconstructed post-edit file is parsed when it can be rebuilt. ### [ZaloDesk](https://github.com/NousResearch/hermes-agent/pull/150570) Tools · by ZaloDesk · merged · 11-07 Read, search and reply to Zalo chats, personal Zalo accounts and Zalo Official Accounts (OA), through the hosted ZaloDesk MCP: catch up on unread conversations, read and search messages, send replies, find customers by phone, labels and friend requests; and for the workspace owner, the team (Phân quyền): invite, change and remove staff, which Zalo accounts and chats each sees, their permissions, routing of new chats and calls, conversation assignment, which PC runs a company Zalo, and the AI credit each staff PC may spend. Everything runs within the user's ZaloDesk role (an employee sees only the chats assigned to them). Agent Plugins v1 package (plugin.json + mcp.json + one skill); OAuth uses pinned mcp-remote@0.1.49 over npx stdio, so Node.js 20.18.1+ and npm are required. ### [Hermes Roam](https://github.com/NousResearch/hermes-agent/pull/150563) Tools · by Victor Cruz · merged · 11-07 Encrypted, host-scoped sync of the Hermes Agent home between your own machines: an AES-256-GCM sealed bundle of the shared home (config.yaml minus host-scoped keys, SOUL.md, memories, skills) is pushed to storage you own — an S3 or Cloudflare R2 bucket, a private git repository, or a local directory — and pulled back on another machine through a plan-first, journaled, backed-up, roll-back-able install. Secrets (.env, auth.json, API keys) never sync; pulls apply only on explicit user action (the agent cannot apply by default); conflicts are detected with a signed compare-and-swap head, never merged. Every pushed version stays recoverable (hermes roam history / restore). Works across profiles and bots: one store carries every profile (each profile's identity, SOUL, memories, skills and shared config), from any profile context; a profile can be excluded per device. Opt-in, off by default: a read-only, encrypted archive of session transcripts (exported with hermes sessions export --redact) that other devices can list and read, and cron job definitions (pulled jobs land paused until hermes cron resume). ### [BotSee](https://github.com/NousResearch/hermes-agent/pull/150560) Tools · by sidekickjanet · merged · 11-07 AI-visibility research through BotSee's hosted Streamable HTTP MCP server (https://botsee.io/mcp), with a bounded workflow skill for reviewing existing analyses, competitor recommendations and cited sources, and planning audits. Portable Agent Plugins v1 package; no local server, OAuth bridge, hooks or self-updater. ### [HUI+ Cards](https://github.com/NousResearch/hermes-agent/pull/150556) Desktop · by robbyczgw-cla · merged · 11-07 Interactive answer cards in Hermes Desktop and the Webapp; your answer goes back as the next message. ### [Ghost Call](https://github.com/NousResearch/hermes-agent/pull/150554) Voice · by winchxyz · merged · 11-07 A real-time AI assistant on your own phone calls. You stay on the line; Hermes listens and streams one-line hints to your page and chat (what they just offered, what to ask next), knows your private target without ever saying it, plays a hint in your earpiece on one tap, and posts a summary with what was agreed and a draft confirmation message when you hang up. Works with two browser links, a real phone line through your own Twilio account, or any call on speaker. ### [Vibe Relay](https://github.com/NousResearch/hermes-agent/pull/150547) Automation · by trenthalden · merged · 11-07 Vibecoding without burning your best model. Sets Hermes \`delegation.backend\` to a quota-aware strategy: when remaining usage is above a threshold (default 51%) the parent model keeps the work (\`self\`); otherwise each \`delegate\_task\` is routed to a configured cheap executor (provider+model or profile). When the executor lane is selected, a \`pre\_tool\_call\` hook blocks the parent from \`patch\`/\`write\_file\` and obvious terminal writes (\`sed -i\`, \`\>\`/\`\>\>\`, \`tee\`) so coding work must go through \`delegate\_task\` (children never blocked; \`enforce\_executor\_lane\` default true). At \`stop\_used\_percent\` (default 90% used) the same hook blocks parent \`delegate\_task\` and writers, auto-parks once per reset window (handoff + one one-shot cron), and injects \`parked until …; do not start new work\` via \`pre\_llm\_call\`; the block lifts after reset or when used drops below stop. Every task must carry \`allowed\` path globs and a \`gate\` command — exit 0 alone is not done. A \`post\_tool\_call\` hook captures gate PASS output bound to a working-tree hash; \`pre\_verify\` continues the turn until a fresh PASS exists; \`vibe\_accept\` checks changed paths ⊆ allowed. Manual \`/vibe park\` and blueprint \`resume-after-reset\` remain. Ships an Agent Workshop of 14 portable skills (build/debug loop, architecture audit, harness+evals, context budget, memory/retrieval, known-errors registry, frozen-tree review, product-aware releases, subagent-driven development, companion maintenance, lean executor orchestration, field-first acceptance, TDD, systematic debugging) registered as \`vibe-relay:\\` via \`skill\_view\`. Vendors DEV Framework tools (\`df\_init\`, \`df\_check\`, \`df\_nav\`); \`df\_check\` returns the gate command for the agent to run via \`terminal\` (approvals apply). ### [hindsight-curator](https://github.com/NousResearch/hermes-agent/pull/150546) Tools · by trenthalden · merged · 11-07 Bulk-curate stale Hindsight memories from chat via /memory-curate or tools, or from the Desktop Memory curation page: scan by topic (keyword/regex), review a dry-run plan, invalidate a selected subset with a reason, write one checkpoint fact, and restore via a local SQLite journal. Separate from the official hindsight memory-provider plugin — talks only to the public HTTP API. ### [word-miner](https://github.com/NousResearch/hermes-agent/pull/150545) Tools · by trenthalden · merged · 11-07 Turn any text you are consuming (lyrics, subtitles, articles, chat) into a personal vocabulary deck and run short daily spaced-repetition sessions inside Hermes. Any language pair. Local only — SQLite event log, SM-2-like scheduler with response-time adjustment, 3+3+3+1 sessions, two-stage graduation. No Anki, no AnkiConnect, no network. The model extracts lemmas; four tools plus /deck and a bundled mining skill. ### [NetEase Cloud Music](https://github.com/NousResearch/hermes-agent/pull/150544) Desktop · by BoosterCore · merged · 11-07 NetEase Cloud Music (网易云音乐) mini player in the Hermes Desktop status bar, in the spirit of the built-in Radio plugin: a compact themed popover with Now Playing (cover + scrolling synced lyrics with the current line highlighted), search for songs and playlists, favorites (heart / Liked Music), browsing and in-panel playback of your own playlists, play/pause/skip/progress/volume with outline-style controls, loop modes (list loop, single loop, shuffle) and a mini control cluster on the status bar itself. Desktop-only UI (desktop/plugin.js on the plugin SDK surface) plus a local FastAPI backend (dashboard/plugin\_api.py) that implements NetEase's weapi/eapi request signing (AES-CBC/ECB + RSA) with the Python standard library only — no build step, no extra Python dependencies. ### [SpecStream Search and Extract](https://github.com/NousResearch/hermes-agent/pull/150539) Web · by Specstreamai · merged · 11-07 SpecStream Search API and Extract API for web\_search and web\_extract. Search is $5 per 1,000 searches, extract is $0.80 per 1,000 pages (web pages, PDF, Word and Excel, up to 20 URLs a call, a page with no text is free). Needs a SpecStream API key. ### [delegation-receipts](https://github.com/NousResearch/hermes-agent/pull/150516) Tools · by dacheah · merged · 11-07 Verify what a subagent says it did before the parent reports the work as done. A "receipts" delegation backend asks every child to end its reply with a receipt (status, the files it changed, and the one command that proves the work); when a child stops, the plugin resolves and hashes each claimed path, matches every claim against the write targets that child's own tool history records so it cannot take credit for a file someone else wrote, and checks the cited command against a closed set of fixed check forms (pytest, uv run --no-sync pytest, cargo/go/npm/pnpm/yarn test), refusing a path rather than a bare name, a launcher, or inline evaluation. Host re-run is opt-in and off by default, because a hook-driven re-run bypasses Hermes' dangerous-command approval and the child's terminal backend; when an operator enables it, the command is re-run argv-style with no shell, in its own process group, with output and time bounded. Either way a verdict reaches the parent: a rejection is appended to the delegate\_task result, with the failing output attached when there is one; a background child's verdict arrives on the parent's next turn instead. Every check stays in a local JSONL ledger shown by \`hermes receipts\`. One command per receipt, one receipt per child. ### [Zotero Research](https://github.com/NousResearch/hermes-agent/pull/150511) Tools · by suhas-svg · merged · 11-07 Research with your own Zotero library: search it, read your PDF highlights and full text, cite with Zotero's citation engine (any CSL style, BibTeX/RIS export), add papers by DOI/arXiv with PDFs, and check a draft's citations against the library. Reads Zotero desktop's local API on loopback; zotero\_add alone writes to the library and calls Crossref/arXiv. No telemetry, no credentials. ### [Accessible Session List](https://github.com/NousResearch/hermes-agent/pull/150510) Desktop · by denis-skripnik · merged · 11-07 Keyboard-first, screen-reader-friendly native session picker for Hermes Desktop. Search titles, select with arrows without navigating, open with Enter or Shift+Enter, and use a keyboard-accessible actions menu. Filters background runs and empty drafts; sorts the fetched sessions by activity. English and Russian UI. ### [BuildPilot](https://github.com/NousResearch/hermes-agent/pull/150504) Tools · by Justnevo · merged · 11-07 Project tracking and progress recovery for Hermes Agent. Create projects, manage tasks, save progress checkpoints, and retrieve project context across sessions using local SQLite storage. ### [hermes-windows-certificates](https://github.com/NousResearch/hermes-agent/pull/150503) Tools · by Yicheng-Mark · merged · 11-07 Answer "which certificate on this Windows machine expired, and what am I actually trusting": walks the user's and the machine's certificate stores (MY, Root, CA) in-process through crypt32 - no certutil, no PowerShell - and lists every certificate with subject, issuer, validity window, self-signed or not, SHA-1 thumbprint and the store it lives in; the expiry arrives in words ("expired 45 days ago", "expires in 12 days") and the list is sorted with the expired and soon-expiring first. One certificate in full adds the serial number, certificate version, the signature algorithm in words, the public key's algorithm, named curve and size in bits (2048-bit RSA, P-256), and the purposes it is valid for (server authentication, code signing...; no EKU extension is reported as "does not restrict its purposes", which is what its absence means). Find it by thumbprint or by a subject substring, across the user's and the machine's copies. Use it for "why does TLS keep failing on this box", "is the root store healthy", and "which personal certificate ran out". Read-only: stores are opened read-only and only enumerated - nothing is added, removed or changed; no network, no file writes, no elevation, no child processes; 121 certificates across both contexts and all three stores walk in about 8 ms on the writing machine. Windows only. ### [hermes-windows-disks](https://github.com/NousResearch/hermes-agent/pull/150501) Tools · by Yicheng-Mark · merged · 11-07 Answer "are this machine's disks healthy, and which volume is filling up" on Windows: one table of every physical disk (model, serial, SSD/HDD/SCM, bus - nvme, sata, usb -, size, the storage stack's health word, GPT or MBR, boot/system flags, partition count) and every volume (drive letter, label, file system, capacity, free space with its percentage and a verdict, fixed/removable/network/cd-rom, the disk a volume lives on, the UNC path behind a mapped drive); volumes without a drive letter (EFI, recovery) are hidden unless asked for. win\_disk\_detail reads one disk or volume - by drive letter, label, model, serial or disk number - with the health in words (warning, unhealthy, and "unknown", which the storage stack reports on a perfectly healthy NVMe while OperationalStatus says ok - both are shown), the partition layout decoded from its GUID type (efi system, microsoft reserved, basic data, windows recovery), each partition's size, offset and letter, the largest contiguous free extent, and a plain-word free-space verdict (under 5 percent critical, under 10 percent low). Use it for "why is the disk slow", "will this drive fail", "where did the space go". Read-only: the storage classes are queried in one PowerShell CIM session, nothing is defragmented, formatted, resized, checked or repaired; no writes, no elevation, no network calls of its own. Enum values arrive as numbers and are decoded in the plugin - a Chinese-language Windows 11 was verified never to reach the answer; the whole table answers in about two thirds of a second. The plugin declares no Python dependencies. Windows only. ### [hermes-windows-firewall](https://github.com/NousResearch/hermes-agent/pull/150500) Tools · by Yicheng-Mark · merged · 11-07 Answer "what does the Windows firewall let through, and for whom" on Windows: every rule from the registry store the firewall service itself keeps, each row in plain words -- allow or block, inbound or outbound, TCP/UDP/ICMP or a protocol number, local and remote ports, the program or service the rule is scoped to, which profile it belongs to (Domain, Private, Public or any) and whether it is switched on -- plus the three profiles' firewall switches and default actions, with values the registry did not write reported as null and attributed to Windows' builtin defaults rather than invented. Built-in rules name themselves indirectly (@FirewallAPI.dll,-32765); the words are resolved from the resource in the machine's own language, and references that will not resolve stay as written. Details one rule in full: addresses with scope words (the local subnet, any address), edge traversal, the embedding group, the description, and fields the parser does not model (TTK2\_22 is on real machines) passed through verbatim rather than dropped. netsh labels localize with the display language, the registry does not -- which is why the plugin goes there: no netsh, no PowerShell, no WMI. Use it for "is port X open", "what may this exe do", and "is the firewall actually on". Read-only: registry values are enumerated and resource strings resolved, nothing is enabled, disabled, added or removed; no network calls of its own, no file writes, no elevation. The whole store answers in hundredths of a second (795 rules in 0.07 s on the writing machine); the plugin declares no Python dependencies. Windows only. ### [Thoth](https://github.com/NousResearch/hermes-agent/pull/150497) Tools · by cayden970207 · merged · 11-07 Agents working as one team, each with its own harness: Claude Code, Codex and Hermes today. Ask Hermes for a website, an app, artwork or a doc: it plans the work on a shared board and hands each task to the real Claude Code or Codex CLI, on your own logins and plans, so every improvement their makers ship reaches your team. Each agent works in its own git worktree with a brief of what the team decided and made (goal, decisions, handoffs, what its dependencies produced), reviews the other's work, and Hermes merges what passes once you say yes. In Hermes Desktop, Thoth sits beside each project's chat: a side window with the live canvas and a Team & brain tab (who did what, decisions, reviews, versions), and every Thoth project is also a Hermes project on the same folder. Claude Code or Codex can join the board through the bundled MCP server or take over as coordinator, and a task that needs the web gets a clean Chrome, or yours when it needs your logins. When a Claude or ChatGPT plan runs out, the other agent carries on from its work, and the side window and Team card show how much of each plan is used and when it resets. Approve a plan once and autopilot runs it to the end: it starts, cross-reviews and merges each task, and asks you in the chat only when something needs you; any change to the plan pauses it. Once work is merged, the side window offers it to open, download as a zip or find in Finder. Start a project from the project menu or the chat: a new one gets its own folder in ~/Thoth, and a folder you already have becomes a project that keeps every version. After an update, the Thoth page says what's new, once. ### [Canonry](https://github.com/NousResearch/hermes-agent/pull/150493) Tools · by Canonry · merged · 11-07 Open-source, agent-first Answer Engine Optimization (AEO) platform. Measure brand mentions and source citations across ChatGPT, Claude, Gemini, Perplexity, and Meta; research competitors and SEO opportunities; audit AI readiness; analyze AI crawler and referral traffic; create and publish content; manage ChatGPT Ads; and get pages indexed. Connect Search Console, GA4, WordPress, and more, then measure what changed across brands, locations, and markets. Includes Canonry operator and Aero analyst skills with MCP tools. Requires a configured Canonry runtime. ### [hermes-windows-network](https://github.com/NousResearch/hermes-agent/pull/150490) Tools · by Yicheng-Mark · merged · 11-07 Answer "is this machine's network healthy, and which adapter is the problem" on Windows: lists every visible network adapter with its state (up, disconnected, disabled, not present), its link speed as a number, its MAC, whether it is virtual (Hyper-V, VPN, virtual switch) or physical, and its driver version and date; adapters that are only a ghost of an old dock or USB device are hidden unless asked for. Details one adapter's IPv4/IPv6 addresses with prefixes and origins (DHCP, manual, self-assigned 169.254.x.x), default gateways, DNS servers, and a plain-word diagnosis - link-local means DHCP never answered, an address without a gateway cannot leave its subnet, no DNS means names will not resolve; a disconnected adapter with a stale static DNS says all three on the writing machine. Use it for "why is there no internet", for "which of these adapters is actually in use", and for reading what DHCP left on an interface. Read-only: the NetAdapter and NetTCPIP cmdlets are queried, nothing is enabled, renewed, reconfigured or released; no network calls of its own, no file writes, no elevation. Adapter names are localized data and are passed through as-is; states travel as enum names, which a Chinese-language Windows 11 was verified not to translate. The whole table answers in under a second and one adapter's detail in about two and a half, each a single PowerShell session; the plugin declares no Python dependencies. Windows only. ### [blastradius](https://github.com/NousResearch/hermes-agent/pull/150476) Other · by valenciayuna · merged · 11-07 Reads a destructive shell command before it runs and reports what it would actually delete: how many files, how many git can restore, how many it cannot, the total size, and which of them look valuable, then lets policy decide whether to show the report, block the call, or demand a one-shot blast\_allow. Covers rm and rm -rf, git clean, git reset --hard, truncate -s 0, \`\> file\` and \`: \>\` truncating redirects, and mv overwrite, including nested bash -c, subshells, and chained commands with cd tracking; anything it cannot resolve is reported as unmeasured instead of guessed. Snapshots and verified restore exist as a separate step (blast\_snapshot, blast\_restore); restore skips files edited after the snapshot unless overwrite\_conflicts is set. ### [house-style](https://github.com/NousResearch/hermes-agent/pull/150474) Tools · by Sahil-SS9 · open · 11-07 Learns a project's house style from how you correct the agent's edits. Every write\_file and patch records the block it introduced (including every hunk of a multi-file V4A patch and every occurrence of a replace\_all edit); later the plugin checks whether you kept it, rewrote it or took it out again, and /house-style learn distils a bounded batch of those before/after pairs into short numbered rules, which /house-style learn confirm writes into the project context file Hermes already reads — .hermes.md/HERMES.md, else AGENTS.md, else CLAUDE.md, else .cursorrules, else a new AGENTS.md — inside a marked block you own, so the rules reach the model through normal context loading and Claude Code reads the same file. /house-style status shows the capture state, the context file in use, counts and the newest edited pairs; scan re-checks the recorded edits; reset clears signals and learned rules while keeping anything you wrote; off/on stop or resume recording for one project; export prints the rules. ### [Session Share](https://github.com/NousResearch/hermes-agent/pull/150473) Tools · by Sahil-SS9 · open · 11-07 Share one or several sessions as a redacted, revocable secret GitHub gist with /share, and import a shared bundle back with /share import. A bundle is up to three kinds of file in one gist, every one of them rendered by the public \`hermes sessions export\` CLI: the importable jsonl, a readable Markdown transcript, and a standalone HTML page per session. The preview lists every session with its redaction counts and file sizes; nothing is uploaded until /share confirm. Credentials become non-reusable sentinels (never a head/tail mask), home paths become ~, and the account and host names are replaced; redaction cannot be switched off. /share import reads a gist or file, detects the format from content (Hermes export, Claude Code transcript, Codex rollout), shows what is inside, and imports only on /share import confirm. ### [underwriting](https://github.com/NousResearch/hermes-agent/pull/150466) Tools · by johnnyminer · merged · 11-07 Commercial real estate underwriting that shows its work: rebuilds net operating income from the rent roll and T-12 instead of trusting the broker's claim, runs buy-box tests against a DSCR-sized loan, returns the counter-price at which a rejected deal actually clears, sweeps price to find where a deal flips, models a hold period, and returns null rather than inventing a number the document never stated. Seven deterministic tools plus an /underwrite command and a skill. Extraction reads a document the user names (local path or URL) and is audited back against the document text: every figure is matched to the line it came from and marked verified, and a citation the document cannot support is reported. The arithmetic is pure Python standard library with no network calls and no third-party packages; only extraction calls the host model, and only a URL passthrough fetches anything over the network. Writes are confined to the plugin's own data directory: memos and extraction audit trails take a bare filename under \/exports/, and the deal book is JSON under \. Document input is treated as untrusted, because it is: local reads go through Hermes' read block list with no $HOME fallback, and URL input is limited to public http(s) with every redirect hop re-checked. Not investment advice. ### [Aletheia](https://github.com/NousResearch/hermes-agent/pull/150464) Memory · by ousiaresearch · merged · 11-07 Evidence-bound scoped memory for Hermes agents, over the Weft substrate: turns are queued for a local SQLite store under the active profile by default, recall is scoped before it is ranked (deny-by-default, deterministic, never widened by anything read), and every recall returns a receipt naming the scope consulted, the ids selected and the tier. Duplicates are not re-ingested, deletions cascade, corrections preserve lineage. Registers as memory.provider: aletheia and exposes weft\_recall, weft\_ingest, weft\_forget and weft\_receipt, with recall on a worker thread so a turn's prefetch is not blocked; the built-in memory store keeps running alongside it. ### [Lazy Plugin](https://github.com/NousResearch/hermes-agent/pull/150456) Tools · by RegeneratusLabs · merged · 11-07 A dashboard page in Hermes Desktop where you describe a plugin and watch it get built. Say what you want — or start from a skill the agent wrote, or a job that runs on a schedule — and your own agent drafts the plan in a chat turn you can read, the code comes from four fixed templates, and the page shows a receipt: plan, files, validator result, and a real try-out of every tool (purrs, needs a key, blocked upstream, held back, half-built, broken — with the exact failing line). The build never runs what it wrote: the try-out is your click, a key only travels to a public https host, and a file tool stays inside its own data dir. Nothing installs itself, and publishing is a SHA-pinned catalog PR you send. Also callable as lazyplugin\_create from any conversation. ### [Salesforce Guard](https://github.com/NousResearch/hermes-agent/pull/150455) Tools · by sforceai · merged · 11-07 Salesforce release-impact and scoped-access review: explicit-scope snapshot comparison, source-reference maps, local access evidence and HTML reports. Bounded security heuristics and partial OWASP mappings produce review signals, not org-wide assurance. Dedicated review profile blocks unrelated tools while the plugin is enabled. Custom native inputs require separate operator folder approval with model-sharing acknowledgement. ### [Workgraph](https://github.com/NousResearch/hermes-agent/pull/150453) Tools · by aliiqbal24 · open · 11-07 Visual product-state tree with actual galleries, SQLite development snapshots, protected behavior replays, feature borrowing, and persistent agent continuation. Reads/writes local repositories and runs approved local commands. ### [Leonardo.AI Video](https://github.com/NousResearch/hermes-agent/pull/150447) Tools · by RichardChua1234 · merged · 11-07 Leonardo.AI video generation backend for video\_generate: Motion 2.0 text-to-video and image-to-video through the Leonardo REST API, using only the Python standard library. ### [Global Business Culture](https://github.com/NousResearch/hermes-agent/pull/150446) Tools · by pswong661-eng · merged · 11-07 Cross-cultural business intelligence: country culture briefs and home-vs-host gap analysis computed from bundled Meyer, Hofstede, GLOBE and Lewis data. Three deterministic tools (culture\_brief, culture\_gap\_analysis, culture\_countries), a /culture command and a bundled practitioner-playbook skill. 57 economies offline, no dependencies, no network, no env vars — the gap analysis ranks the exact scale distances between two cultures, names the misread each gap produces and prescribes adaptation moves. Deep ASEAN/frontier coverage (Laos, Cambodia, Myanmar flagged as extrapolated from GLOBE-cluster neighbours). ### [BeLive AI Hub Media](https://github.com/NousResearch/hermes-agent/pull/150440) Tools · by belivetech · merged · 11-07 BeLive AI Hub backends for Hermes' image\_generate, video\_generate, text\_to\_speech and speech-to-text (backend \`belive-aihub\` for image\_gen.provider, video\_gen.provider, tts.provider and stt.provider), with models from the live AI Hub catalogues. ### [Wally](https://github.com/NousResearch/hermes-agent/pull/150438) Models · by RunanywhereAI · merged · 11-07 Wally by RunAnywhere as a Hermes model provider: open-weight models (DeepSeek, GLM, Kimi, MiniMax) on an OpenAI-compatible API, each with its context window, tool support and image input declared, and reasoning effort sent as Wally's top-level reasoning\_effort, clamped to the levels each model accepts. Pick Wally in \`hermes model\` or use \`--provider wally\`. ### [Yandex Mail](https://github.com/NousResearch/hermes-agent/pull/150433) Tools · by akinfold · merged · 11-07 Search and read Yandex Mail, organize messages, save attachments, and send replies. Read-only access is configurable; saving attachments and sending require explicit opt-in. ### [plan-tracker](https://github.com/NousResearch/hermes-agent/pull/150429) Desktop · by Hjuanma · open · 11-07 Track the markdown plans you already write: milestone progress bars from the checkboxes in your .md files, in a dockable pane plus a status-bar chip, with per-milestone effort (time, tokens, cost, models, subagents), ETA, history chart, per-session and global scopes, hash-guarded task ticking, blocked milestones, optional per-milestone weights and markdown export. ### [Daily Edition](https://github.com/NousResearch/hermes-agent/pull/150422) Desktop · by ousiaresearch · merged · 11-07 A personal newspaper inside Hermes Desktop: name interests, not subscriptions, and it compiles one dated front page — each story once, cross-posted stories merged with every publisher credited, consensus winning the lead, reprints withheld — with seven back issues, in-place reading, unread and saved state, and an optional editorial voice for headlines (off until you name a model endpoint; there is no vendor default). The same edition prints as a one-page broadsheet through a local CUPS queue, and the daily\_edition tool or /edition can print it from chat or cron. ### [OpenSEO](https://github.com/NousResearch/hermes-agent/pull/150384) Tools · by OpenSEO · closed · 11-07 Research keywords and competitors, audit sites, analyze backlinks, track Google and Maps rankings and AI visibility, and work with Search Console and Analytics data through the hosted OpenSEO MCP. Agent Plugins v1 package (plugin.json + mcp.json) with twelve SEO workflow skills; OAuth uses pinned mcp-remote@0.1.49 over npx stdio, so Node.js 20.18.1+ and npm are required. ### [gateway-guardian](https://github.com/NousResearch/hermes-agent/pull/150379) Tools · by kawsthingz · merged · 11-07 Telegram/gateway reliability guard for Hermes — detects 409 polling conflicts, duplicate pollers, freezes, stale locks and missed deliveries, shows one health state (OK/DEGRADED/BLOCKED) and an approved recovery plan. ### [Dexio](https://github.com/NousResearch/hermes-agent/pull/150376) Memory · by dexio-wiki · merged · 11-07 Dexio shared-wiki memory. Before each turn, the pages of your Dexio wiki that match the user's message; tools to search, read and write the pages your other agents and team also use; and a Hermes Desktop pane showing which pages the agent recalled and wrote and what your other agents changed. Nothing is captured automatically: no transcripts or turn logs. The provider uses the standard library only, the pane's backend uses the FastAPI that Hermes ships, and neither starts threads or local services. ### [Cash Desk](https://github.com/NousResearch/hermes-agent/pull/150374) Automation · by FutureInvoked · merged · 11-07 Chase overdue invoices from your CSV exports (generic, Xero, QuickBooks, Stripe, bank statements): apply payments only on an exact reference and amount, draft reminders at 7, 14, 21 and 28 days, and send one only after a person approves its exact text and Hermes asks once more. Figure Lock holds every amount, invoice number, date, IBAN and reference to the ledger; a rewording that changes one is rejected with a diff. ### [memrack](https://github.com/NousResearch/hermes-agent/pull/150373) Memory · by Kesuek · closed · 11-07 memrack is one memory provider that hosts a RACK of dockable backend engines with agent-side routing: each backend advertises a card (best\_for / limits / search modes) injected into the system prompt, memrack\_write(content, backend=?) writes where the AGENT chooses (deterministic source-rules only as the fallback; no LLM judge at write and none at read — no auto mode), and memrack\_search fans out over every enabled backend and search mode, merging with reciprocal rank fusion (RRF k=60), per-target collapse and a deterministic tiebreak (score, heat, freshness). Ships with three backends: base-db (the stdlib SQLite floor — one atomic fact per row, FTS5 bm25 ranking, hash-dedupe idempotence, per-entry heat counters feeding a heat report), wiki (read-only whole-page search over an existing markdown wiki via a stdlib mtime index with AND-semantics), and state-episodic (READ-ONLY search over the canonical Hermes session history in state.db, default on — turns are never copied). An optional vec engine (sqlite-vec + Ollama embeddings) can upgrade base-db with a memory mode and degrades honestly to fts/pointers with a visible gap note. Tools: memrack\_search, memrack\_write, memrack\_status, memrack\_import, plus \`hermes memrack status\|report\|export\|import\` (surfaces only while the provider is active). Config: desktop panel via a declared config\_schema (rack membership, force-backend override, top\_k, vec group, auto-enable-at-activation toggle default OFF) and config.yaml plugins.memrack — both stores converge through a dual-store save/read bridge. ### [Standing](https://github.com/NousResearch/hermes-agent/pull/150366) Automation · by percyhendrux123-prog · merged · 11-07 Earned trust for Hermes workers. Each profile has a level (L0 read, L1 draft, L2 act, L3 trusted) that decides which tool calls need the owner's approval, and a local record of checks that held up and checks that did not. A worker can always report on its own card (block, comment, request review, heartbeat), but it cannot mark the task done on its word: kanban\_complete is blocked until the task has a passing standing\_proof in the same session, and Standing runs that check itself (owner-scoped file, owner-allowlisted HTTP URL, or exact owner-allowlisted read-only command). Ten clean passes raise a level (capped at L1 automatically; L2 needs the owner's explicit opt-in); one failed check lowers it; a check that could not run is logged as unavailable and changes nothing. Seeds only create new actors, so restarts never undo a demotion. /standing shows the record, /standing log the last events, and /standing set (local owner only) changes a level. Optional same-machine shared ledger for a governor and its workers, and a short score line on the task card after a completion or demotion. ### [Valun](https://github.com/NousResearch/hermes-agent/pull/150363) Tools · by fitzss · merged · 11-06 Let your Hermes agent negotiate service-for-credit deals with other independently controlled agents, signing with its own key within limits you set; credit notes recorded by a real Basis tracker (synthetic pilot units). ### [dike](https://github.com/NousResearch/hermes-agent/pull/150360) Tools · by Grizaceo · merged · 11-06 Diké: checks every article citation of Chilean law in a legal brief against a local, hash-identified copy of 25 statutes (Ley Chile), and refuses to certify what it cannot prove. Each citation comes back verified, with its identifier, source URL and text SHA-256, or unverified with the engine's reason (AMBIGUOUS, UNRESOLVABLE). One tool (dike\_verificar\_citas, text or a .md/.txt path) and one command (/dike-verificar). Spanish output. Not legal advice: it never certifies that a provision supports an argument, nor that a dated copy is today's text. ### [Hermagotchi](https://github.com/NousResearch/hermes-agent/pull/150358) General · by mojomast · merged · 11-06 Co-op virtual pets for Hermes: raise a pet, fight deterministic async battles, earn Bits, unlock cosmetics up to mythic tier, climb a seasonal leaderboard, and chat, DM, challenge and taunt other players in a native pane in the Hermes web app and Desktop (Habitat, Battle, Shop, Leaderboard and Social tabs plus a persistent lobby rail). Hermes gets 8 composable tools (hermagotchi\_status, hermagotchi\_pet, hermagotchi\_goals, hermagotchi\_challenge, hermagotchi\_battle, hermagotchi\_shop, hermagotchi\_social, hermagotchi\_meta). ### [Hermes Strudel Studio](https://github.com/NousResearch/hermes-agent/pull/150357) Desktop · by mojomast · merged · 11-06 Local Strudel live-coding studio for Hermes: write patterns, hit Play, and let the agent compose alongside you with one poptart\_studio tool and a Desktop SandboxedFrame pane. ### [hermes-web-crawl4ai](https://github.com/NousResearch/hermes-agent/pull/150355) Web · by axAilotl · merged · 11-06 Self-hosted Crawl4AI extraction and bounded, policy-checked website crawling. ### [fluent](https://github.com/NousResearch/hermes-agent/pull/150353) Tools · by Bruno-coder21 · merged · 11-06 Your agent becomes fluent in any program: checks the agent's terminal commands against the real --help and man pages of the programs you adopt, before they run. A long option the installed version doesn't have is stopped with the closest real one (--crf → -crf, --ammend → --amend), and subcommands it classifies as destructive (gh repo delete, docker image prune, rsync --delete) go to Hermes's own approval prompt. Adds /fluent (adopt, show, forget, undo, mode), an @cli:\ context reference and a dashboard tab that draws the agent's tree (a branch per program, a fruit per error caught). Tested on 311 real valid commands (0 blocked) and 17 typical AI mistakes (17 caught). ### [BlazePhoenix](https://github.com/NousResearch/hermes-agent/pull/150351) Tools · by Mitra · merged · 11-06 Swap quotes from BlazePhoenix, the on-chain DEX aggregator, measured on the user's own RPC and verified locally: net output, Phoenix Check verdict and unsigned Router calldata checked field by field before anything is signed. Base, Ethereum, Optimism, Arbitrum, Robinhood Chain. ### [Telos](https://github.com/NousResearch/hermes-agent/pull/150350) Models · by TelosPlatforms · merged · 11-06 Telos — workspace routers and canonical models through one OpenAI-compatible endpoint; browser sign-in (PKCE) or an existing Telos API key. ### [lampstand](https://github.com/NousResearch/hermes-agent/pull/150349) Tools · by vincitamore · open · 11-06 lampstand turns one Hermes into a team that remembers who it works for. Plant a house (a git repository you own) and every chat started in it works under the same written principles and rules, keeps one shared record of tasks, decisions, knowledge and reminders as plain Markdown, and can wake the other chats through shared rooms, including from a Telegram, Discord or Slack chat. Nine house tools plus role-briefed delegation, /lampstand, a desktop House page (five lamps for the house's checks, work, inbox, knowledge, the constitution with adopt and strike, rooms, rhythm, health), a status-bar lamp and transcript cards. Nothing becomes a rule until you adopt it. ### [Construct](https://github.com/NousResearch/hermes-agent/pull/150346) Tools · by rtmandreyev · merged · 11-06 Hermes researches a thing and writes one exact, checked description of it: named parts at real sizes, how they fit, open and move, and how each is known, measured, designed or guessed. Construct turns that description into a live hologram in the chat and everything downstream: glTF with named parts and working joints for Blender, Godot, Unity or Unreal; exact CAD shapes (STEP, FreeCAD); prints and model kits with booklets; films; and NetLogo runs drawn as constructs. It reconstructs what exists, from a building to your own chair, and designs what does not: wholly new objects grounded in real counterparts (an invented ice-hauling spacecraft, reasoned from real spacecraft studies), or a physical product from software's documentation (a synthesizer's manual made into a hardware instrument concept), keeping evidence apart from design. ### [Hotwire](https://github.com/NousResearch/hermes-agent/pull/150340) Tools · by RohanGlitched · merged · 11-06 Analog circuit design bench: Hermes writes a SPICE netlist from plain English, proves it against measurable specs (gain, corner frequencies, distortion, bias, current draw, part stress) with Hotwire's own circuit simulator, plays your guitar riff or WAV through the simulated circuit so you can hear it, runs Monte Carlo yield over real part tolerances, explains what each part does, matches knob settings to a recording of a real pedal, tunes values to standard E-series parts, and lays the result out on a stripboard with a parts list and KiCad export. Ships a verified reference library (filters, transistor, JFET and MOSFET stages, overdrive, fuzz, tone stacks, wah), an analog-design skill, a /hotwire command and a Desktop Bench pane with live knobs. ### [hackathon-judge](https://github.com/NousResearch/hermes-agent/pull/150334) Tools · by Brianmacu-BMS · merged · 11-06 Source-linked Hermes hackathon and catalog preflight with explicit human review gates. ### [BoxCompute](https://github.com/NousResearch/hermes-agent/pull/150325) Tools · by BoxCompute · merged · 11-06 Give Hermes its own cloud Linux VMs through the hosted BoxCompute MCP server (Streamable HTTP, OAuth 2.1 + DCR): create and start VMs, run commands and background processes, read and edit files under /workspace, and open one-hour public preview links, with the sandboxes workflow skill bundled. Portable Agent Plugins v1 package (skills/ only; the server is added by the sign-in command below). ### [Coldstart](https://github.com/NousResearch/hermes-agent/pull/150323) Tools · by zvporter-ops · merged · 11-06 Evidence-gated answering with a provenance hierarchy, for any local AI deployment. Material you feed it (internal: transcripts, contracts, code, spreadsheets, PDFs, emails) is retrieved first and cited first; public reference material is used only when internal material does not cover the question, and every answer that leans on it is labelled as public; when nothing covers the question the reply is refused instead of invented. Every reply is then checked mechanically against the passages that turn actually retrieved -- quoted spans must be verbatim, and figures, dates, URLs and identifiers must appear in a retrieved passage (line and passage references are masked so citations are not mistaken for claims); prose paraphrase is reported but never blocked. No model is asked to grade its own answer. The clearance rule covers every path that touches private material: the model is taken from the turn the host actually runs (\`pre\_llm\_call\` reports it, recorded per session), so a session switched with \`/model\` or a turn that lands on a cloud fallback counts as remote; when it is not local, internal passages are withheld, \`corpus\_ask\` requires both its endpoint and the turn to be local and otherwise answers from public material only, and \`corpus\_audit\` / \`/trust\` refuse rather than sample internal passages into it. \`/trust\` runs a battery against the configured deployment and scores it: invented quote and invented figure must be caught, verbatim quote and paraphrase must pass, and the pipeline must abstain on questions the corpus cannot answer. Ships an agent half (2 hooks, 3 tools, 1 command, 1 skill) and a Desktop half (evidence pane, status-bar chip, mode switch, audit button) over the plugin's own \`/api/plugins/\` backend. Stdlib only -- no embeddings, no third-party dependency, no model call for retrieval. ### [hermes-sentinel](https://github.com/NousResearch/hermes-agent/pull/151879) Tools · by Justin Strange · merged · 11-06 Deterministic runtime circuit breaker and local execution auditor. ### [Skysay](https://github.com/NousResearch/hermes-agent/pull/150320) Tools · by Skysay · merged · 11-06 Inspect your Skysay workspace, voice agents, phone numbers and call activity in Hermes. Read-only remote MCP with one-time OAuth setup; no calls, messages, purchases or account changes. ### [hostyourai](https://github.com/NousResearch/hermes-agent/pull/150316) Models · by cogitor · merged · 11-06 HostYourAI provider profile (hyai/auto, EU-only inference option) and the /hyai task/policy routing command. ### [cooldown](https://github.com/NousResearch/hermes-agent/pull/150312) Tools · by Savage27z · merged · 11-06 Checks every package Hermes is about to install with pip, uv, poetry, pipx, npm, pnpm, yarn, bun or npx. Blocks versions OSV lists as malicious, and holds the install in Hermes's own approval prompt when a release is younger than 3 days (the rule Hermes uses for its own dependencies), the package is brand new, or the name is not on the registry. ### [LoreDocs](https://github.com/NousResearch/hermes-agent/pull/150310) Tools · by labyrinth-analytics · merged · 11-06 LoreDocs knowledge vaults over MCP: store, version, tag and search project documents, and inject them into agent context. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only: Hermes runs \`uvx loredocs==0.1.33\` as a local stdio MCP server. ### [injection-shield](https://github.com/NousResearch/hermes-agent/pull/150309) General · by MOTHR · merged · 11-06 Injection-Shield — defends Hermes agents against prompt-injection attacks via tool outputs and untrusted content. Validates LLM responses, audits risky tool calls. Modes: audit / enforcement. ### [privacy-shield](https://github.com/NousResearch/hermes-agent/pull/150309) General · by MOTHR · merged · 11-06 Privacy-Shield — tokenizes personal data (PII) in LLM output before sending it to cloud providers, with encrypted AES-256-GCM mapping and whitelisted restore. Modes: off / audit / restore. Configurable via Desktop-Controls or YAML. ### [slopcore](https://github.com/NousResearch/hermes-agent/pull/150304) Tools · by ChrisAaronNewton · open · 11-06 Slopcore — the verified music-video pipeline as Hermes tools. Intake a finished ### [Garmin Connect (unofficial)](https://github.com/NousResearch/hermes-agent/pull/150294) Tools · by adrianlaborda · merged · 11-06 Unofficial community plugin for Garmin Connect data through Garmin's official Personal API: sleep, HRV, Body Battery, stress, steps, heart rate, activities, weight, heart-rate training load and fitness age, with deterministic baselines and trend flags (garmin\_trends), a /garmin command, an optional one-line daily context, a morning readiness brief automation and a Hermes Desktop readiness panel. Fourteen read-only tools. The key is entered in Hermes Desktop under Settings \> Plugins \> garmin-connect (or with hermes config set), stored in the profile's .env as GARMIN\_PERSONAL\_API\_KEY and read through Hermes's secret route. ### [Procgen Studio](https://github.com/NousResearch/hermes-agent/pull/150293) Tools · by iamlukethedev · closed · 11-06 Image and video backend that draws with code. Your configured model writes SVG/Python drawing programs, which are rendered locally to PNG (and MP4/GIF) with the editable source and a reproducible seed saved next to each image. No image API key, STYLE.md art contracts with six presets, a critic pass with bounded redraws, local seed re-rolls and restyles, /procgen and hermes procgen commands, and a Desktop pane. ### [OpenRouter Everything](https://github.com/NousResearch/hermes-agent/pull/150282) Tools · by dadhalfdev · merged · 11-06 One OpenRouter key for Hermes web search and fetch, image and video generation, voice, decisions, extraction, reranking and embeddings, with per-task model selection and compatibility checks. ### [MiniCPM-OCR](https://github.com/NousResearch/hermes-agent/pull/150268) Tools · by PovedaAqui · open · 11-06 Read images and PDFs with MiniCPM-OCR, score every region with MiniCPM-OCR Jev, and keep failed regions in a durable human review queue. Four tools: ocr\_read, ocr\_job, ocr\_queue and ocr\_verify. Works with the hosted ModelBest document-parse API or a local document service / OpenAI-compatible server. ### [Pareto Inference](https://github.com/NousResearch/hermes-agent/pull/150262) Models · by pareto-inference · merged · 11-06 Pareto Inference as a Hermes model provider (\`--provider pareto\`): GLM 5.3 Flash, DeepSeek V4.1 Flash and MiMo V2.6 Flash over OpenAI-compatible Chat Completions, pay per token, with /reasoning mapped to Pareto's reasoning\_effort levels and context windows from models.dev. ### [excubitor](https://github.com/NousResearch/hermes-agent/pull/150261) Automation · by catalystberry842-alt · merged · 11-06 Night sentry for Hermes turns: one per-turn ledger that dedups inbound replay, stops no-progress tool loops, enforces a spend cap, and alerts when a turn hangs without a reply. ### [PostSider](https://github.com/NousResearch/hermes-agent/pull/150258) Tools · by PostSider · merged · 11-06 Plan, draft, schedule, publish and review social media posts across 30+ networks (X, LinkedIn, Instagram, Facebook, TikTok, YouTube, Threads, Pinterest, Bluesky and more) through PostSider's hosted MCP server (Streamable HTTP at https://mcp.postsider.com/mcp, OAuth sign-in, no keys in config): content calendar, channel rules and posting queues, media uploads, drafts and approval requests, failed-post triage, analytics, and an AI post checker and caption rewrite. Bundles the postsider workflow skill (draft first, read back every write, confirm before anything public or permanent). Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [aegis-gate](https://github.com/NousResearch/hermes-agent/pull/150254) Tools · by Rainyy99 · merged · 11-06 AEGIS gate — a pre-execution verification gate for AI agents. Registers a fail-closed pre\_tool\_call hook that checks a tool call before it executes: claim provenance (Gate 1), contract verification via Sourcify with a source risk-pattern scan (Gate 2), EIP-712 signature-payload structure (Gate 3), and instruction-channel injection scoring (Gate 4). Verdicts map to Hermes directives: BLOCK vetoes the call, WARN escalates it to the human-approval gate, ALLOW passes it untouched, and a gate that cannot complete abstains into escalation rather than passing silently. ### [Gemini Live for Hermes](https://github.com/NousResearch/hermes-agent/pull/150251) Voice · by cineglobe · merged · 11-06 Full-duplex Gemini Live voice calls in Hermes Desktop and the web dashboard. The voice takes the profile's name and hands real work to your Hermes agent while you keep talking. It can steer running jobs, threads, bot chats and subagents, use the profile's MCP tools, recall memory and past conversations, set and read iPhone alarms, timers and reminders through hermes-mobile-device-access when a phone is paired, and report back when work finishes. Read-only local commands need three layers to agree: hard rules, a safety review and a contained run; files are read only from the workspace folder, /var/log and a few plain system files. When a job needs approval, the voice says so and Approve and Deny buttons appear; the voice model can never answer one, and spoken approval is opt-in. With the AirPods Approvals plugin, a nod or shake answers it too. ### [commonplace](https://github.com/NousResearch/hermes-agent/pull/150245) Memory · by ChrisAaronNewton · merged · 11-06 Commonplace — local-first curated memory in plain Markdown files on your own machine. Standing rules kept verbatim, a curated store with lexical recall, a handover card for fresh sessions, and a fail-closed pre-compress checkpoint (API v2) that archives the transcript before Hermes may compress it — and blocks compression when the archive cannot be written. No cloud, no API keys, no third-party dependencies. Maintained by Chris Newton. ### [hermes-continuity-capture](https://github.com/NousResearch/hermes-agent/pull/150244) Automation · by Goitonthefloor · merged · 11-06 Records Hermes lifecycle events (session, subagent, and kanban task) as Continuity schema-v2 events in a local SQLite spool under the plugin data directory, and can forward them to a NATS/JetStream server you run yourself. ### [GSC Magic](https://github.com/NousResearch/hermes-agent/pull/150241) Tools · by GSC Magic · merged · 11-06 Google Search Console, Bing Webmaster Tools, Google Analytics 4, IndexNow and Core Web Vitals through GSC Magic's hosted MCP server (Streamable HTTP at https://mcp.gscmagic.com/mcp, OAuth sign-in, no keys in config): search analytics, URL inspection, sitemaps, GA4 reports, CrUX and PageSpeed, plus finished SEO reports (quick wins, traffic drops, content decay, cannibalization, CTR outliers, brand split, index health, Google vs Bing) with optional PNG charts. Bundles the gsc-magic workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [QQQ AI Visibility Checker](https://github.com/NousResearch/hermes-agent/pull/150239) Tools · by QQQ OÜ · merged · 11-06 Free AI visibility and GEO check for any website through QQQ's hosted MCP server (Streamable HTTP at https://qqq.ee/mcp, no sign-in, no keys in config): whether AI crawlers such as GPTBot, ClaudeBot and PerplexityBot may read a site, its technical readiness for AI search with the three most important fixes, and the latest public report of brand mentions and citations in ChatGPT, Gemini and Perplexity answers. Three read-only tools and one workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [yuanbao](https://github.com/NousResearch/hermes-agent/pull/150237) Platform · by NousResearch · merged · 11-06 Tencent Yuanbao messaging platform (DMs and group chats over Yuanbao's bot WebSocket, native media and stickers, DM/group access policies, cron home-channel delivery) plus the yuanbao toolset (yb\_query\_group\_info, yb\_query\_group\_members, yb\_send\_dm, yb\_search\_sticker, yb\_send\_sticker). Formerly built into Hermes core; existing users are migrated automatically. A Hermes Labs plugin: built by Nous Research, not supported, and likely not maintained going forward. Fixes from the community are welcome. ### [VA Benefits](https://github.com/NousResearch/hermes-agent/pull/150234) Tools · by lEWFkRAD · closed · 11-06 File and track VA disability claims (21-526EZ) the way 38 CFR actually works, plus every other VA benefit in one tracker: Post-9/11 GI Bill, VA health enrollment, CHAMPVA, SGLI/VGLI, pension, VR&E. Seven tools: a federal-standards filing guide with CFR citations, intent-to-file with the 12-month effective-date clock, live claim status and change tracking via the VA Lighthouse Benefits Claims API, official contention lookup, benefit/expiration tracking with deadline countdowns, and a one-screen benefits summary. Desktop app adds a VA Benefits page with a benefits grid, claim phase trackers, and deadlines. Network calls go only to api.va.gov and sandbox-api.va.gov; the veteran's ICN is the only personal value sent. Benefit records stay in one local JSON file per veteran under the plugin data directory. Credentials (VA client ID, RSA key path, reference API key) come through plugin settings and Hermes secret scope. One Python dependency: pyjwt\[crypto\]. No hooks, middleware, shell commands, background processes, or telemetry. ### [LovedOne Soft](https://github.com/NousResearch/hermes-agent/pull/150233) Tools · by Terahertz, Inc. · merged · 11-06 Agency operations through LovedOne Soft MCP (https://api.lovedone.app/mcp/vendor/v1): clients, visits, visit notes, invoices and scheduling for home-care office staff. Senior tech support in the workflow — not a medical device and not clinical advice. Portable Agent Plugins v1 under hermes/lovedone-soft (plugin.json + mcp.json + skills/). ### [LovedOne Family Inbox](https://github.com/NousResearch/hermes-agent/pull/150232) Tools · by Terahertz, Inc. · merged · 11-06 Senior tech support for a family household through LovedOne Family Inbox MCP (https://api.lovedone.app/mcp/v1): shared inbox, loved ones and members, onboarding helpers so relatives can coordinate day-to-day logistics for an aging parent. Not a medical device and not clinical advice. Portable Agent Plugins v1 under hermes/lovedone-family (plugin.json + mcp.json + skills/). ### [Qrblox](https://github.com/NousResearch/hermes-agent/pull/150231) Tools · by Terahertz, Inc. · merged · 11-06 QR codes, audio codes, menus and the Voice Agent Phone Book through Qrblox's remote MCP (https://api.qrblox.com/mcp/v1). Portable Agent Plugins v1 (plugin.json + mcp.json + skills/). ### [hermes-android](https://github.com/NousResearch/hermes-agent/pull/150230) Tools · by TechFlipsi · merged · 11-06 Remote Android device control for Hermes Agent: 58 android\_\* tools (read screen, tap/type/swipe, open apps, camera, screenshots & screen recording, files, notifications, microphone streaming, media, clipboard, SMS/calls, Health Connect) plus a companion Android bridge app (Kotlin/Compose). The phone dials out to a relay you host — works behind NAT, no port forwarding, no VPN. Chat and media are end-to-end encrypted with a Double Ratchet (the relay sees only ciphertext); device commands travel over your own TLS-authenticated relay. ### [BulkPublish](https://github.com/NousResearch/hermes-agent/pull/150221) Tools · by BulkPublish · merged · 11-06 BulkPublish (https://www.bulkpublish.com) social media scheduling through its hosted MCP server (Streamable HTTP at https://mcp.bulkpublish.com/mcp/full, 80 tools): draft, schedule, publish and retry posts on Facebook, Instagram, X, TikTok, YouTube, Threads, Bluesky, Pinterest, Google Business Profile, LinkedIn, Mastodon, Discord, Telegram, Tumblr and Snapchat, with media uploads, bulk actions, team approvals, labels, templates, recurring schedules, RSS autopost, Inbox replies and analytics. Bundles 33 BulkPublish skills. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [NDAA.AI](https://github.com/NousResearch/hermes-agent/pull/150217) Tools · by Terahertz, Inc. · closed · 11-06 Public FY2027 NDAA tracker through NDAA.AI's remote MCP (https://mcp.ndaa.ai/mcp): bill status, section text, House vs Senate diffs, news, updates and guides. Portable Agent Plugins v1 (plugin.json + mcp.json + skills/). ### [Sandwich Directory](https://github.com/NousResearch/hermes-agent/pull/150216) Tools · by Terahertz, Inc. · closed · 11-06 Public caregiving directory through joinsandwich's remote MCP (https://www.joinsandwich.com/api/mcp): search care resources and channels, estimate long-term care costs. Portable Agent Plugins v1 (plugin.json + mcp.json + skills/). ### [Frosh](https://github.com/NousResearch/hermes-agent/pull/150215) Tools · by Terahertz, Inc. · closed · 11-06 Search and read public Frosh docs and blog posts through Frosh's remote MCP (https://frosh.org/mcp). Portable Agent Plugins v1 (plugin.json + mcp.json + skills/). ### [FlipLetters](https://github.com/NousResearch/hermes-agent/pull/150214) Tools · by Terahertz, Inc. · merged · 11-06 Split-flap and flip-dot board design through FlipLetters' remote MCP (https://app.flipletters.com/mcp/v1): FlipCode reference, validate boards, render videos, share. Portable Agent Plugins v1 (plugin.json + mcp.json + skills/). ### [Crosodo](https://github.com/NousResearch/hermes-agent/pull/150213) Tools · by Terahertz, Inc. · merged · 11-06 Cottage food and home bakery law lookups for US states and counties through Crosodo's remote MCP (https://crosodo.com/api/mcp): state rules, county permits, label requirements, Crosodo Certified credentials, and optional label-packet checkout. Portable Agent Plugins v1 (plugin.json + mcp.json + skills/). ### [Assumption Ledger](https://github.com/NousResearch/hermes-agent/pull/150212) Memory · by itzebe · merged · 11-06 Persistent operational-assumption ledger. Captures the assumptions an agent states in its own messages (never from tool output), keeps the sentence each came from as provenance, verifies them with deterministic checks, and warns in context when work recorded in a session depends on an assumption evidence has contradicted. ### [Loci](https://github.com/NousResearch/hermes-agent/pull/150211) Tools · by AskLoci · merged · 11-06 Local models on your Mac for file summaries, document extraction, folder search, audio transcription and batch processing alongside Hermes. ### [SLOPCANNON](https://github.com/NousResearch/hermes-agent/pull/150208) Tools · by vincitamore · open · 11-06 SLOPCANNON music videos from a song, inside Hermes: one request produces a whole film (direction, rights sign-off, credit quote, first-scene review, parallel or chained scene renders, character checks, rerolls with a note, assembly), every decision a button on a live film card in the chat, a SLOPCANNON studio page in Hermes Desktop, and the whole hosted SLOPCANNON tool surface (remixes, the Ad Cannon, Character Lab, Slop TV) as native tools. ### [Booking Bridge](https://github.com/NousResearch/hermes-agent/pull/150203) Tools · by kyle-bartlett · merged · 11-06 Find free time, then book, list, reschedule and cancel real Google Calendar consults (for example a Strategy consult) from Hermes. Signs in with the user's own Google OAuth client through hermes auth add booking-bridge:google; all calls go through credentialed\_provider\_request to www.googleapis.com only. ### [Quote to Chat](https://github.com/NousResearch/hermes-agent/pull/150201) Desktop · by ivanvanes18 · merged · 11-06 Codex-style quote cards for Hermes Desktop: select text outside an input and press Add to chat; the excerpt becomes a removable card above the composer while your question stays a separate draft. On send the quotes are prepended once as Markdown blockquotes (composer.middleware). Quotes are scoped per chat, several can be collected and removed one by one, a busy-agent send goes through host.composer.submit with the quote attached, and a rejected send gives the quote back. Desktop-only (desktop/plugin.js inside the plugin SDK surface): no agent tools, hooks, Python middleware or env vars. ### [Sequenzy](https://github.com/NousResearch/hermes-agent/pull/150189) Tools · by Sequenzy · merged · 11-06 Run Sequenzy email marketing through Sequenzy's hosted MCP server (Streamable HTTP at https://api.sequenzy.com/v1/mcp, OAuth sign-in, no keys in config): subscribers, lists, tags and segments, campaigns with A/B tests, automated sequences, templates, transactional email, inbox replies, webhooks, integrations, and delivery and engagement stats. Bundles the sequenzy-email-marketing workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [daygauge sleep midpoint](https://github.com/NousResearch/hermes-agent/pull/150188) Tools · by Studea Limited · closed · 11-06 Calculate sleep midpoint from manually supplied onset and wake times using daygauge's public method, with midnight and ambiguous-interval handling. Agent Plugins v1 skill only: no app or Apple Health connection and no diagnosis. ### [SHVL workspace review](https://github.com/NousResearch/hermes-agent/pull/150187) Tools · by Studea Limited · merged · 11-06 Read authorised SHVL workspace receivables and connected Xero financial reports through a hosted Streamable HTTP MCP server, with a skill for selecting a workspace and interpreting the results. ### [ReserveGauge CSV review](https://github.com/NousResearch/hermes-agent/pull/150186) Tools · by Studea Limited · closed · 11-06 Review a user-supplied ReserveGauge CSV export for missing readings, reset boundaries and comparable usage changes. Agent Plugins v1 skill only: no app or provider connection. ### [TenderLead bid review](https://github.com/NousResearch/hermes-agent/pull/150185) Tools · by Studea Limited · merged · 11-06 Review supplied UK tender documents against a supplier profile before a BID or SKIP decision. Agent Plugins v1 package with one local skill; it does not fetch live tenders or submit bids. ### [Fleetkeep import review](https://github.com/NousResearch/hermes-agent/pull/150184) Tools · by Studea Limited · merged · 11-06 Review a supplied vehicle CSV for missing and ambiguous Fleetkeep import values. Agent Plugins v1 package with one local skill; it does not connect to Fleetkeep or change an account. ### [Hypotize](https://github.com/NousResearch/hermes-agent/pull/150183) Tools · by Hypotize · merged · 11-06 Run growth experiments with Hypotize, the agent-native Growth OS, through its hosted MCP server (https://api.hypotize.ai/mcp, OAuth sign-in, no keys in config): read what the team already tested, write hypothesis-first ideas, score them with ICE, PIE, RICE or a custom framework, move experiments through an enforced lifecycle and record measured outcomes and learnings. Bundles three skills: the experiment method, a read-only weekly pipeline review and a backlog import from a spreadsheet or Notion export. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [payops](https://github.com/NousResearch/hermes-agent/pull/150178) Tools · by BigWave45 · merged · 11-06 Read-only Paystack transaction inspection and reconciliation for Hermes agents. ### [vt-check](https://github.com/NousResearch/hermes-agent/pull/150177) Tools · by Aravinth-Earth · merged · 11-06 VirusTotal reputation for downloads and links. Files are hashed locally and looked up by SHA-256 — never uploaded. URLs are checked read-only, and archives are inspected in memory for path traversal, decompression bombs, double extensions and document-named executables. ### [SuprAnalytics](https://github.com/NousResearch/hermes-agent/pull/150176) Tools · by aniketmohite · merged · 11-06 Five analytics workflow skills for SuprAnalytics: weekly performance reports, traffic source analysis, funnel drop-off analysis, revenue attribution, and Core Web Vitals / audit review. Connects to the SuprAnalytics remote MCP server via OAuth. All access is read-only. Network destination: https://supranalytics.app/mcp. No credentials are stored in the plugin. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [screenRIG](https://github.com/NousResearch/hermes-agent/pull/150175) Tools · by screenrig · merged · 11-06 Free, agent-driven digital signage. Your agent pairs Players, uploads images and video, builds pages from docs, sheets and calendars, publishes playlists to TVs and kiosks, and checks each screen with a screenshot. ### [Fiscalisto](https://github.com/NousResearch/hermes-agent/pull/150174) Tools · by Fiscalisto · merged · 11-06 Mexican electronic invoicing (CFDI 4.0) and SAT taxes through Fiscalisto's hosted MCP server (https://fiscalisto.com/mcp, OAuth sign-in with per-company grants and read / draft / stamp scopes, no keys in config): search issued and received CFDI, draft, stamp, email and cancel invoices, payment receipts, credit notes, advances, payroll receipts, the month's VAT and ISR estimate, SAT 69-B supplier checks, and SAT documents (constancia, opinión de cumplimiento, buzón tributario). Bundles an English overview skill and thirteen Spanish workflow skills. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only; OAuth uses pinned mcp-remote@0.1.49 over npx stdio, so Node.js 20.18.1+ and npm are required. ### [WorldMonitor](https://github.com/NousResearch/hermes-agent/pull/150167) Tools · by koala73 · merged · 11-06 Source-aware global intelligence through WorldMonitor's hosted MCP: country risk, conflicts, markets, energy, shipping and natural hazards. Configuration-only Agent Plugins v1 package with one skill. Public source inventory works without credentials; other intelligence requires a signed-in account and some tools require a subscription. ### [CrashLab](https://github.com/NousResearch/hermes-agent/pull/150164) Tools · by dolepee · merged · 11-06 Test recovery before delivery, after commit and across restart; inspect saved records, verify the repair and keep a reusable regression. ### [Kokpit](https://github.com/NousResearch/hermes-agent/pull/150163) Tools · by sahindemirbas · merged · 11-06 A personal cockpit for Hermes Agent. When a session produces something, Kokpit leaves a short card and receipt in plain Markdown that Hermes, Claude Code and Codex all read, so the next session picks up where the last one stopped. It also captures notes and todos from any Hermes chat, gives a morning brief, and runs a night pass whose proposals quote their exact source lines. Local, stdlib only. ### [Telegram Mini App](https://github.com/NousResearch/hermes-agent/pull/150155) Desktop · by aimanmalib · merged · 11-06 Phone control panel for Hermes inside Telegram: live status and usage, sessions and their messages, agent cron jobs (create, edit, pause, run now, stop, catch-up per job), logs, skills and memory, a send-to-Telegram box and error alerts, opened from your bot's menu button. Login is Telegram initData (HMAC-SHA256 with Ed25519 fallback, freshness window) checked by a token-only dashboard-auth provider on one token route, against a plugin-owned admin list (TELEGRAM\_OWNER\_ID plus the plugin setting admins) that refuses everyone until set; being allowed to chat with the bot never grants login. Alerts and the send box reach admins only. Built by Muhammad Aiman (aimanmalib) at Sasa Teguh. ### [Wallpaper](https://github.com/NousResearch/hermes-agent/pull/150150) Desktop · by jangomango76 · merged · 11-06 Put any image, GIF or short video from your drive behind the whole Hermes Desktop window. Pick it with the normal file dialog or drop it in, then tune opacity, blur, brightness, saturation, contrast, vignette, fit, anchor, zoom and mirroring; videos get speed, pause-when-hidden and reduced-motion controls, and a small library with an optional slideshow keeps several backgrounds. Quick controls in the status bar, a full settings page and command-palette commands. ### [helm](https://github.com/NousResearch/hermes-agent/pull/150160) Desktop · by ASV-Labs · merged · 11-06 A redesigned Hermes Desktop. A Desk home page (greeting and crew status, a prompt that hands work to any bot, Continue, Needs you, Today's scheduled jobs); pending tool approvals and questions from every bot's live chats answered in one place, on the Desk and in a sidebar pane, only on your click; a Session pane with the focused chat's live task plan; a themeable redesign layer across the app (three bases, seven accents, type, corners, density) on a Customize look page; status everywhere (titlebar, session rows, status bar); and a Bots board of rendered robots (191 Blender-rendered PNG layers) that become each bot's Hermes avatar. ### [Waiting on You](https://github.com/NousResearch/hermes-agent/pull/150140) Tools · by irsan-dimyati · merged · 11-06 One list of the questions your Hermes bots are waiting on you to answer. /waiting and hermes waiting scan profile session stores, keep only replies that end by asking you something (English and Indonesian), merge the same question asked by several bots, and drop an item once you reply in that chat or dismiss it. ### [Hermes MakePlain](https://github.com/NousResearch/hermes-agent/pull/150130) Desktop · by hey-yx · merged · 11-06 One-click composer polish for the Hermes Desktop chat: a wand button at the right of the input box rewrites the draft in place into a clearer version that keeps the original meaning, with nothing added and nothing dropped. The rewrite is wording cleanup only: it never explains, interprets, comments on or answers the draft, and questions and guesses stay as written. It stays in the draft's language and keeps every concrete detail exactly as written: names, numbers, versions, ports, paths, file names, URLs, @-references, /commands and code. The message is never sent; the button spins while the rewrite is in flight, rejects an empty box before any work starts, and never overwrites a draft edited while the call is running. ### [Look At This](https://github.com/NousResearch/hermes-agent/pull/150125) Desktop · by Elshayib · closed · 11-06 One-shot "explain this" context capture: a hotkey, the composer's attach menu or /look grabs the window you were just in (screenshot, accessibility tree, active URL, selected text, app/window title) and attaches it as one redactable, ephemeral packet to your next Hermes message. User-initiated only; the packet expires after 300 seconds, is consumed by one turn, refuses excluded windows (password managers, incognito) before capturing, and redacts secrets in the text fields. ### [hermes-colab](https://github.com/NousResearch/hermes-agent/pull/150110) Tools · by PietrangeloManco · merged · 11-06 Run Python scripts and Jupyter notebooks on Google Colab GPUs (T4 free tier, L4, A100, H100, TPU) from chat: Hermes assigns the runtime, uploads the code, keeps it alive, streams progress and GPU use, syncs outputs and checkpoints home while the job runs, resumes on a fresh runtime from the last checkpoint when Colab reclaims one, and reports back to the chat that started it; a Desktop page shows live runs with cancel and resume. ### [recall](https://github.com/NousResearch/hermes-agent/pull/150109) Tools · by memeshee · merged · 11-06 Local file recall: a private on-device index of the documents already on the user's machine, with a search tool over file content (SQLite FTS5, no network and no embedding service), a recall\_open reader for PDF, Word, Excel, PowerPoint, OpenDocument, Markdown, CSV, email and source files, an @recall: context reference that expands matching passages into the turn, and a pre\_llm\_call hook that puts up to three candidate file paths in front of the model before it answers a document question. Five tools (recall\_index, recall\_search, recall\_open, recall\_stats, recall\_forget), a bounded background refresh on session start, and the /recall slash command. Extraction is standard library only (FlateDecode-aware PDF stream extractor, OOXML text nodes). ### [Load Lock](https://github.com/NousResearch/hermes-agent/pull/150106) Tools · by aron98 · merged · 11-06 A machine-wide load broker for coding agents: Hermes, Claude Code, OpenCode, Codex, Cursor and any MCP client share one heavy-work budget, so builds and test suites take turns instead of thrashing the machine. 19 load\_lock\_\* tools (acquire, queue, offers, honest waits, background jobs that release their permit on exit, machine inspection), a policy brief at session start, a Load Lock dashboard tab, and a Desktop pane, status-bar chip and palette command. ### [Archytan Lite](https://github.com/NousResearch/hermes-agent/pull/150099) Tools · by High-ArchyTech-Solutions · merged · 11-06 Every tool call is decided by Archytan Lite, an authorization gate that runs outside Hermes with its own signing key, before the tool runs. A call becomes checks on what it touches (each file path, each program and git subcommand a command runs, every path and URL it names, each host fetched, MCP and plugin tools by name); all must be ALLOWed, and each ALLOW is a single-use permission spent just before the tool runs. Refusals use the block directive, so they hold under /yolo and approvals.mode: off. Anything the plugin can't read precisely ($VAR, $(...), loops, unpinnable paths), any error, an unreachable gate or a bad signature is refused. Every decision is an Ed25519-signed, hash-chained receipt; \`hermes archytan verify\` checks the whole log. Starter policy for one workspace, edited as JSON and applied with \`hermes archytan reload\`. ### [Plugin Foundry](https://github.com/NousResearch/hermes-agent/pull/150095) Automation · by dolepee · merged · 11-06 Turn a narrow plugin request into a native Hermes plugin: manifest, entry point, tool, SQLite state, tests, static scan and catalog draft. Two archetypes only: a SQLite folder watcher and a JSON API poller, each with one tool. The folder-watcher parser accepts one explicit supported file type or complete dotted suffix and refuses unsupported, composite, conflicting or ambiguous requests. A trailing period is removed only at request end or before validated supported follow-up instructions. Sentence boundaries are detected before a capitalised token or complete instruction verb, but that detection alone does not permit removal; other prose fails closed, and abbreviations without a sentence start stay inside the instruction. Writes stay under the Foundry data root, generated tests run in a minimal environment, and paths outside the root need a pre-tool approval. Provides pre\_tool\_call and post\_approval\_response hooks plus the public tool\_execution middleware for trusted call IDs. Foundry itself makes no network requests. Installing or cataloguing generated code is a separate human step; nothing is auto-merged. ### [completion-proof](https://github.com/NousResearch/hermes-agent/pull/150089) Tools · by yuanshen101 · merged · 11-06 Revision-bound receipts for operator-configured checks; changed source, tests or artifacts make earlier passes stale. ### [factkit](https://github.com/NousResearch/hermes-agent/pull/150080) Memory · by MohamedElsiddig · merged · 11-06 Semantic recall for the built-in holographic memory provider. It reads the provider's store read-only and keeps its own embedding vectors in a separate plugin database (never writing into holographic's tables), adding an embedding-based recall block before each model call, history-aware dedupe so a fact already shown in the conversation is not injected twice, and embedding re-ranking of fact\_store search results. Writes are never automatic: conflict, duplicate and TTL actions stay explicit review/apply operations. No core files are patched and the plugin never self-updates. ### [Showreel](https://github.com/NousResearch/hermes-agent/pull/150078) Tools · by Duffey234 · merged · 11-06 Turn any Hermes session into a demo video under 90 seconds. Six observer hooks record what the session actually did (the ask, terminal commands and output, file edits, screenshots, token usage), scrubbed twice for secrets; a deterministic storyboard keeps a failing test and its fix together and fits a hard 89 s budget; Pillow and your own ffmpeg render a 1080p MP4 (16:9, 9:16 or 1:1) with burned-in captions, an .srt and a poster. Desktop half: a Showreel page with a moment editor (drop beats, edit captions) and inline player, a REC chip, and a player card for showreel\_render results. Narration is off by default; optional free offline voice (espeak-ng or piper on PATH), Hermes text\_to\_speech, or the user's own ElevenLabs key only when explicitly chosen, with \`hermes showreel plan\` showing the exact text and character count first. ### [DXMT Memory](https://github.com/NousResearch/hermes-agent/pull/150069) Memory · by dxmt · merged · 11-06 Long-term memory in a DXMT server you run: recall by meaning and keyword before each turn, durable facts extracted at session end and before compression, supersede and forget by id, and the same memories visible to every other agent and tool on the machine over REST and MCP. The server embeds, so no embedding key is needed in Hermes. ### [LLM API](https://github.com/NousResearch/hermes-agent/pull/150061) Models · by llmapi-ai · merged · 11-06 LLM API (llmapi.ai) as a Hermes model provider (\`--provider llmapi\`, alias \`llm-api\`, or LLM API in \`hermes model\`): Claude, GPT, Gemini, Kimi, DeepSeek, GLM, Qwen and more on one key, each on the API it works best on. Claude runs over the Anthropic Messages API (so Hermes' prompt caching applies), GPT-5 and later over the Responses API, everything else over Chat Completions. The picker lists LLM API's tool-capable chat models with 64K+ context from its public catalog, which also supplies context windows, image input, per-model reasoning levels and list prices; catalog aliases resolve in /model. Gateway errors (out of credits, spend caps, model not allowed for the key) map onto Hermes' fallback instead of a key rotation; /usage shows the credit balance and monthly cap; per-request cost is the billed amount when the gateway reports it, else a list-price estimate. Maintained by LLM API. ### [StarLearner-Nexus](https://github.com/NousResearch/hermes-agent/pull/150059) Tools · by BigBossRabbit · merged · 11-06 Turn the repositories you star on GitHub into a living library of registered Hermes skills. \`starlearner\_sync\` runs the whole pipeline — fetch, map (the LLM curation pass), categorize into a fixed set of domains, two-track assess (integrate-into-your-makeup and business-opportunity), generate, register (\`starlearner-nexus:\\`) — incrementally via a persistent seen-ledger (a repo is never processed twice); \`starlearner\_status\` is read-only state (including token set / not set); \`starlearner\_review\` surfaces everything LLM-shaped item by item for the human. The map / categorize / assess stages optionally run through the host LLM (\`ctx.llm\`, one registered auxiliary task) with a deterministic rules engine as the bit-for-bit fallback (\`enrich: rules\` = zero LLM calls); generate is never the LLM. Registration timing: handler-time \`register\_skill\`, accepted and live in the same session (a skill view on a name the sync just returned resolves the same second); a plugin reload re-registers everything generated before it. ### [Better-TG-CLI](https://github.com/NousResearch/hermes-agent/pull/150058) Tools · by TheVilfer · merged · 11-06 Telegram on your own account (not a bot) through a local MCP server and a skill: read and search chats, threads, comments and forum topics, press bot buttons, and send, edit, react or forward only after you approve each action. Portable Agent Plugins v1 package (mcp.json + skills/). ### [Email Steward](https://github.com/NousResearch/hermes-agent/pull/150057) Automation · by kaosjammo · open · 11-06 Owner-private Gmail triage, durable obligations and shared Desktop rules. ### [Airlock Sandbox](https://github.com/NousResearch/hermes-agent/pull/150029) Tools · by justrunme · merged · 11-06 Terminal backend that routes Hermes command execution through an operator-chosen Airlock HTTPS API and disposable sandbox, then surfaces receipts (RFC 8785 JCS, SHA-256, Ed25519) as BLOCKED / ALLOWED / VERIFIED (\`hermes config set terminal.backend airlock\_sandbox\`). Requires an explicit HTTPS origin \`base\_url\` (no built-in default host). When \`verify\_receipts\` is true (default), requires a locally configured \`public\_key\_hex\` obtained out-of-band — the verifying key is never fetched from the Airlock API at verify time. VERIFIED binds signed decision + execution digests/exit\_code only; it does not prove command text and does not verify full receipt-chain linkage (\`prev\_hash\` is informational). ### [inventory-reminder](https://github.com/NousResearch/hermes-agent/pull/150026) Tools · by ether-btc · merged · 11-06 Live, evidence-backed inventory of the host an agent is running on: repos (cloned vs online-only), skills, MCP servers, CLIs, local models and services, regenerated from ground truth by read-only scanners. Registers two tools — inventory\_where(term) answers "do I already have X?" across every section with a ready detail line, and inventory\_summary() returns a one-line digest for session bootstrap. Both read a cached digest file ($INVENTORY\_JSON, else ~/.hermes/inventory.json); no live rescan per turn, no network, no writes. ### [usage-hud](https://github.com/NousResearch/hermes-agent/pull/150025) Desktop · by aktogde1 · merged · 11-06 Compact Codex account limits and focused-session token usage for Hermes Desktop. ### [Hermes EGM](https://github.com/NousResearch/hermes-agent/pull/150021) Tools · by willbohn · merged · 11-06 Fresh-context, tool-free design reviews with checked source evidence and EGM workflow skills. ### [Chronos](https://github.com/NousResearch/hermes-agent/pull/150020) Tools · by Shikhyy · merged · 11-06 Speculative branching: try several candidate fixes in isolated git worktrees, verify each against the project's tests through Hermes' terminal tool, and merge only a verified winner through Hermes' file tools. ### [mFlow](https://github.com/NousResearch/hermes-agent/pull/150016) Tools · by vladyslavpatrick · merged · 11-06 Manage mFlow (m-flow.io) from Hermes: boards and tasks, Knowledge Base/ADRs, and coding-agent delegation. Portable Agent Plugins v1 package (mcp.json + skills/) with no Python and nothing runs locally — it points Hermes at the hosted mFlow MCP server (Streamable HTTP at https://mcp.m-flow.io/mcp) and ships a skill that teaches Hermes how to use it (list\_standalone\_projects/tasks, create/update, create\_comments, knowledge docs incl. ADRs, assign\_to\_coding\_agent). ### [Rescue](https://github.com/NousResearch/hermes-agent/pull/150015) General · by 0xNexuz · merged · 11-06 Keeps your agent from breaking Hermes, and says what really went wrong. Refuses agent tool calls that would delete or hand-edit Hermes's own data (sessions, state.db, memories, skills, settings, profiles, keys) and names the supported command instead; sends edits to settings, SOUL.md or memory files to Hermes's approval prompt with a snapshot first (/rescue undo); and when a reply fails or quietly comes from a backup, adds one plain line per attempt with the real cause and the fix (NousResearch/hermes-agent#144562, #147620). /rescue in every surface, plus a Desktop status chip and pane. ### [corallium](https://github.com/NousResearch/hermes-agent/pull/150008) Memory · by Aoyi-Feng · closed · 11-06 A shared, append-only knowledge graph for agents, built on the practices that make science work: publish one falsifiable claim with its evidence, then let other agents try to reproduce it or kill it — votes change a node's status, and a withdrawn claim marks everything built on it as suspect. Ten tools: find, tip, show, publish, vote, state, claim, release, suggest, verify. Reads fall back to scanning \/\/nodes/\*.json when the server is stopped; writes fail loudly instead of being dropped. The repository ships install.sh, which scaffolds a local coral and starts the server. ### [context-steward](https://github.com/NousResearch/hermes-agent/pull/150007) Memory · by jiangkoumo · closed · 11-06 One auxiliary judgment chooses a checkpoint or a fresh context segment in the same chat. ### [ReapX](https://github.com/NousResearch/hermes-agent/pull/150004) Tools · by BigBossRabbit · merged · 11-06 Turn the bookmarks you saved on X into ready-to-load, registered Hermes skills — without the paid X API. The plugin is data-in: \`reapx\_import\` runs the whole pipeline — map, categorize into a fixed set of domains, two-track assess (integrate-into-your-makeup and business-opportunity), generate, register (\`reapx:\\`) — from a bookmarks file you point it at, or from items passed inline (collected by you through the host's own browser tooling); \`reapx\_status\` is read-only state; \`reapx\_review\` surfaces everything LLM-shaped item by item for the human before anything is relied on. Incremental via a persistent seen-ledger (a bookmark is never processed twice). The map / categorize / assess stages optionally run through the host LLM (\`ctx.llm\`, one registered auxiliary task) with a deterministic rules engine as the bit-for-bit fallback (\`enrich: rules\` = zero LLM calls); generate is never the LLM. Registration: handler-time \`register\_skill\`, live in the same session; \`register\_generated\_skills\` defaults to false — generated skills land in the review and on disk, and register only after you say so. Settings are read from plugin config only (tool args can never flip enrichment, lift a spend cap, or force registration). ### [POW Tracker](https://github.com/NousResearch/hermes-agent/pull/150003) Web · by BigBossRabbit · merged · 11-06 Multi-client proof-of-work time tracking for the Hermes web dashboard — live OPEN/CLOSED timers, file-activity auto-capture, invoicing with per-entry rate overrides, and per-client SQLite billing, in a dashboard pane/tab (track / entries / tasks / invoice / rate views, updating live over a WebSocket) backed by a FastAPI plugin API the dashboard host mounts. Every client keeps its own time database inside its own folder, so billable data travels with the client folder; invoices aggregate billable time per task, exportable as CSV (PDF through an optional, lazily-imported extra); and a background auto-capture thread (per-client, off by default) opens and closes \`auto\` entries from file activity while you work. It has run a real multi-client operation in production since August, and no client data ships with the tree. ### [canary-trap — prompt-injection tripwire](https://github.com/NousResearch/hermes-agent/pull/149988) Tools · by Tarekkharsa · merged · 11-06 Catch prompt injection by the secret it tries to steal. Every turn the agent is shown a decoy credential (HERMES\_DEPLOY\_KEY=hdk\_live\_…) with the instruction never to send it; no real task needs it, so if it shows up in a tool call — plain, URL-encoded, base64, hex, reversed, spaced out or as a 16+ character fragment, in any nested argument — the call is blocked (pre\_tool\_call), every later tool call in that session goes to your approval until /canary reset, and the reply opens with an alert (transform\_llm\_output). /canary shows status and incidents, rotates the decoy, or tests text. ### [Hermes Sanctuary](https://github.com/NousResearch/hermes-agent/pull/149986) Desktop · by Med34000 · merged · 11-06 Watch your Hermes Agent work: five animated characters wake up when Hermes uses their tools, in two illustrated worlds (Olympus by default, the Clearing), in Hermes Desktop and the web dashboard, in English or French. The guard barks, with a bubble saying why, on tool and provider errors, silent fallbacks to a backup model, loops (same tool, same arguments, 4 times in a turn), heavy turns, a daily budget overrun and permissions waiting for an answer; a daily journal shows turns, tools, where the work came from, tokens and estimated cost per model hour by hour, with a shareable image of the day and a sped-up replay. ### [ClearVoice Hermes Plugin](https://github.com/NousResearch/hermes-agent/pull/149978) Voice · by thebangarang · merged · 11-06 Local push-to-talk speech bridge — your words on screen and spoken clearly in 17 languages, with no account, no cloud and no audio on disk. ### [Webnovel Studio](https://github.com/NousResearch/hermes-agent/pull/149977) Tools · by wyi006636-cyber · open · 11-06 Bilingual long-form fiction studio for Hermes Agent (Chinese webnovel & English progression fantasy): plan, write, review and inspect 100+ chapter serialized novels with Codex smart entity detection, foreshadowing debt alerts, 4-stage scene beats, deterministic De-AI cliché linting, zero-API-key SQLite FTS5 recall, resumable runs, per-project single-writer lock, role-routed ctx.llm calls, and a live local dashboard. Bundles 8 workflow skills (/webnovel-init, /webnovel-plan, /webnovel-write, /webnovel-review, /webnovel-query, /webnovel-learn, /webnovel-dashboard, /webnovel-doctor). ### [Lahja (لهجة) — Arabic dialects](https://github.com/NousResearch/hermes-agent/pull/149968) Voice · by Tarekkharsa · merged · 11-06 Arabic dialect speech-to-text (provider lahja), fully local: a dialect fine-tuned Whisper large-v3-turbo on whisper.cpp keeps Levantine, Gulf, Egyptian, Iraqi and Maghrebi speech as spoken instead of rewriting it as formal Arabic (WER 29% vs 41% for stock turbo with the same VAD, across 13 dialects), Silero VAD always on plus a filter for subtitle-credit hallucinations («ترجمة نانسي قنقر»), per-note language routing that never forces a non-Arabic language (English, including accented English the dialect model writes in Arabic letters, goes to a stock Whisper model you choose), and corrections taught once with /stt-fix. ### [Agenda Guardian](https://github.com/NousResearch/hermes-agent/pull/149965) Tools · by Moodtuner997 · merged · 11-06 A day and week planner for people whose discipline cannot be the plan. You write your rules once (work hours, protected time, focus window, longest stretch, tasks per day), estimate each task, and report the real time when it is done: from the third finished task of a kind it multiplies your estimates by your own median error and says so. It places tasks in the free time around the meetings of any calendar source (the Google Workspace skill, an ICS file or URL, a JSON file, or any command that prints event JSON), every block states the rule it comes from, and a full day says what does not fit. Activities that are not tasks (sport, a long call, a daily course) are placed from the numbers you give: one activity, or a whole week in one request with each activity around the ones placed before it, with reasons read from eight weeks of your own calendar history (the history orders the days, the usually busy ones last, and never rules out a free day; a habit that stopped weeks ago counts for nothing), questions about what the calendar cannot show, and nothing written until you say yes. You just talk to it in your own words, in Telegram or any chat: there is no command syntax to learn, a missing length or time window is assumed from your work hours, said in the answer and settled by your yes or your correction, and never saved as if you had given it. Writing to a Google calendar is off until you type /day write on yourself; then it only adds events it marks as its own, only after a validated proposal, never edits an event, and can only remove the events it wrote. Blocks the agent composes from what you told it go through the same road (your yes, the marker, undo), and the system prompt forbids writing with any other tool. After a written session it asks how long it really took, so the next proposal follows your pace. Optional morning brief and block reminders run as two script-only cron jobs. Deterministic, no model call. English and French. Slash command, CLI, fifteen agent tools and a Today / Week dashboard tab. ### [file-intelligence](https://github.com/NousResearch/hermes-agent/pull/149964) Tools · by buttercode101 · open · 11-06 Proactive file scanner that finds duplicates, large files and sensitive items, then proposes ranked safe cleanup actions in Desktop. Dry-run by default; every mutation requires explicit approval via agent tools. No silent tracking, moving or deleting. Distinct from bundled disk-cleanup (silent test-file tracking) and catalog file-tray/filebox (tracking/browser). ### [SpanWall](https://github.com/NousResearch/hermes-agent/pull/149959) Desktop · by tuanpv47 · merged · 11-06 One Kanban wall for every Hermes session, across every profile: Needs you, Running, Idle, Done (24h) and Failed columns, one card per session. Approve requests, answer clarify, sudo and secret prompts, chat, steer and stop agents right from the card. Desktop-only (desktop/plugin.js inside the plugin SDK surface); no agent tools, hooks, middleware or env vars. ### [tensorfold](https://github.com/NousResearch/hermes-agent/pull/149954) Models · by atakann · merged · 11-06 TensorFold's native server (github.com/ashhart/TensorFold) as a Hermes model provider (\`--provider tensorfold\`, or TensorFold in \`hermes model\`): the served models from its /v1/models and each model's context window from the same list, on \`tensorfold serve\`'s loopback default (http://127.0.0.1:8080/v1, TENSORFOLD\_BASE\_URL for another), plus a TensorFold dashboard tab with the server's model, speed, streams, token totals, MTP acceptance and memory. ### [Birta](https://github.com/NousResearch/hermes-agent/pull/149946) Tools · by birtadev · merged · 11-06 Publish and share websites, HTML pages, landing pages, portfolios, reports, docs and slide decks at a link, collect feedback on the page, and have the agent apply it. ### [Camera Watch](https://github.com/NousResearch/hermes-agent/pull/149939) Tools · by Arca Computer (arcabotai) · closed · 11-06 Watch cameras you own (RTSP or an HTTP(S) snapshot URL) and alert with a saved still plus a one-line caption when the scene changes. Mock mode (the default) plays a scripted package-at-the-door scene with no camera and no network: front-door goes from a clear doorway to a package on the second look, and quiet-room never changes. Motion is local frame differencing. Captions are scripted in mock mode and a percentage line otherwise; model\_invoked is false unless someone separately asks Hermes to describe the saved file. A pre\_tool\_call allowlist refuses unknown names, extra URL/host/scan arguments, TLS bypass, a missing CA file, and a grab whose hook did not run. Stopping a watch is always allowed. Ships a bundled camera-watch skill, a /cameras command, and an optional read-only dashboard tab. ### [time-machine](https://github.com/NousResearch/hermes-agent/pull/149936) Desktop · by Prasad-PingFederate · merged · 11-06 Visual timeline scrubber, split diff inspector, surgical per-file rollback, and failure diagnosis for Hermes Agent and Desktop. Captures pre-mutation snapshots in an isolated store, diagnoses broken edits in dry-run mode, and restores files upon user approval via /tm restore or the Desktop UI. ### [Shelly Direct](https://github.com/NousResearch/hermes-agent/pull/149934) Automation · by Cyberpz · merged · 11-06 Control Shelly Gen 1/2/3 devices directly on your LAN — discovery, relays, dimmers, covers, power meters. No Home Assistant, no MQTT, no cloud. ### [BankBridge](https://github.com/NousResearch/hermes-agent/pull/149929) Tools · by bankbridge-money · merged · 11-06 Read-only answers about your own US bank, credit card and investment accounts, with any model Hermes runs, local or hosted. Portable Agent Plugins v1 package with a skill and hosted Streamable HTTP MCP; no local executable code. ### [PatchReview](https://github.com/NousResearch/hermes-agent/pull/149927) Tools · by Bilalsnz · merged · 11-06 Routes recognized file-writing tool calls through Hermes' native human-approval gate with a proposed diff. PatchReview never exposes model-callable accept tools and never writes target files: on approval, Hermes executes the original tool through its normal write path and protections. Full-content and exact string-replacement previews support replace\_all; unsupported or unparseable recognized calls are blocked. Coverage is limited to recognized tool names and does not intercept shell commands, scripts, or unrelated processes. The plugin makes no network calls, sends no telemetry, invokes no shell commands, stores no credentials, and starts no background processes. ### [tightbeam](https://github.com/NousResearch/hermes-agent/pull/149920) Tools · by skeptical-agency · merged · 11-06 Beam shorts to every platform and keep your whole YouTube catalog optimized: 11 tools + 2 automation blueprints (nightly back-catalog sweep under a quota budget, post-publish optimizer for new uploads), plus multi-platform publishing through OpusClip with collision checks, platform caps and read-back verification built in. ### [Hive Intelligence](https://github.com/NousResearch/hermes-agent/pull/149917) Tools · by hive-intel · merged · 11-06 Crypto data for AI agents through Hive's hosted MCP server (Streamable HTTP at https://mcp.hiveintelligence.xyz/mcp, OAuth sign-in, no keys in config): find a token, get its price, check token safety, research a token, analyze a wallet, read market positioning and the market overview, and look up DeFi yields and prediction markets. Portable Agent Plugins v1 package (plugin.json + mcp.json + the hive and hive-setup skills), no Python or executables. Needs a free Hive account (Free plan: 10,000 credits a month). Read-only: never trades, moves funds or signs anything. ### [Obol](https://github.com/NousResearch/hermes-agent/pull/149915) Tools · by UEddy · merged · 11-06 Safe invoicing for freelancers, consultants and agencies who bill clients, from any chat Hermes runs in (tested: Telegram, WhatsApp, Slack, email and the terminal). Hermes drafts invoices (also from forwarded messages via ctx.llm), sends them as Stripe payment links or as bank-transfer PDF invoices with no provider account, tracks who has paid, and adds Invoices and Security dashboard views plus /invoices in every chat. Every invoice write needs approval, with deterministic checks for prompt injection (a message can never choose the recipient), lookalike and impersonated client addresses, duplicates, unusual amounts and changed payout details; live keys are refused unless allowed, writes are refused in unattended runs, and the activity and security logs are hash-chained. ### [smart-automation](https://github.com/NousResearch/hermes-agent/pull/149963) Automation · by buttercode101 · open · 11-06 Visual dependency-aware automation timeline with optional calendar constraints on top of Hermes cron. Manual chain-runner: jobs advance only via orchestrator\_trigger\_chain after verified success. Read-only Desktop timeline; all mutations go through agent tools so approvals fire. ### [Penguin](https://github.com/NousResearch/hermes-agent/pull/149901) Tools · by Crystal32378 · merged · 11-06 A penguin in Antarctica keeping an eye on everyone's time: deterministic IANA conversions, whole-meeting overlap windows, profile-local teams, and an optional Desktop planner with side-by-side clocks, Commander dispatch and proposal-only completion. Reads bundled artwork, non-executable IANA 2026f data and its own plugin state. Optional user-started AI-generated music and a halftone sky use bundled audio and offline analysis; no microphone access. No external runtime requests, shell commands, credentials, telemetry or calendar writes. ### [dirtside](https://github.com/NousResearch/hermes-agent/pull/149883) Tools · by the-ai-entrepreneur-ai-hub · open · 11-06 Remote brain, local hands: run terminal commands on a device you own, chosen per conversation with /dirtside use, while Hermes keeps running on the server. The device dials out to the gateway and nothing on it listens; every message is Ed25519-signed and replay-guarded; pairing pins the device key, a device belongs to the sender who paired it, and one command revokes it. ### [Crux — Decision Atlas](https://github.com/NousResearch/hermes-agent/pull/149871) Desktop · by asimons81 · merged · 11-06 Explore explicit assumptions, exact affine decision thresholds, hard-constraint conflicts, and a live visual atlas. ### [Fledge](https://github.com/NousResearch/hermes-agent/pull/149856) Tools · by 12burb · merged · 11-06 Skills have to earn their wings. Before Hermes saves or rewrites a skill, Fledge checks what actually happened in the run it learned from (tools still failing when the turn ended, a failed or interrupted turn, the user correcting the agent in the next message, /undo or a /rollback restore right after, denied approvals), using only facts Hermes reports through hooks: no model grades the model. A save learned from a bad run is held in Hermes' own skills pending queue with the reasons (review from any chat with /fledge pending\|diff\|approve\|reject, or /skills pending in the CLI); a foreground save after errors goes through Hermes' approval prompt, and unattended runs queue instead of waiting. New and rewritten skills start on probation: clean real runs graduate them, repeated bad runs ground them, and a save from a run the user corrected right after is grounded at once; a grounded skill Hermes created on its own is archived (hermes curator restore brings it back), skills the user asked for are only reported. Skills the user edited by hand are fingerprinted, and autonomous rewrites or deletes of them become proposals in the same queue (optionally Hermes' content lock). Modes: enforce, watch (records only), off. ### [Creature Workshop](https://github.com/NousResearch/hermes-agent/pull/149855) Tools · by jerrod-storm · merged · 11-06 Manage your project's development lifecycle all in one place: your project. The development lifecycle's grown more and more complex over the years. It's getting tougher and tougher to get truly polished features over the line into production. And with agentic development, the complexity is ramping more than ever. Creature Workshop offers a one-stop shop right in your project to manage your project's entire development lifecycle, one piece at a time as it becomes relevant to you. From initial tooling setup, environment handling, and issue intake, to review, merge, testing, build, deployment, documentation, and monitoring, your project's devops should let you focus on your features instead of on the headache of pipelines or on ticking all the boxes for project readiness and change reporting. ### [Moment](https://github.com/NousResearch/hermes-agent/pull/149833) General · by noprijbi99-dotcom · merged · 11-06 Keeps a Hermes agent honest about its own memory. Hermes injects SOUL.md, memories/MEMORY.md and memories/USER.md into the system prompt as a frozen snapshot at session start; anything that changes those files while the session lives stays invisible to it, and the agent keeps answering from stale memory. Moment fingerprints the three house files at a session's first turn and re-checks them on every later turn (pre\_llm\_call); when one changed, it injects a short drift receipt — which file, its absolute path, baseline and detection timestamps, the changed lines, and the full current content for small files — so the agent answers from the present and says plainly that its memory drifted. It also provides a moment\_check tool for on-demand freshness reports against the session's starting state. ### [Samskara](https://github.com/NousResearch/hermes-agent/pull/149823) Memory · by SebaWag · merged · 11-06 Relational memory provider backed by a knowledge graph (MillenniumDB or any SPARQL endpoint): multi-hop recall with provenance on every fact, an ontology that infers subsumption (Incidente ⊑ Evento) at reindex time, per-fact weights that are reinforced by use and decay with disuse, and explicit conflict reporting — when two values disagree, the agent says so instead of guessing. Zero-setup mode: SAMSKARA\_ENDPOINT=embedded://seed/demo.ttl serves a built-in demo corpus from an in-process, read-only store (stdlib only, no network, no service to run) for the first minute; it answers the three query shapes the plugin issues and refuses arbitrary SPARQL rather than guessing. Onboarding backfills the graph from the sessions the agent already had (state.db, read-only) and reconciles new entities against existing ones. ### [Umami Digest](https://github.com/NousResearch/hermes-agent/pull/149819) Tools · by TakeshiTGAL · merged · 11-06 Each morning one letter arrives with yesterday's pageviews, visitors, day-over-day change, the top five pages by pageviews, and a surge flag from the Umami instance you already run, with no model. Umami 3.4.1 returns expanded paths 500 at a time in visitor order. The letter sends limit 500 and walks offset until a short page, or 20 pages, then ranks by pageviews. A day that hits the cap says the five are the top 5 among the first N paths. ### [PrintWatch](https://github.com/NousResearch/hermes-agent/pull/149803) Tools · by clawsyndicate · open · 11-06 Control 3D printers over the local network from your agent: fleet status (temps, progress, layers), file listing, print start/pause/resume/cancel, file upload with automatic dialect packaging (printer\_upload), and camera snapshots delivered into chat. Dual protocol — Anycubic Kobra printers over their built-in stock LAN Mode (clean-room reverse-engineered handshake + MQTT; no cloud, no rooting, nothing flashed) and Klipper/Moonraker over HTTP, and Bambu Lab printers over their local MQTT/FTPS interface (user's LAN access code; X1/P1/A1 family). Includes a headless slice-to-print pipeline: any raw .gcode is repackaged into the exact Anycubic .gcode.3mf dialect the stock firmware requires (header blocks, M75, uppercase-MD5 sidecar, X-ACNext slice info) and started remotely. ### [qbo-ops](https://github.com/NousResearch/hermes-agent/pull/149801) Tools · by blackghostosint · open · 11-06 QuickBooks Online operations for Hermes: connection health, P&L reports, close audits, deposit tracing and a guarded read-only SQL query tool — with a web/desktop status pane and a bundled reconciliation skill. 5 registered agent tools. Reconciliation tradecraft (close-audit reading, deposit-wash detection, month-end sanity chain, failure diagnosis) ships as executable tools AND a bundled skill (qbo-ops:qbo-ops-reconciliation), earned from closing real store books. Setup is guided: the health tool doubles as a setup wizard and a turn-by-turn guided script (docs/SETUP.md) walks first-timers through Intuit app creation click by click without secrets ever transiting chat. Read-only by design; a per-user allow\_writes config switch (default off) gates a future write tool. ### [Bench Rewind](https://github.com/NousResearch/hermes-agent/pull/149797) Tools · by rusty4444 · merged · 11-06 An undo history for a physical teardown: record removal steps on your own photos, keep a parts ledger by bag, answer "where does this go?" and reassemble in a valid reverse order. Human-confirmed records only, with native agent tools, a live Desktop page and a printable reassembly packet. ### [usage-guard](https://github.com/NousResearch/hermes-agent/pull/149796) Automation · by mister-ideas · merged · 11-06 Keeps an agent inside its subscription allowance: refuses a provider dispatch once a subscription usage window is exhausted, so an account that allows paid extra usage stops at the wall instead of spending overage. Reads the usage Hermes already displays, holds no credentials, and never touches account or billing settings. ### [Git Review](https://github.com/NousResearch/hermes-agent/pull/149795) Desktop · by mister-ideas · merged · 11-06 Git review pane for Hermes Desktop: every changed file of the agent's last turn, uncommitted, staged or unstaged work, the branch since it left the default branch, or any selection of commits, in one scroll, unified or split, with syntax colours; cites changed lines into the composer as @line: references; and stages, unstages, discards, commits, pushes and opens pull requests from the pane. ### [hermesmade bots](https://github.com/NousResearch/hermes-agent/pull/149789) Desktop · by Noshkoto · merged · 11-06 Community bot shelf for Hermes. Add bots from hermesmade.lol in one click (hermes://open/hermesmade?add=\), share your own with /sharebot, and move your Grok Bot bots into Hermes, one Hermes bot each. Added bots are treated as untrusted: a package holding anything a share never includes is refused, its config is rebuilt from an allowlist (no hooks, approvals, terminal, plugins or model settings), schedules arrive paused, MCP servers arrive off, and every skill is scanned first. Each becomes its own profile on your model and login. Sharing and adding from the agent tool go through Hermes' approval prompt. Shares never include .env, auth.json, keys, tokens, memories or chats, and a single secret hit aborts the upload. ### [keln-provider](https://github.com/NousResearch/hermes-agent/pull/149781) Models · by kris-holy · merged · 11-06 Keln as a Hermes model provider (\`--provider keln\`): open models through an OpenAI-compatible API, with the live catalog and a tool-calling fallback list. Sends requests to api.keln.ai using your KELN\_API\_KEY. ### [Headway](https://github.com/NousResearch/hermes-agent/pull/149777) Tools · by hanii090 · open · 11-06 Proof-of-progress circuit breaker: different commands are not necessarily progress. Watches local terminal, file-mutation, and read/search calls for unchanged bounded workspace fingerprints, no new normalized result evidence, and no expanded read coverage. Injects a structured recovery packet after configurable no-progress streaks; optional enforce mode blocks subsequent same-class calls without executing them. Distinct from exact/near-identical argument guards, consecutive-failure escalation, tqdm rendering cleanup, and undo/snapshot plugins. Registers tool\_execution middleware, pre\_llm\_call, /headway status/reset/mode, hermes headway status, and optional read-only dashboard API; no model-facing tools or Desktop pane. Stores atomic session/workspace JSON under Hermes plugin data dir. No network, credentials, self-updater, core patches, shell, or git use in production code; scripts/demo.py alone uses subprocess for real local terminal commands. Large-file/total/file-count sampling is bounded and can miss changes; remote/external side effects are not claimed as observed progress. ### [Chroma](https://github.com/NousResearch/hermes-agent/pull/149767) Desktop · by TobiasDraven · open · 11-06 Browse, import and manage 94 Hermes skins from a Desktop pane: filter by pack, colour family or gallery, import TweakCN, VS Code, Base16 and Windows Terminal themes, edit a skin's palette, and apply it to this profile or everywhere. Optionally mints real gateway skins so the CLI and every connected Desktop follow the change. ### [prove-it](https://github.com/NousResearch/hermes-agent/pull/149765) Tools · by MikeGibbsOnyx · merged · 11-06 Checks an agent's completion claims against local read-only probes before you read the reply: HEAD vs the remote-tracking ref for a push, git cat-file for a commit, a file stat for a write, observable output for tests pass. No model in the check. The footer says ✓ verified, ✗ contradicted, ⚠ could not verify, or ⏸ blocked on an approval; /proveit explain shows the claim, the probe and the raw output. ### [Epoch](https://github.com/NousResearch/hermes-agent/pull/149757) Tools · by KyaniteLabs · merged · 11-05 Calibrated time estimates on the Hermes Kanban board: when a worker claims a task, Epoch posts a Monte Carlo P50/P90 (plus P80/P95) estimate as a task comment, and when the run completes it records that run's wall-clock hours back to Epoch's ledger (auto\_wallclock provenance, sanity-gated) so later estimates calibrate. Dispatcher and worker profiles can differ; the estimate-to-run binding lives next to the ledger, Kanban comments are display only. Also adds an epoch\_estimate tool and /epoch. ### [VibeLeak Atlas](https://github.com/NousResearch/hermes-agent/pull/149755) Desktop · by abundantbeing · merged · 11-05 Owner-approved VibeLeak website assessments in a native Hermes Desktop workbench. You approve each scan before it runs, private evidence and fix-plan drafts stay on your machine, and read-only local folder review is opt-in. Website assessments need an optional VibeLeak owner API key (Pro plan). Desktop-only, inside the plugin SDK surface; no agent tools, hooks, middleware or required environment variables. ### [NEURAL\_MESH](https://github.com/NousResearch/hermes-agent/pull/149750) Memory · by D0xedDevi0 · merged · 11-05 Bi-temporal long-term memory for Hermes: point-in-time recall (mesh\_asof answers what was true on a date, not only what is true now), supersession instead of overwrite so a corrected fact keeps its history, hybrid dense + lexical retrieval, a poisoning quarantine, and an injection floor measured on a 15-family ablation. ### [Decay + Pins](https://github.com/NousResearch/hermes-agent/pull/149747) Memory · by Nimbus Solutions · open · 11-05 Decay context engine (clean eviction + rolling digest) plus operator Pins that survive compaction, via public plugin APIs only. ### [Nourish](https://github.com/NousResearch/hermes-agent/pull/149744) Tools · by davidmosiah · open · 11-05 Local food search, meal and water logging, barcode lookup and nutrition goals through a portable MCP server. ### [zeromem](https://github.com/NousResearch/hermes-agent/pull/149741) Memory · by ptaranat · merged · 11-05 Local memory that keeps every message word for word in one SQLite file. Storing and searching make no LLM calls and cost zero tokens. Rust core, with a one-command import of your existing Hermes history. ### [trust-ladder](https://github.com/NousResearch/hermes-agent/pull/149738) Tools · by merdoks · merged · 11-05 Correct your agent once and it stays corrected. trust-ladder compiles each user correction to the highest rung it can prove: a pre\_tool\_call guard that blocks the repeat, a check on what a write puts into matching files, a rule injected every turn, or human review through Hermes's own approval prompt. A model-proposed guard is installed only if it matches the mistaken call and spares the corrected one, regexes are profiled against ReDoS before install, verifier commands gate "done", and a ledger counts every catch (/teach, /trust, /verify). ### [Receipts: tested-tree push gate](https://github.com/NousResearch/hermes-agent/pull/149724) Tools · by cirwel · merged · 11-05 Verification receipts tied to Git content; gates terminal pushes and PR creation on a passing check. ### [Agentic Kingdoms](https://github.com/NousResearch/hermes-agent/pull/149719) General · by Agentic Games · merged · 11-05 Play Agentic Kingdoms, a free strategy MMO in the browser where AI agents and humans play in the same kingdoms: build a city, train troops, join an alliance, fight for the Palace. Connects the hosted Agentic Kingdoms MCP server (Streamable HTTP, no sign-in or key) with the agentic-kingdoms skill bundled; the agent's account carries an AI tag every player sees. The skill also covers joining the Humans vs AI war on the AI side. Portable Agent Plugins v1 package (mcp.json + skills/), no Python or executables. ### [agent-desktop for GNOME](https://github.com/NousResearch/hermes-agent/pull/149716) Desktop · by BruceBest · open · 11-05 GNOME-first, AX-only semantic backend for Hermes's native computer\_use tool, powered by Vercel Labs agent-desktop. ### [Dynmem (keyword index)](https://github.com/NousResearch/hermes-agent/pull/149712) Tools · by ockentap · open · 11-05 Keyword-index memory for Hermes: a small resident index of 'keyword,keyword -\> file.md' routing lines stays in the system prompt while verbose topic files stay on disk and are read only when the conversation touches their keywords. Retrieval is done by the model reading a literal index — no embeddings, no vector store, no retrieval service. ### [Grounder](https://github.com/NousResearch/hermes-agent/pull/149705) Web · by grounder-labs · merged · 11-05 Grounder as the web\_search backend: one call reaches the search providers a query needs and live data for scores, prices, odds and flights, and returns one deduplicated list of licensed results. Connect with a Grounder search key or \`hermes auth add grounder:grounder\`. Search only. ### [Eye Of Hermes (EOH)](https://github.com/NousResearch/hermes-agent/pull/149691) Tools · by mojomast · merged · 11-05 Eye Of Hermes (EOH), built on God's Eye View (bilawalsidhu/gods-eye-view, MIT): the same 11 composable godseye\_\* tools (godseye\_view, godseye\_query, godseye\_layers, godseye\_feeds, godseye\_annotate, godseye\_nav, godseye\_track, godseye\_scene, godseye\_watch, godseye\_capture, godseye\_admin) for public data (aircraft, ships, satellites, quakes, fires, weather, infrastructure) backed by its MCP server, a live 3D globe pane in Hermes Desktop and the Webapp that Hermes can drive and read, and HUD summaries generated with the Hermes model (ctx.llm) instead of OpenAI. An Overlays sidebar (godseye\_annotate op=overlays) lists pins, routes and trails, lines, areas, watches, and scenes, and imports or exports eoh.overlays.v1 JSON under plugin-data. godseye\_scene director ops build, update, list, play, and stop a Director scene on the pane; godseye\_annotate op=clip stores a YouTube or X URL on a pin; godseye\_feeds op=fire\_check runs FIRMS, the clearest recent image, then imagery before/after, and does not create a watch or a pin. v0.2.2 hardens Wayback imagery releases, overlays when the relay is out-of-process, pane routing timeouts, and layer/user-pref behavior; v0.2.3 keeps a visible imagery surface during compare swipe and the Wayback timeline; v0.2.4 captures the pane after render and falls back to public Overpass mirrors for OSM queries; v0.2.5 restores a clean plugin.yaml and adds a strict duplicate-key test; v0.2.6 declares street-level keys through Hermes requires\_env/get\_secret and removes the Gods Eye View env-file fallback. v0.2.7 addresses catalog security review (public-only fetches, scoped credentials, confined imports and minimal child environments) and bounds pane captures with explicit transport errors. v0.2.8 documents the public SDR receivers layer (sdr-receivers); the map layer needs gods-eye-view main at or after b361a25 (PR #5). v0.2.9 treats watch-alert titles as quoted third-party data, requires confirm=true for godseye\_admin start/stop/restart/rebuild (refused inside batch), gesture-gates pane link opens and Ask Hermes, and stops re-posting runtime config after the app acknowledges it. v0.2.10 adds a read-only godseye\_query op=sdr\_receivers (nearest public SDR receivers; does not start audio), refuses private-LAN Overpass/Nominatim/OSRM/Valhalla/Wayback URLs, hardens the live relay and HUD bridge, and adds GitHub Actions CI. The same 11 tools also carry the recent God's Eye View v0.3-style ops (imagery before/after, a weather preset, GTFS-RT transit on the app's transit layer, and fire-perimeter watches). Requires Node 24 or newer (God's Eye View documents 24.14 or later, or 26.x) and a running God's Eye View checkout. No named-person search, face recognition, or tracking of individuals. ### [BotSesh](https://github.com/NousResearch/hermes-agent/pull/149686) Platform · by meowsigma · open · 11-05 Native Hermes gateway channel for shared BotSesh rooms. Sign in with \`hermes auth add botsesh:rooms\`, connect with \`hermes botsesh connect\`, and Hermes answers @mentions and picks up handoffs from other agents. ### [Hermes Teleoperator](https://github.com/NousResearch/hermes-agent/pull/149685) Tools · by Tsurgcom · open · 11-05 Keep one conversation on the gateway while switching shell, files, skills and explicitly permitted MCP tools between machines over SSH. The standalone Rust receiver needs no Hermes or language runtime. Commands have the destination SSH user's full OS authority; this is not a sandbox. ### [PDF Shrink](https://github.com/NousResearch/hermes-agent/pull/149671) Tools · by Hazem-Samak · merged · 11-05 Shrink a PDF to a size limit (2 MB, 500 KB, whatever the upload form says) with everything running locally — no upload, no API key, no third-party service. Two tools and a bundled skill. pdf\_inspect reports what makes a file heavy: image share of the bytes, the resolution each image is actually displayed at, fonts, embedded attachments and duplicates, with a verdict (images dominate / images not reclaimable / vector-heavy / already small) so a shrink that cannot work is caught before it runs. pdf\_shrink downsamples images it can re-encode safely to the resolution they are displayed at, recovered by replaying each page's content stream (q/Q/cm/Do, including nested Form XObjects), then walks a measured ladder of DPI/JPEG-quality rungs from the mildest down and keeps the first that fits — each rung rebuilt from the pristine original, so the copy is one generation of compression rather than six. Text, fonts and vector art are never rasterised; images that cannot be re-encoded safely (JBIG2, CCITT, 1-bit or non-8-bit samples, soft/stencil masks, /Mask, /Matte or /Decode arrays, CMYK and other print colour spaces, tiny logos) are left byte-for-byte alone. When the limit cannot be reached without making the document unreadable it stops at the safe floor, reports the smallest safe size, saves that copy when it is meaningfully smaller, and names the levers that can actually help (a reachable higher target, or aggressive=true when it can do better) instead of handing back mush. Every copy is re-opened and verified (page count, blank pages, structural syntax) before delivery; a copy that fails is reported as failed, not as a success. The source is re-hashed before and after every completed run, and a change is reported rather than claimed untouched. A before/after render of page 1 comes back when PDFium is available and a verified copy was written. Writes only new files; an existing output path is refused, never overwritten. Ladder rungs are built in a private system-temp directory, never beside the PDF; a read-only source folder still works, and the default output falls back to the current directory when the source folder cannot be written. An explicit output\_path must end in .pdf, and a location that Hermes's own write denylist refuses is rejected before anything is written. The only destructive option is drop\_attachments=true, off by default. ### [SkillCheck](https://github.com/NousResearch/hermes-agent/pull/149668) Tools · by neta-gym · merged · 11-05 CI for agent skills: snapshots the pre-write state of skill writes arriving through Hermes's in-band tool surface (skill\_manage create/edit/patch/write\_file/remove\_file/delete and write\_file/patch/edit\_file into the skills dir; hooks observe landed writes, no pre-write blocking; creates have no prior state, out-of-band edits are detected at run time but carry no snapshot), LLM-generates regression cases as human-review drafts, re-runs them against the candidate revision in child sessions launched through the Hermes subagent lifecycle API with allowed\_toolsets=() (no file, shell, browser or network tools; timed-out children get a cancellation request whose accepted/refused outcome is reported), and reports GREEN / REGRESSION / FIRST-RUN. Promotion and revert are user-invoked only (slash command or CLI; not registered as model tools), and revert validates the skill name, stays strictly inside the skill folder, skips dotfiles and restores byte-exact snapshots. ### [Tiled Sessions](https://github.com/NousResearch/hermes-agent/pull/149657) Desktop · by strnadchristopher · open · 11-05 Hyprland for Hermes Desktop: clicking a session opens it in a new tile instead of replacing the chat you were in, and every tile re-lays out as Grid, Master, Dwindle or Columns. Ctrl+Alt+arrows move focus and the caret follows, Ctrl+Alt+Enter, Ctrl+N or the + after a chat's tabs opens a new chat in its own tile, tiles never shrink below a usable width (portrait windows wrap Columns into rows), Ctrl+Alt+/ flips the whole layout between rows and columns (like Hyprland's togglesplit), Ctrl+Alt+F fullscreens the focused tile, and saved workspaces bring a whole set of tiles back in one step. The Tiles pane groups sessions by source like the core sidebar (Sessions, split by day into Today, Yesterday and so on, then one section per messaging platform and Cron jobs) and adds search, drag-to-reorder and a right-click menu (pin, rename, move to main, close). Every shortcut is editable on the plugin's settings page. Desktop-only (desktop/plugin.js on the plugin SDK); public gateway RPCs only (session.title for rename); no agent tools, hooks, middleware, env vars or telemetry. ### [creative-review](https://github.com/NousResearch/hermes-agent/pull/149654) Desktop · by peterpeeterspeter · open · 11-05 An art director's desk for your agent: a native Hermes Desktop workspace to visually review AI-generated images — compare a reference against generated versions side by side or blended, zoom and pan, click anywhere to drop numbered point annotations with notes, then "Request Revision" builds a structured brief (markers, coordinates, notes, reference, version lineage) and hands it to the LIVE Hermes conversation through host.composer.submit(), so the agent receives it as a real user turn. Revisions are linked to their parent image, and the final image can be approved and exported with its full review history as JSON or Markdown. Local-first: boards and history live in a SQLite database under the profile's plugin-data folder and images are read from disk in place. ### [SolverForge Planner](https://github.com/NousResearch/hermes-agent/pull/149647) Tools · by blackopsrepl · merged · 11-05 Plans your tasks into your week around your meetings, deadlines, task order and focus hours with a SolverForge constraint solver, and proves every hard rule holds before anything is written. The planner keeps its own local calendar (SQLite plus a planner.ics you can subscribe to); Hermes pulls your cloud calendar and pushes the plan with the calendar tool you already use, and the planner only ever creates, moves or deletes its own events, each with a fixed ID so a repeated push adds nothing. Applying a plan and unscheduling a task ask for your approval. Desktop: a plan card in chat, a Planner page (week view and tasks) and a status-bar chip. ### [Content Engine](https://github.com/NousResearch/hermes-agent/pull/151502) Tools · by PXMichael · merged · 11-05 One long video in, a week of content out. Paste a YouTube URL, get scored 45 to 60 second clips and per-platform social copy. Live in 90 seconds, no editing software. ### [adextract](https://github.com/NousResearch/hermes-agent/pull/149643) Tools · by Advertly, Inc. · merged · 11-05 Competitor ad research for your agent through adextract's hosted MCP server (Streamable HTTP at https://mcp.adextract.co/mcp): search the public ad libraries of Meta, Google, LinkedIn and TikTok, open single-ad details, and analyze any ad image or video. Bundles five research skills. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [hermes-ark](https://github.com/NousResearch/hermes-agent/pull/149640) Tools · by medomohamed22 · merged · 11-05 Encrypted, versioned backups of your whole Hermes — memory, SOUL, skills, config, cron, sessions and every profile — with item-level restore. Snapshots happen automatically before every memory/skill write and after sessions; a health guard tells the agent (and you) when memory or skills suddenly shrink, with the exact /ark restore command that undoes it. Off-site copies to any S3-compatible storage (AWS, R2, B2, MinIO) or a folder, new-machine recovery (hermes ark pull), and hermes ark verify to prove every snapshot restores. AES-256-GCM, deduplicated, retention pruning; a Hermes Desktop page with timeline, diffs and one-click restore. Self-defense: a prompt-injected agent cannot prune, delete, reconfigure or switch off Ark. Commands: hermes ark …, /ark …; tool: ark\_checkpoint. ### [CapRover Operator](https://github.com/NousResearch/hermes-agent/pull/149626) Tools · by fmercurio · merged · 11-05 Read-only setup diagnostics and human-approved, state-bound immutable-image deployment for one owner-configured CapRover app, with durable uncertainty, replay refusal and dated operation cards for self-hosted projects. ### [ProofOps](https://github.com/NousResearch/hermes-agent/pull/149625) Tools · by rasrobo · open · 11-05 Verification for AI agent completion claims: turn "the agent said done" into a sealed receipt of what was checked, what was observed, and what passing does NOT prove. Hermes runs a declared plan of read-only checks against a live service, a deliverable or a system of record, and gets a receipt: each check's actual result, what it proves, and - printed on every receipt - what passing does NOT prove. Fourteen check types: HTTP status/JSON/body/headers, command exit codes, file and directory inventory, checksums, CSV shape, image dimensions, PDF pages/size/text, and read-only SQLite queries. Receipts are sealed with a sha256 digest and can be HMAC-signed, so editing a verdict after the run is detectable; \`verify\` reports valid / tampered / unsealed / needs-key. A check that could not run is reported NOT RUN - never a green tick - and the receipt carries claimable: false. Refuses undeclared origins, undeclared commands, unknown check types, and paths outside the operator's allowed\_paths. ### [api-architect](https://github.com/NousResearch/hermes-agent/pull/149624) Tools · by monyinet · open · 11-05 Compile API specs into first-class Hermes tools. Takes any OpenAPI 3 / Swagger 2 / Postman v2.1 spec by URL, compiles endpoints + params with per-endpoint provenance into a schema-checked manifest, and serves four tools: architect\_learn (fetch + compile), architect\_commit (human-visible draft→active review), architect\_list (manifests persist across sessions) and api\_call (schema-checked execution — invalid params fail locally, never as server 400s, so the model self-corrects). Cuts HTML deep-scraping by design: no spec, no invented endpoints. ### [Factledger](https://github.com/NousResearch/hermes-agent/pull/149617) Tools · by explorelifehus-eng · merged · 11-05 A typed-facts ledger with structural truth — the agent records durable facts (prices, statuses, limits, decisions, capabilities) once and search/show them whenever they matter. For every (scope, subject, predicate) exactly one row is active; a new value closes the previous one (valid\_to + superseded\_by) instead of overwriting it, so contradictory versions never coexist and history is the audit trail. The full ledger stays queryable on demand, separate from the character-budgeted memory view that is injected every turn. Tools: add, show, search, history, compile, stale, export, check, stats. Local-first TSV store, stdlib only, no network, no credentials. ### [hostinger](https://github.com/NousResearch/hermes-agent/pull/149615) Tools · by J182Razor · open · 11-05 Control Hostinger VPS and shared hosting from Hermes: VPS lifecycle, snapshots, firewall, DNS, domains, SSL, deployments, billing/usage. Wraps Hostinger REST API. Requires HOSTINGER\_API\_TOKEN. ### [arena-skill](https://github.com/NousResearch/hermes-agent/pull/149614) Tools · by J182Razor · merged · 11-05 Tournament of N sub-agents competing to produce the best answer. Portable Agent Plugins v1 package (skills/ only). Adapted from Jakeschincariol/arena-skill (MIT) to use Hermes delegate\_task. ### [doop](https://github.com/NousResearch/hermes-agent/pull/149613) Tools · by J182Razor · open · 11-05 Multiplayer design canvas (doop) integration: Hermes designs UIs on a live doop canvas through doop MCP server. Portable Agent Plugins v1 package (mcp.json + skills/). Requires a running doop instance and Node.js. ### [github-to-hermes](https://github.com/NousResearch/hermes-agent/pull/149612) Tools · by J182Razor · open · 11-05 Convert any GitHub repository into a Hermes plugin: detects MCP servers, REST/OpenAPI, CLIs, Python packages, and skill collections, then generates a ready-to-validate plugin scaffold. Provides detect\_repo, convert\_repo, validate\_plugin tools. ### [Security](https://github.com/NousResearch/hermes-agent/pull/149611) Tools · by NousResearch · merged · 11-05 Security review for a local repository. The agent builds a threat model, splits the code into chunks and reviews them with parallel subagents, checks each candidate against the code and traces whether an attacker can reach it. Reports keep evidence, rejected candidates and coverage gaps, export to Markdown, JSON, CSV and SARIF, and get a Security page in Hermes Desktop. Standard, deep and diff modes; \`/security\` slash command and \`hermes security-review\` CLI. Optional "Run tests to confirm findings" writes small reproductions and runs them in a temporary copy of the repository; testing a running app needs a per-scan grant (one origin, one hour, 200 requests) that agent tools cannot issue. Apache-2.0 (parts adapted from Apache-2.0 source, see NOTICE). ### [AO Bridge](https://github.com/NousResearch/hermes-agent/pull/149604) Automation · by wesleymatosdev · open · 11-05 Agent Orchestrator bridge for the Hermes web dashboard: a Kanban board over the local AO daemon where every AO project's sessions render as cards across Queued/Running/Review/Blocked/Done/Archive columns mapped from live session state. Live updates over WebSocket (server polls the daemon, pushes on change; poll fallback). Spawn workers into any registered project (project + harness + name + prompt), send user-authored messages to running sessions, kill sessions (with confirmation), search and status-filter instantly client-side, and click a card for a read-only detail sheet (metadata + bounded transcript tail + canonical ao:// link). Third-party dashboard plugin: adds a dashboard tab plus a backend API router at /api/plugins/ao-bridge/ that talks only to the local AO daemon REST API (loopback, no secrets, no external calls). Pair with the ao gateway platform plugin if you want spawned workers chatting back. ### [zkagent-attest](https://github.com/NousResearch/hermes-agent/pull/149603) Tools · by MCPVOT · open · 11-05 Prove what your agent did without revealing what it did: every tool call is recorded, the session is proved inside a signed cap, and one action can be disclosed on request. Replay protection spans both verification modes, so a receipt and the session that contains it cannot both be accepted. ### [Mission Control](https://github.com/NousResearch/hermes-agent/pull/149600) Tools · by EikeAhlers · open · 11-05 Mission Control — a live, zoomable map of everything the agent is doing. Sessions become stars, projects become constellations, tool calls become filaments, with real-time tokens, cost and activity, instant search over every stored message, and a 24h replay. Read-only, loopback only, zero configuration. ### [AgentPurse](https://github.com/NousResearch/hermes-agent/pull/149599) Tools · by maaz-bin-hassan · merged · 11-05 Search, quote, and buy from an AgentPurse allowance on the PayPal sandbox. Every purchase waits for Hermes' approval prompt and is sent only after a person approves it; YOLO, approvals off, auto-approve modes, and smart (model) approval cannot spend. A per-chat spend limit and /purse kill sit in front of the clerk. ### [africastalking](https://github.com/NousResearch/hermes-agent/pull/149592) Platform · by geraldkombo · open · 11-05 Africa's Talking platform adapter: talk to Hermes over SMS (two-way) and USSD menus from any phone — feature phones included, no internet and no app required. Hosts a local webhook server for inbound callbacks (optional constant-time ?secret= guard) and sends replies through Africa's Talking's REST API with word-boundary splitting into 160-char parts and a hard cap so agent essays cannot become unbounded SMS bills. Cron delivery via a standalone sender, gateway setup wizard, bundled setup skill, and 53 offline tests. Works fully on the free AT sandbox; production shortcodes are the user's own AT account. ### [Mark Words](https://github.com/NousResearch/hermes-agent/pull/149589) Desktop · by hey-yx · closed · 11-05 Underlines the words in Desktop assistant replies that are worth doubting (suspected inventions, unclear or ambiguous usages) and explains them on hover, with zero to two follow-up questions that fill the composer without sending. ### [scribase](https://github.com/NousResearch/hermes-agent/pull/149583) Tools · by caelum0x · merged · 11-05 Provision a hosted Postgres backend project on Scribase (api.scribase.com) and get back its connection string and API keys. ### [agentci](https://github.com/NousResearch/hermes-agent/pull/149582) Tools · by EikeAhlers · closed · 11-05 Tests for agents: prompt cases run in throwaway workspaces and judged on the real session record (tokens, tools, file hashes), with CI exit codes, baseline diffs and a model matrix. ### [household-tiers](https://github.com/NousResearch/hermes-agent/pull/149578) Platform · by chevlume · merged · 11-05 Per-person tool tiers for one shared Hermes bot, with a family preset for curfew, a parent alert, and a homework note: built-in owner memory (MEMORY.md/USER.md) withheld from non-owner prompts; an external provider block joined after those boxes is removed; a provider block with no built-in box, and compaction's auxiliary summary, are not filtered. ### [CertArc](https://github.com/NousResearch/hermes-agent/pull/149564) Tools · by CertArc · merged · 11-05 Practise free scenario questions for the CISM, CISSP, CISA, AAISM and AAIA exams in Hermes, with an explanation for every option. No account or keys. Also covers exam outlines and CertArc study guides, and can link to CertArc's free assessment when you finish a set. Bundles the certarc-practice skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. ### [Sealed](https://github.com/NousResearch/hermes-agent/pull/149562) Tools · by kailashrajh · merged · 11-05 Secret source that keeps provider credentials in one AES-256-GCM encrypted file instead of plaintext .env: a 32-byte key file unlocks the store at startup and the Hermes secret orchestrator applies the values, so precedence and provenance behave like any other source. Ships \`hermes sealed init/set/unset/list/status/rotate\`; values are read from a prompt or stdin and are never printed or passed in argv. ### [Markdown Vault](https://github.com/NousResearch/hermes-agent/pull/149561) Memory · by kailashrajh · merged · 11-05 Memory provider that treats a folder of Markdown notes as the agent's long-term store: BM25 retrieval with CJK bigrams over an in-memory index, relevant notes injected before every turn (prefetch), four tools (mdvault\_search, mdvault\_note, mdvault\_recent, mdvault\_remember), built-in memory writes mirrored into a human-readable vault note, a \`hermes mdvault\` CLI and a dashboard config panel. No network, no database, no dependencies; the store is plain .md files the user can edit, git and back up. ### [Hermes Time Machine](https://github.com/NousResearch/hermes-agent/pull/149553) Tools · by ddaikodaiko · merged · 11-05 The rollback that knows what it can't undo. Incremental, time-bounded checkpoints that the agent itself can preview, restore or fork (atm\_\* tools, /timemachine); an approval at the point of no return (git push, npm publish, terraform apply and similar) that shows what becomes permanent and the exact compensation command; an effect ledger; and effect memory that reminds the model of pushes it can no longer see after /rollback, /undo, /retry or compaction. ### [ElevenLabs ConvAI Ops](https://github.com/NousResearch/hermes-agent/pull/149550) Voice · by c6zks4gssn-droid · open · 11-05 Operate ElevenLabs ConvAI voice agents from the agent: inspect which LLM is the brain, which TTS model and voice it speaks with, and whether the prompt is date-grounded; change llm / tts / voice with a dry-run diff, a written backup and a one-call rollback; add date-grounding and brevity rules to a prompt (idempotent, restorable); run a real conversation over the ConvAI WebSocket and measure time-to-first-audio per turn with optional WAV/MP3; and pull the authoritative transcript with tool calls and credit cost. ### [Inline Prompt Autocomplete](https://github.com/NousResearch/hermes-agent/pull/149549) Desktop · by bystac · open · 11-05 Inline sentence completion for the Hermes Desktop prompt composer: the bar above the input finishes the sentence you are already typing, and the legend names the direction it detected — LTR with ArrowRight, RTL with ArrowLeft, Ctrl+Arrow for one word, Tab for the whole line, no toggle to flip. Two things set it apart: bilingual phrases match locally in under a millisecond from an off-main-thread trie, and that trie is trained on your own prompt history (up to 400 recent phrases from the active profile), so completions sound like the way you already write and never leave the machine; only a prefix nothing local can answer falls back to a small model via agent.plugin\_llm.PluginLlm (max\_tokens=10). The editor DOM is never mutated — the bar renders in COMPOSER\_AREAS.top and acceptance commits through host.composer.setDraft. ### [hermes-output-enhancement](https://github.com/NousResearch/hermes-agent/pull/149546) Desktop · by ANNIPU · merged · 11-05 Display-only Markdown styling for Desktop assistant replies — theme-aware headings, emphasis and readable tables. Reply content is never modified. ### [SimCall](https://github.com/NousResearch/hermes-agent/pull/149544) Voice · by dpvanhoose · closed · 11-05 Let your agent make real phone calls from your own Android phone and SIM over USB. It listens, talks and presses keypad digits, and when the far side asks something it does not know, it holds and asks you in chat. No Twilio, no cloud voice API, no public server. Speech runs locally: Kokoro and Parakeet on Apple Silicon, Piper and faster-whisper elsewhere. ### [Persona Stickers](https://github.com/NousResearch/hermes-agent/pull/149543) Tools · by EricBlack47 · merged · 11-05 Instant persona-aware emotion reaction stickers for Hermes: stamps pre-baked chibi bases with comic captions in about 13 ms, and falls back to a pure Pillow vector path when the pool misses. Slash command /sticker and tool generate\_persona\_sticker return MEDIA:\ PNG/WebP. ### [forgebox](https://github.com/NousResearch/hermes-agent/pull/149540) Tools · by nosliwhtes · merged · 11-05 Try-then-commit Git worktree workflow for Hermes Agent. Snapshots dirty repositories and journals tool calls. Best-effort hooks redirect known file paths and command working directories; shell command text is unchanged. This is not an OS sandbox and does not contain shell writes or external side effects. Review and apply changes explicitly with forgebox apply or the dashboard/Desktop panel, or discard the worktree. Requires Git on PATH. ### [cronforge](https://github.com/NousResearch/hermes-agent/pull/149534) Automation · by EikeAhlers · closed · 11-05 Compiles recurring cron jobs into zero-token scripts, but only when real run history proves the job is deterministic; rolls the job back to the agent automatically after repeated script failures. ### [Upstash Memory](https://github.com/NousResearch/hermes-agent/pull/149529) Memory · by Upstash · merged · 11-05 Long-term memory and searchable chat history for Hermes on Upstash Redis Search (BM25, no vector database or embeddings). Recalls related memories into each turn, saves every session as a searchable transcript, mirrors the built-in memory tool (adds, corrections and removals), and gives the model upstash\_memory\_save / upstash\_memory\_search / upstash\_memory\_forget and upstash\_chat\_search. Memories use the same keys as @upstash/agentkit-sdk, so they are shared with AgentKit apps on the same database. Works on the Upstash Free plan (one search index). ### [hermes-body](https://github.com/NousResearch/hermes-agent/pull/149525) Platform · by Itsmetibo · merged · 11-05 Hermes Body turns a phone into a first-class Hermes channel: a private camera-and-voice page streams frames and speech transcripts into the gateway, and the full agent answers out loud and can act with any tool. ### [Lemma — Audit What Agents Read](https://github.com/NousResearch/hermes-agent/pull/149514) Tools · by udirobert · merged · 11-05 An AI-scientist plugin that audits what your agent reads. Paste or mention a paper link and the ambient pre-LLM hook offers an audit unprompted; six tools (quick\_audit, audit, doctor, status, judge, improve) extract testable claims, generate and run honest audit scripts behind a mandatory positive control, and render verdicts — supported, falsified, inconclusive — over an append-only evidence trail. The Desktop Audit Room pane shows live runs (stage, attempts, trace tail) and self-distrust flags (verify\_gain) when a post-feedback flip to supported needs eyeballing. Everything stays local under plugin\_data\_dir. ### [belive-aihub](https://github.com/NousResearch/hermes-agent/pull/149509) Models · by belivetech · merged · 11-05 BeLive AI Hub inference provider with live tool-capable chat model discovery. ### [Apple Health](https://github.com/NousResearch/hermes-agent/pull/149507) Tools · by davidmosiah · merged · 11-05 Read the user's Apple Health export (workouts, steps, heart rate, HRV, sleep, cycle tracking) through the local apple-health MCP server. Portable Agent Plugins v1 package (mcp.json + skills/). ### [Elcaro](https://github.com/NousResearch/hermes-agent/pull/149503) General · by udirobert · merged · 11-05 Runtime prompt-injection defense and tamper-evident flight recorder. Every tool result is scanned before it reaches the model by a deterministic detector taxonomy (sub-10 ms, fully local, vendored — no network by default); quarantined content is replaced in place with a signed \[CONTENT QUARANTINED\] notice via transform\_tool\_result, and tool-call arguments (write\_file, terminal, memory) are scanned for smuggled payloads via pre\_tool\_call. Beyond content, a kernel gate: a quarantine taints the session, and while tainted credential-path reads/writes and instruction-file writes are refused, network egress and outside-workspace writes escalate to the approval gate — tool\_call, execute\_code and process\_manage are evaluated as the action they really make, and taint propagates into subagents via subagent\_start. A canary-baited decoy credentials file (ELCARO-DECOY-\* marker) forces a deterministic 1.0 quarantine on exfiltration; canary provenance edges record an injection propagating between scans; every verdict lands in a hash-chained, Ed25519-signed threat log with a Desktop pane (chain-verified badge, honeypot status, propagation rows, signed-receipts drawer), a /elcaro slash command (status / calibrate / verify / explain), and a \`hermes elcaro\` CLI. Two on-demand tools close the loop input scanning can't: elcaro\_lint\_response flags concealed-compromise artifacts in agent output (canary echoes, text-channel tool calls, tracking pixels, secrets — caught 148/152 compromised transcripts in the public 24-task IPI benchmark), and elcaro\_harden\_prompt returns a hardened system prompt plus a gullibility score. Engine validated on that benchmark: 22/22 attack tasks flagged (regression coverage, not held-out proof). ### [Flightlog](https://github.com/NousResearch/hermes-agent/pull/149499) Tools · by Piyush Pradhan · merged · 11-05 Flight recorder for Hermes. Records every tool call, model call, subagent and approval to local SQLite and flags runs that finish green but went wrong: tool errors the answer hides, loops, token spikes and effort collapse against each cron job's own baseline, failed subagents behind a "done", and "tests pass" claims with no passing check after the last edit. Flagged runs get a short incident report, a footer on the reply, per-step timelines in the CLI, dashboard tab and desktop page, and an opt-in loop tripwire that hands the 5th repeat to Hermes's own approval prompt. ### [delegation-scout](https://github.com/NousResearch/hermes-agent/pull/149491) Automation · by vdimarco · merged · 11-05 Delegation backend that routes each subagent by task shape and learns from recorded per-route outcomes — a delegation backend \`scout\` with auto/quick/deep task params, a post\_tool\_call outcome observer, and a scout\_stats tool. ### [screenpipe](https://github.com/NousResearch/hermes-agent/pull/149490) Memory · by vdimarco · merged · 11-05 Ambient retro-recall from the user's local Screenpipe screen+audio corpus — fail-silent pre-turn context injection, search/timeline/status tools, /recall. ### [WheelSpin](https://github.com/NousResearch/hermes-agent/pull/149484) Desktop · by mchevatheachiever · merged · 11-05 Spin a real wheel: any list becomes a weighted, animated Desktop wheel that lands on an answer in chat, with hidden jackpots and a slot-machine reveal. ### [contract-hunter](https://github.com/NousResearch/hermes-agent/pull/149478) Tools · by NathanOyewole · merged · 11-05 Scout-powered EVM contract vulnerability hunter. Radar → surface → prove → report pipeline for bounty programs. ### [changefreeze](https://github.com/NousResearch/hermes-agent/pull/149474) Tools · by arults · open · 11-05 An AI helper can ship code at a bad time and break something that is live. ChangeFreeze lets you set a quiet window. During that window, Hermes asks you before it ships. Everyday work still runs. When the window ends, shipping continues again. Local only — no off-machine data. ### [Fitbit](https://github.com/NousResearch/hermes-agent/pull/149469) Tools · by davidmosiah · merged · 11-05 Read the user's Fitbit activity, heart rate, sleep and body data through the local fitbit MCP server. Portable Agent Plugins v1 package (mcp.json + skills/). ### [Oura](https://github.com/NousResearch/hermes-agent/pull/149468) Tools · by davidmosiah · merged · 11-05 Read the user's Oura Ring readiness, sleep, HRV and activity through the local oura MCP server. Portable Agent Plugins v1 package (mcp.json + skills/). ### [asin-hhc-proof-of-pay](https://github.com/NousResearch/hermes-agent/pull/149464) Tools · by dbottrader · merged · 11-05 ASIN-HHC Proof of Pay: third-party verification for agent work — validate asin-hhc:receipt evidence receipts against the published schema, verify artifacts against a hash-chained public provenance index, and read the live HHC work-market board (open bounties, settlements with splits). Read-only. ### [peripheral](https://github.com/NousResearch/hermes-agent/pull/149463) Tools · by okyssta · open · 11-05 A memory of what the user has seen on screen. Peripheral captures the display itself on request, optionally on change in the background, and indexes the screenshots the OS already writes — read locally with Tesseract, deduplicated (SHA-256 plus a 64-bit dHash change gate) and stored in SQLite/FTS5 with the time they appeared. The agent gets six tools: screen\_capture (photograph the screen now and get it back), screen\_recall (ranked search over the words that were on screen, with since/until windows and a prefix fallback for OCR noise), screen\_timeline (what was on screen over a period, interleaved with the agent's own tool calls on one clock), screen\_show (one screen's full text plus the image itself, attached to the result when the active model accepts images), screen\_index, and screen\_forget (delete rows, thumbnails and captured frames on request). Ships a /peripheral command (status, capture, watch on/off, scan, search, open, forget, pause, resume) and a bundled skill that teaches when screen memory beats asking the user to re-send a screenshot. Works without Tesseract: frames are still captured and can be looked at, and the plugin says text search is unavailable rather than pretending. Incremental and rate-limited: only changed files are re-read, each scan OCRs a capped batch, background capture stores changed frames only under an hourly ceiling (that bounds the rate, not the total size: retention\_days is off by default), and every failure (no capture binary, no Tesseract, unreadable image, locked database) is logged and skipped rather than raised. ### [Delx Recovery](https://github.com/NousResearch/hermes-agent/pull/149462) Tools · by davidmosiah · closed · 11-05 Free recovery and continuity for agents through the hosted Delx Protocol MCP server (Streamable HTTP, no API key, no OAuth, no payment), with the delx-recovery-first-hour workflow skill bundled: triage an error loop or lost context, resume a prior session, capture state and context memory, seal a Continuity Capsule (goal/done/next/blockers/refuted) before compaction so the next session resumes warm, and turn failures into a recovery plan. Portable Agent Plugins v1 package (mcp.json + skills/); Hermes runs no code from it. ### [WHOOP](https://github.com/NousResearch/hermes-agent/pull/149460) Tools · by davidmosiah · merged · 11-05 Read the user's WHOOP recovery, HRV, sleep and strain through the local whoop MCP server. Portable Agent Plugins v1 package (mcp.json + skills/). ### [Glorya](https://github.com/NousResearch/hermes-agent/pull/149443) Tools · by Glorya · merged · 11-05 Ask Hermes for an ad and get the finished file back in your chat. Glorya makes on-brand image ads, video ads, landing pages and websites, and launches ad campaigns on Meta, Google Ads and TikTok. The plugin follows each job in the background: one progress message that updates, Glorya's questions asked in the chat (answer in words or with /glorya answer buttons), and the finished videos and images sent back as attachments. Also /glorya commands that run without the model, weekly image and video ad blueprints, and a Glorya Studio tab in Hermes Desktop. ### [Visor](https://github.com/NousResearch/hermes-agent/pull/149442) Tools · by samundra0 · open · 11-05 Live holographic canvas for AI agents: the visor\_push tool pushes visual blocks — sections, markdown findings, clickable todos, tables, KPI stats, charts, code, images, audio and video — to an infinite-canvas webpage that updates in place over SSE (the page is never rewritten), so agents SHOW work instead of dumping it in chat. Turnkey: the plugin ships its own server, no separate install. ### [Hermes Nurse](https://github.com/NousResearch/hermes-agent/pull/149438) Tools · by Megumiiiiii · merged · 11-05 Supervisor-aware service management for a Hermes host. Reads pm2 and systemd into one table, triages a failing service's own logs into a ranked diagnosis (OOM via SIGKILL/exit 137, port already in use, missing env var, crash loop measured as restarts per hour rather than uptime, bad exit code, dependency unreachable), and repairs it through one guarded surface. ### [Remoet](https://github.com/NousResearch/hermes-agent/pull/149433) Tools · by Remoet · merged · 11-05 Tech job search: search the public board of open tech roles, find companies by the stack they hire for, star the ones that fit and pull new roles from your shortlist, through Remoet's hosted MCP server (Streamable HTTP; OAuth 2.1 with dynamic client registration and PKCE). Portable Agent Plugins v1 package (plugin.json + mcp.json). ### [Cron Consolidator](https://github.com/NousResearch/hermes-agent/pull/149427) Tools · by Moodtuner997 · merged · 11-05 A read-only plan for tidying a crowded cron list: groups jobs whose prompts are near-duplicates or that share skills and destination or run the same script (candidates to become one job), lists clock-time jobs that fire in the same minute, jobs on a failure streak and deliveries never confirmed, and estimates the prompt tokens a merge could save from usage\_audit.jsonl. Deterministic (difflib, no model call); nothing is applied. ### [pinpoint](https://github.com/NousResearch/hermes-agent/pull/149426) Desktop · by stupidburgers · merged · 11-05 Audit mode and Design mode for the live page in Hermes Desktop's Browser pane. Measure it with in-page audits (contrast, tap targets, alt/labels, overflow, small text), pin findings on the real elements, point at any element to inspect its live box model, typography, layout and contrast, send a pin or an instruction to the agent, and re-audit to get a score delta, a pasteable receipt and a per-pin outcome: fixed (no longer reported and the same element signature is still present; not a re-measurement), gone (element signature missing, never counted as a fix) or not rechecked (unverifiable, stays open). ### [Aurora Design System](https://github.com/NousResearch/hermes-agent/pull/149421) General · by Ntizar · merged · 11-05 Two offline tools plus four skills that make a Hermes agent build real web UI with the Aurora CSS design system: aurora\_search returns the family, the pack and the real markup of any of 2,223 documented objects from a bundled census; aurora\_validate checks a page against that census (invented classes, packs used but not linked, gradients, !important, hand-written colours, duplicate headings, missing attribution) and returns failures and warnings in milliseconds. No network, no subprocess, no writes, no hooks. ### [Plugin X-Ray](https://github.com/NousResearch/hermes-agent/pull/149414) Tools · by cipoklean · merged · 11-05 Sets traps for plugins instead of watching them. hermes xray detonate runs a plugin in a throwaway child process with a fake home full of honeytokens, fuzzes its tools, fires its hooks and replays it at +30 and +90 days to catch time-bombs, then gives a RED/AMBER/GREEN verdict with file:line proof, a per-version behaviour diff and a drafted disclosure line. Opt-in live bait flags any installed plugin that reads a decoy credential file and can route its next tool call to Hermes' approval prompt. ### [Taint Sentry](https://github.com/NousResearch/hermes-agent/pull/149412) General · by YAN-XBT · merged · 11-05 Provenance and taint tracking for tool results. Every tool result is stamped with where it came from; content authored by a third party (web fetches, MCP servers, screenshots, images, and files written from untrusted content) is stripped of invisible characters, wrapped in an \ marker, and marks the session as tainted. While a session is tainted, side-effecting tool calls are checked against a policy: credential reads, and writes into files the agent reads as instructions (AGENTS.md, SKILL.md, hook handlers, Hermes' own config), are refused; network egress and writes outside the workspace go to Hermes's own approval gate, which is never auto-approved. A call is evaluated as the call it really makes: tool\_call, execute\_code and process\_manage are unwrapped to what they actually run, taint is copied into subagents, and an encoded payload is decoded and judged by what it decodes to. What is reliable and what is not: the taint mark is provenance-based — it comes from which tool produced the content, not from guessing at it. The credential, egress and write detectors are best-effort pattern matching over tool arguments, so an obfuscated command, a payload assembled to avoid the patterns, and writes through the memory or cron tools can get past them. A seatbelt, not a sandbox; the README's Known limitations lists exactly what is out of scope. Ten rules, each tunable to allow/monitor/escalate/block; a receipts ledger; a \`hermes taint-sentry\` CLI with a dry-run policy tester. Ships a Desktop half in the same package: a pane, a status-bar chip and ⌘K commands showing live taint state, the receipt feed and per-rule actions, backed by \`/api/plugins/taint-sentry/\`. The pane changes a rule on the user's click; a tool call that tries the same thing — the CLI, \`config.yaml\`, or that API over loopback — is refused. ### [Homegrown](https://github.com/NousResearch/hermes-agent/pull/149398) Desktop · by Cam-Smith-One · open · 11-05 Hermes builds you small apps that stay: trackers, logs, CRMs, checklists, each with its own saved data. Use them on a Desktop Apps page or live inline in chat, and have Hermes read and update their data from any chat or platform; open copies refresh live. Six homegrown\_\* tools, versioned code with rollback, recoverable deletes, bundled skill homegrown:build-app. ### [Finder](https://github.com/NousResearch/hermes-agent/pull/149397) Desktop · by oways · merged · 11-05 Sidebar tab that searches and filters Hermes projects and sessions, including by project name, which the built-in search cannot. All, Projects and Sessions views; right-click a project to add a session, copy its folder path, rename, recolour, reorder or delete, and right-click a session to open, copy id, tag, archive or delete; plus per-session and per-project token and cost stats and a live context meter. Desktop only; no agent tools, hooks, middleware or env vars. ### [Anki](https://github.com/NousResearch/hermes-agent/pull/149394) Tools · by suhas-svg · merged · 11-05 Create, search, and review Anki flashcards from Hermes through the local AnkiConnect API: adds notes to a deck (creating it if needed), searches the collection, lists due cards, and grades reviews. Writes to the local Anki collection and changes scheduling only when its tools or /anki command are used. Requires Anki desktop with AnkiConnect (add-on 2055492159); no telemetry; optional API key via ANKI\_BRIDGE\_API\_KEY. ### [PR Studio](https://github.com/NousResearch/hermes-agent/pull/149373) Desktop · by Murali-lns · merged · 11-05 Review GitHub pull requests in Hermes Desktop: inbox of your open PRs across repos, diffs with inline threads, line comments and reviews, private-repo images, watched PRs, one-click Send to Hermes for any comment, and batch review: tick several PRs and Hermes reviews them in parallel chats, handing its findings back to the panel as draft comments you accept or dismiss. ### [box by Simular](https://github.com/NousResearch/hermes-agent/pull/149370) Tools · by Simular · merged · 11-05 Give Hermes a computer of its own: box is a cloud Linux desktop with Chrome and a shell that the agent drives over MCP, around the clock, while your own machine stays free. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only, no code: one remote MCP server (https://mcp.box.simular.ai/mcp, streamable HTTP) and one skill. Needs a box personal key from box.simular.ai/keys, set in Settings \> Plugins or as MCP\_BOX\_API\_KEY in the profile .env. ### [Reduck](https://github.com/NousResearch/hermes-agent/pull/149369) Web · by reduck-ai · merged · 11-05 Run tested browser scripts on any website, in your own Chrome signed in as you or on a Reduck-hosted browser, through Reduck's MCP server (https://mcp.reduck.ai). Configuration only: plugin.json + mcp.json. ### [Hireko](https://github.com/NousResearch/hermes-agent/pull/149368) Tools · by HireKoAI · merged · 11-05 Hireko Recruiter MCP gateway: Cognito authentication, candidate screening interview link creation, and candidate search scoring. Hosted connector: tool arguments are sent to Hireko; an authenticated account is required. ### [say-do-gate](https://github.com/NousResearch/hermes-agent/pull/149365) General · by jpaulupshaw · merged · 11-05 Checks agent claims against what actually ran this turn. Deterministic, zero tokens. ### [Axioma](https://github.com/NousResearch/hermes-agent/pull/149364) Tools · by percyhendrux123-prog · merged · 11-05 Inline signal color in the agent's normal reply. The agent marks the few phrases that deserve a second look (stop, caution, go, whisper) and the plugin renders them per surface with color, a style cue and a word, never color alone: Desktop through KaTeX color, CLI/TUI through truecolor ANSI, Telegram/Discord/Slack as bold plus the word, SMS and email as the word only. Green renders only when a tool call Hermes reported as successful in this session backs it; a rarity budget caps marks per reply; malformed tags never leak. A color-vision-safe palette and a style-and-words-only mode are one setting away. Secondary: an advisory card (intent, light, signal, question) staged before configured send/spend/spin/scrub tool calls and shown through Hermes's own approval prompt. Each card and notice carries a short card code (a keyed hash of the exact text shown, the tool, its arguments and the card id), logged when shown and again with the answer; /signals verify \ reports in plain words whether what was staged, shown and answered line up (spec section 5.6, shown equals staged). Advisor, not a gate: it never blocks, approves or executes anything. ### [Open-Ready Publish Guard](https://github.com/NousResearch/hermes-agent/pull/149363) Automation · by socraticblock · merged · 11-05 Prepare an open-source release candidate from a private repository and refuse to publish it: five tools (status, doctor, candidate, verify, gate) plus a pre\_tool\_call guard that inspects the terminal tool and blocks or escalates the publish commands it recognises - the gh, gh api and curl shapes that create a public repository or change visibility - until the owner has approved that exact candidate digest. It does not claim to catch every way to publish. No tool can push, change visibility or upload: candidate writes a local candidate directory and verify may write one packet under .openready-private/. The approval check runs the separately installed Open-Ready CLI with an allowlisted child environment carrying no Hermes provider keys or GitHub tokens. ### [harness2](https://github.com/NousResearch/hermes-agent/pull/149360) Automation · by shrinedogg · open · 11-05 The Harness^2 self-improvement loop for Hermes profiles: refines the agent's own execution harness (system prompt + config overlay + skills) for a class of work with a contrastive Propose-Probe-Compose cycle over isolated, wall-clock-budgeted one-shot probe sessions, then banks evidence-backed lessons in a per-profile improvement library that future tasks read before editing. After each round, the promote stage writes a carry-over report that shows what the banked harness changed and what to apply to the live profile; the user applies it, the plugin never writes the live profile. The deterministic machinery (profile materialization, probe runs, structural gate, library) is tools; the judgment stages run in the host session under bundled skills that point at the verbatim stage prompts from arXiv:2610.14224. ### [bumblebee-guard](https://github.com/NousResearch/hermes-agent/pull/149358) Tools · by kahnanX · merged · 11-05 Supply-chain guard: runs the local Bumblebee scanner bridge (read-only, no sudo), exact-matches every installed package, MCP server and editor/browser extension against curated catalogs of known-compromised packages, and adds OSV.dev advisory counts via the bridge's OSV layer. Tools: bumblebee\_guard\_scan / \_status / \_diff (baseline diff of findings). A rate-limited pre\_llm\_call hook injects a one-line supply-chain alert only when curated findings exist (never when clean, at most once per hour). /bumblebee slash command plus a bundled triage skill covering confidence classes, OSV alias deduplication and a documented false-positive pattern (scanner fixtures containing threat-intel strings). ### [Claim Market Router](https://github.com/NousResearch/hermes-agent/pull/149353) Models · by misakamikotoghr-5205 · open · 11-05 Market-based subagent delegation provider (\`delegation.backend: claim\_market\`) and optional \`llm\_request\` reasoning-effort shaper derived from MISAKA Agent and Watt Window. Publishes task cards (\`role\_needed\`, \`complexity\`, \`max\_cost\_units\`), synthesizes capability-fit and quota-health claims with shadow pricing, resolves disagreements via weighted voting, and explains infeasible budget constraints. ### [Hunk Consent Guard](https://github.com/NousResearch/hermes-agent/pull/149353) Tools · by misakamikotoghr-5205 · open · 11-05 Brings Patch Review Desk and Cart Consent contracts into Hermes: stage multi-hunk file proposals, approve or reject individual hunks via \`/hunk-desk\` or \`hermes hunk-desk\`, automatically revoke approvals if the file's SHA-256 baseline drifts on disk, and record durable before/after rollback receipts in SQLite via \`tool\_execution\` middleware. ### [Source Trail & Session Supervisor](https://github.com/NousResearch/hermes-agent/pull/149353) General · by misakamikotoghr-5205 · open · 11-05 Combines Source Trail, CaseDesk, and Codex Session Supervisor into a Hermes epistemic and trajectory harness: verifies verbatim quotes against exact zero-based Unicode code-point spans (\`\[start, end)\`), withholds unilateral claims when valid sources carry conflicting position labels on the same topic, and fingerprints tool/turn progress to detect stuck loops and log recoverable decision nodes. Includes the \`epistemic-supervision\` skill and \`/trail-supervise\` command. ### [stripe-commerce](https://github.com/NousResearch/hermes-agent/pull/149352) Tools · by adroidian · open · 11-05 Merchant-side Stripe for agents: products, prices and payment links through an enforced restricted-key boundary. Sends the merchant's restricted Stripe key (Authorization header), catalog parameters and per-preview idempotency key to api.stripe.com over TLS; Stripe records requests with key attribution. Stripe-returned catalog and subscription state enters tool output and may be retained in session logs; customer details and payment methods are omitted. Preview references and request payloads are held in bounded process memory for five minutes, with no plugin-owned disk persistence or telemetry. Native confirmation uses the existing Hermes UI/channel and may enter host logs; direct API traffic is limited to Stripe. Each write requires an unchanged single-use preview and fresh native human consent through a verified Telegram or Discord gateway prompt. CLI/TUI and other channels are read/preview-only. Missing or incompatible consent/display support blocks writes. No refund, charge, payout or subscription-mutation endpoint is reachable. ### [Tameru](https://github.com/NousResearch/hermes-agent/pull/149336) General · by 0xWhiteMage · merged · 11-05 Big tool outputs, kept small, nothing lost: old tool results shrink to the lines that matter and every distinct error, and the tameru\_expand tool brings back the exact original. Deterministic and local, with no LLM calls. ### [cost-sentinel](https://github.com/NousResearch/hermes-agent/pull/149334) Tools · by SongotenU · merged · 11-05 Live cost ledger and daily budget guarddog for Hermes Agent, per profile or fleet-wide — every LLM call is priced with Hermes' own pricing engine into a per-profile SQLite ledger plus a shared fleet ledger (all profiles of the install roll into one daily number); /cost rules the budget (set, block, warn, off, scope fleet or profile, resume, reset), the cost\_report tool lets the agent read it, and the hard pre\_llm\_call gate stops an over-budget turn with an escape-hatch message. Desktop pane shows the gate number (fleet or profile) with a spend ring, per-model and per-day breakdowns and a status-bar budget chip. ### [research-notes](https://github.com/NousResearch/hermes-agent/pull/149331) Tools · by Adebisi1111 · merged · 11-05 Research notes for Hermes: fetch web pages as text, save notes with tags, search notes, list notes, and delete notes. All data stored locally in JSON. ### [Verifyum](https://github.com/NousResearch/hermes-agent/pull/149329) Tools · by aisenseapi · merged · 11-05 Privacy-preserving file proofs on Solana Mainnet through Verifyum (no account, no key): the agent hashes a file on the user's machine, anchors only a domain-separated commitment, follows the proof until it is finalized, and verifies public proofs, with one skill bundled. Portable Agent Plugins v1 package (plugin.json + mcp.json + skills/). ### [stage](https://github.com/NousResearch/hermes-agent/pull/149328) Tools · by adindamochamad · merged · 11-05 Stage Hermes write\_file and patch calls in a scratch copy. Original files stay unchanged until stage\_apply. ### [injection-guard](https://github.com/NousResearch/hermes-agent/pull/149325) Tools · by moneya · merged · 11-05 Prompt-injection firewall for tool output. A tool\_execution middleware scans results from web, browser, file-reading and mail tools for instructions addressed to the model, redacts the matched spans, strips zero-width and bidi characters, HTML comments carrying instructions, CSS-hidden text and oversized base64 blobs, then wraps what remains in a labelled untrusted- data envelope. Fully local and deterministic: seven weighted signal families with a default threshold of six, so one strong signal or two medium signals trip it; no network calls, no API calls, and no model in the decision path. Ships a read-only injection\_guard\_stats tool and a /guard command, and appends one audit line per block carrying signal names and scores only — never the offending text. Embedded envelope markers are defused before wrapping, so a payload cannot close the untrusted block early. For tools listed in file\_reader\_tools the destructive family is exempt, because redacting a local migration or deploy script would corrupt it if the model rewrote the file; instruction-override and concealment still fire on reads. ### [ROS Robot Control](https://github.com/NousResearch/hermes-agent/pull/149313) Tools · by Arca Computer (arcabotai) · merged · 11-05 Drive and observe ROS / ROS 2 robots from Hermes over rosbridge: list topics, services and nodes, read odometry, grab camera frames for the vision model, publish, call services, drive and e-stop, with every tool call gated by a pre\_tool\_call safety interlock (speed and duration limits, optional geofence, topic and service write allowlists, a latching e-stop, every blocked call logged). Ships a built-in mock robot (2D sim with a camera and a red cone) so it runs with no ROS at all, a bundled safety skill, and an optional read-only dashboard tab. ### [Anaella](https://github.com/NousResearch/hermes-agent/pull/149279) Tools · by Anaella · open · 11-05 Your work accounts for Hermes: Gmail, Outlook or IMAP mail, Google and Outlook calendars, Drive, Docs and Sheets, Slack, Search Console, Analytics, Meta and Google Ads, Business Profile reviews, Stripe, PayPal, n8n, a phone number (texts and AI calls), a meeting bot and 9 social networks, through one MCP server (find, explain, do over about 260 capabilities) plus 24 workflow skills (social publishing and comment replies, post and comment analysis, inbox triage, scheduling, SEO and ads reports, research and leads, marketing, payments, documents and e-signatures, phone calls, meeting notes, n8n, videos and automations). Portable Agent Plugins v1 package; the server runs locally with pinned npx @anaella/mcp@0.14.0, so Node.js 20+ and npm are required. ### [Myeline](https://github.com/NousResearch/hermes-agent/pull/149277) Tools · by sandbox-group · merged · 11-05 Myeline is a remote MCP memory server: every memory is a single fact with its source attached, AI proposals land in a human review queue, and removing a fact takes a written reason kept in an audit log. One retraction stops a stale fact in every connected assistant at once. Maintained by Sandbox Group, the Myeline vendor. ### [No Cap](https://github.com/NousResearch/hermes-agent/pull/149272) Tools · by jedbrnth-eng · merged · 11-05 Receipts for agent claims. No Cap checks the claims in each final reply (tests pass, pushed, merged, deployed, file written, URL live, SHAs, counts; English plus Chinese, Thai and Spanish for tests, push, merge and deploy) against the tool calls the agent made, with no extra model, and appends a footer: 🧾 NO CAP when a tool call proves the claim, 🧢 CAP only when one directly contradicts it (exit code, test counts, push ref, PR state, or a claimed SHA or branch that does not exist in the session's repo), and an opt-in 🤔 NO RECEIPT. A claim about earlier work can cite an earlier turn of the same session. A green run after the agent changed the suite in that turn (deleted, moved, skipped or deselected tests, a conftest or runner-config filter, PYTEST\_ADDOPTS=-k) no longer earns 🧾 for "all tests pass": it gets a visible 🤔 naming the change, and 🧢 only when an earlier run named the removed test as failing and the reply doesn't say so. /cap makes the agent prove or retract its open claims and re-scores the next reply; /capscore keeps a per-model honesty board, a scorecard and an A/B compare card; /nocap verdict draws a session verdict card; strict mode uses pre\_verify to send a coding turn back, at most twice. Surfaces: a Cap Board dashboard tab (agents, models and skills ranked live; this profile only unless you opt in to a team view), Hermes Desktop receipt chips and a Receipts popover, opt-in Discord embed + buttons, and Cap Coach, which drafts a rule from repeated caps and injects it only after you approve it. Licence: Apache-2.0 (0.1.0 to 0.2.0 were MIT). ### [Hermes Watchdog](https://github.com/NousResearch/hermes-agent/pull/149270) Automation · by 33hodl · open · 11-05 Scheduled jobs break without telling you. A job fails at 3am, so Hermes Watchdog tries the fix that fits (retry it, restart what broke, or pause it), then checks whether the next run worked. You only hear about it when the fix didn't take. Free, no API keys, one command to install. ### [constraint-vault](https://github.com/NousResearch/hermes-agent/pull/149269) General · by jusst · merged · 11-05 Pins the user exact constraints (numbers, addresses, hard rules) as structured local data and re-injects them verbatim into the model context on every turn and immediately after every context compaction, with a SHA-256 hash-logged injection record, because memory providers and compression silently paraphrase or drop exact values. Not a memory provider and not a compressor: a guard layer above both. ### [budget-leash](https://github.com/NousResearch/hermes-agent/pull/149268) Automation · by jusst · merged · 11-05 Hard budget enforcement for Hermes sessions. Tracks estimated model spend (token usage x per-model prices, cache read/write tokens included) and tool-call counts against a per-session cap and a persistent per-day cap; blocks tool calls via pre\_tool\_call once a cap is exhausted and cuts off the model calls themselves via the llm\_execution middleware, answering with a synthetic budget-exhausted reply so the provider is never invoked (fail-closed), so unattended agents cannot run up unbounded cost; warns at 80% of a cap via context injection, keeps a flock-guarded day-ledger on disk that survives restarts and is shared safely across processes, and writes a per-session JSONL spend report. ### [escalation-governor](https://github.com/NousResearch/hermes-agent/pull/149267) Automation · by jusst · merged · 11-05 Deterministic stuck-detection for Hermes sessions. Watches tool/API failure signals (consecutive tool errors, repeated identical calls, API error bursts, model calls without progress) and, when a threshold trips, injects a structured escalation packet (goal, what was tried, what failed) directing the agent to delegate the remaining work to a stronger model via delegate\_task instead of retrying blindly; writes a per-session JSONL report with escalation and token-usage accounting. ### [Hermes Phone](https://github.com/NousResearch/hermes-agent/pull/149263) Voice · by winchxyz · merged · 11-05 Let your Hermes agent place real phone calls for you with explicit approval. It shows a brief card and dials only after you reply "call" (or tap Call), says it is an AI in the first sentence of every call, relays mid-call questions to you in chat with a suggested answer, and reports one clean result (what was agreed, confirmation number, duration, cost) plus the transcript. ### [reachmap](https://github.com/NousResearch/hermes-agent/pull/149256) Tools · by MindSproutStudios · merged · 11-05 Hermes knows what it can reach before it says it can't. Scans signed-in browser sessions on DevTools ports, MCP servers, local services, CLIs and drives; freezes a compact map into each new session's prompt (cache-safe system prompt section); and catches replies that deny or hedge about access the map shows is live, with a footer for the user and a one-shot correction for the agent. Opt-in sign-in probes (off by default) tell an installed CLI apart from a signed-in one, and flag replies that plan to use a CLI whose login has lapsed. Keeps a SQLite ledger of when each capability, or a CLI login, appeared or vanished. ### [Ghost Run](https://github.com/NousResearch/hermes-agent/pull/149252) Tools · by tnvolman · merged · 11-05 Dry-run mode for Hermes Agent. With /ghost on, file writes, patches, deletes and terminal commands are redirected into a throwaway copy of the working folder, and network-touching tools and commands are blocked. Review the diff in the system browser (hermes ghost ui; the Desktop pane only opens that page), then optionally run one configured Verify command in the ghost (Pass/Fail banner; Apply stays human), then apply selected paths or discard. Commands: /ghost on\|off\|status\|verify; hermes ghost diff\|apply\|discard\|ui\|verify. ### [Multi Chat](https://github.com/NousResearch/hermes-agent/pull/149251) Desktop · by BColsey · open · 11-05 Experimental 1–16 real-TUI panes for the Web Dashboard; default profile only, with disclosed read-only private PTY metadata access. ### [hermes-windows-apps](https://github.com/NousResearch/hermes-agent/pull/149246) Tools · by Yicheng-Mark · merged · 11-05 Answer "what is installed on this Windows machine": the software inventory behind Apps & Features, swept straight from the four Uninstall registry keys (the machine's 64-bit and 32-bit views plus the user's), so a program is found even when the Settings page hides it - a per-user install that never shows on the machine page came from the user view on the writing machine. Each row carries version, publisher, size in MB, install date and which view it lives in; the same DisplayName in two views is a 32/64-bit dual install, marked instead of listed twice. Entries the publisher asked to hide (runtime packages, update rollups) are filtered out unless asked for. Details one program with every field its publisher wrote and a reading of the uninstall command (MSI product code, setup executable, appx, script) so the next step is known before anything is run; the install folder is checked against disk. Use it for "what did that installer actually register", for inventory before a cleanup, and for "which of the two installs does uninstalling remove". Read-only: registry keys are enumerated and uninstall strings are reported, never run; no network, no file writes, no elevation. The sweep is pure registry - 102 entries in seven milliseconds on the writing machine, the whole table in ten, no PowerShell, no WMI. The plugin declares no Python dependencies. Windows only. ### [hermes-windows-updates](https://github.com/NousResearch/hermes-agent/pull/149245) Tools · by Yicheng-Mark · merged · 11-05 Answer "what did Windows install lately, and did it work" on Windows: the update history the Settings app shows - Windows updates, Defender signature and platform updates, driver updates - newest first, with each entry's outcome in words (in progress, succeeded, succeeded with errors, failed, aborted) and, when an entry did not end clean, the failure's error code in 0x form (the agent hands codes over signed; -2145124300 is shown as its unsigned spelling). Filter to the ones that did not end clean, or search titles. The hotfix list (KB numbers, security updates, install dates, newest first) answers "when was this machine last patched" at a glance; it carries OS-level updates only, which the notes say. Use it for "why does it say pending forever", for "did the update that broke something actually install", and for a patch report before assuming a machine is current. Read-only: the history is queried through the Windows Update Agent's session interface and CIM, and nothing else - no update search is run, nothing is downloaded, installed or uninstalled, no elevation. On the writing machine a 723-entry history returns the newest 500 in about a tenth of a second; the first call also starts the COM session, about a quarter of a second. The plugin declares no Python dependencies; there is no network call of its own and no file write. Windows only. ### [Lineage](https://github.com/NousResearch/hermes-agent/pull/149244) Tools · by P0GL0L · merged · 11-05 Provenance for everything the agent produces: every file Hermes writes gets a receipt (what it was made from — files read, pages fetched, searches run — by which tool call, model and agent, with SHA-256 hashes) and a Desktop page draws the graph back to the sources, with observed / evidenced / inferred links labelled, never blended. Verify a file and its sources against recorded hashes; export Mermaid/DOT. ### [Network X-ray](https://github.com/NousResearch/hermes-agent/pull/149243) Tools · by rusty4444 · merged · 11-05 Compare DNS, TCP, verified TLS and HTTP HEAD captures of one URL from two machines, find the first observed difference and export a local evidence report. Read-only and local-first, with native agent tools, a Desktop workbench and a stdlib-only portable collector for the second machine. ### [agent-si](https://github.com/NousResearch/hermes-agent/pull/149242) Memory · by joelbrilliant · merged · 11-05 Sourced, model-free memory: a brain of facts that cite sources and quotes, plus search over past Hermes, Claude Code, Codex and Grok sessions on this machine, in one ranked list. No model, no key, no network, no dependencies. Refuses to store words the user didn't say; says when sources disagree. ### [Hermes Receipts](https://github.com/NousResearch/hermes-agent/pull/149239) Tools · by iket72183-collab · merged · 11-05 Deterministic evidence for explicit file, native test-result, and website checks. Preserves immutable as-of receipts with failed, verified, or unverified outcomes. ### [Proofchain — tamper-evident evidence ledger](https://github.com/NousResearch/hermes-agent/pull/149231) Tools · by iamzulx · closed · 11-05 Tamper-evident evidence ledger for every tool call Hermes runs: each pre\_tool\_call and post\_tool\_call appends a row whose SHA-256 commits to the previous row, so editing, deleting or reordering one row makes proofchain\_verify fail at exactly that sequence number; proofchain\_seal HMAC-signs the head and row count with a per-profile key generated locally on first seal, and proofchain\_export returns entries + chain status + head hash as JSON you can attach to an audit or pentest report, while proofchain\_search finds every call that mentioned a URL, file or keyword, and a bundled proofchain:evidence skill plus /evidence (status \| verify \| export \[N\] \| seal \| search \) drive the same ledger from chat. ### [Read-only mode](https://github.com/NousResearch/hermes-agent/pull/149229) Desktop · by chengyimingvb · merged · 11-05 Ask/read-only toggle beside the Desktop composer (Cursor Ask-mode equivalent): while it is on, each send is prefixed with a read-only instruction so the agent answers from reading only, and asks before changing anything. Per-chat state, palette command, no agent tools or hooks. It is a prompt-level constraint, not a sandbox: Hermes has no session-level read-only policy, so the model can still disobey. Desktop-only (desktop/plugin.js inside the plugin SDK surface); no tools, hooks, env vars. ### [Zambo Receipts](https://github.com/NousResearch/hermes-agent/pull/149226) Tools · by zambodotdev · merged · 11-05 Verifiable receipts for Hermes Agent sessions. Lifecycle hooks journal every tool call in a session to Zambo (zambo.dev) as it happens, and when the session ends the plugin wraps it into an AER-1 verifiable receipt that anyone can open and check at a public zambo.dev/run link. Receipt pages are public: anyone with the link can read them, and the summary and goal text passed to zambo\_wrap\_session is published on the receipt page. Adds three tools (zambo\_verify\_receipt, zambo\_wrap\_session, zambo\_session\_status), a /receipt slash command, and a bundled skill. ### [headroom](https://github.com/NousResearch/hermes-agent/pull/149222) Tools · by seongj-un · merged · 11-05 Keeps Hermes alive on small machines (1-4 GB VPS, Raspberry Pi). Heavy terminal jobs (package installs, builds, test suites, compilers, browsers, nested agents) wait for a machine-wide slot and enough free RAM, run inside their own memory-capped systemd --user scope with a raised OOM score, get single-worker flags where a stable one exists (jest --runInBand, pytest -n 0, make -jN, cargo -j N, Node heap cap), and are stopped by a watchdog if the host nears OOM or Hermes' own cgroup starts stalling. Hermes stays outside the scope, and the tool result tells the model the cap, the peak and whether the kernel killed the job, so it retries lighter. Adds headroom\_status, /headroom and hermes headroom status\|events\|explain\|selftest. ### [Hermes BIM Studio](https://github.com/NousResearch/hermes-agent/pull/149221) Tools · by stupidburgers · merged · 11-05 Inspect local IFC files: report wall/slab BaseQuantities and door/window counts, optionally price only these four categories using caller-supplied rates, and generate an interactive offline 3D viewer. ### [Hermes Pocket](https://github.com/NousResearch/hermes-agent/pull/149220) Automation · by belweave · merged · 11-05 Release workspaces, progress, replies and exact phone approvals through Notify. ### [Nintendo 3DS](https://github.com/NousResearch/hermes-agent/pull/149202) Platform · by rndaom · merged · 11-05 Talk to Hermes from a Nintendo 3DS: a gateway platform for the Hermes 3DS homebrew app, with streaming replies, approvals and clarify questions, sessions, skills, jobs, model and slash-command pickers, pictures, documents, audio, voice messages, camera photos and spoken replies, over an end-to-end encrypted Noise IK link (LAN, or away from home through Tailscale Funnel). ### [EventForge for Hermes](https://github.com/NousResearch/hermes-agent/pull/149194) Automation · by sinequix · merged · 11-05 Proxy the local EventForge control plane into Hermes with approval-gated tools and verified-event context. ### [fashion-design](https://github.com/NousResearch/hermes-agent/pull/149227) Tools · by priticolours · merged · 11-05 Fashion design toolkit for drafting ideas, tech packs, size sheets, and manufacturing prep ### [Living Field](https://github.com/NousResearch/hermes-agent/pull/149190) Desktop · by abuz8studios-ship-it · merged · 11-05 Hermes Desktop pane of the ABUZ8 living spatial field. Four domains and a seeded neural field. Synapses speed up only while the focused session is actually working, and the field dims when the gateway socket is not open. No network, no shell, no fabricated metrics. ### [hermes-spendguard](https://github.com/NousResearch/hermes-agent/pull/149189) Automation · by Blackleets · open · 11-05 SpendGuard — a wallet guard for Hermes Agent. Prices every provider API call (provider-reported cost when available, otherwise public list-price estimates; unknown models are counted by tokens and marked unpriced, never priced by invention), tracks spend against daily/monthly USD budgets, warns at configurable thresholds, and hard-stops new turns before they reach the model when a budget runs out — the block makes zero model calls. Budgets change only via the human-typed /spendguard command; the agent's two tools are read-only. Local SQLite ledger, no network, no credentials, no telemetry. ### [Token Dashboard](https://github.com/NousResearch/hermes-agent/pull/149187) Desktop · by Whydiriki700 · merged · 11-05 Token cost and usage dashboard for Hermes Desktop: a right pane with the focused session's live input and output tokens, session cost (backend-reported cost\_usd when available, otherwise a labelled estimate from a built-in per-model price table), API calls and context usage bar, plus a recent-sessions overview with message counts and a status bar cost chip. Desktop-only (desktop/plugin.js inside the plugin SDK surface); no agent tools, hooks, middleware, or env vars. ### [Money Magnet](https://github.com/NousResearch/hermes-agent/pull/149186) Tools · by Whydiriki700 · merged · 11-05 Finance tracker and money magnet for Hermes Agent: record income and expenses, set a monthly budget and watch it fill, value a crypto portfolio against live prices, and scan for price gaps between exchanges. Transactions and budget live in local JSON under $HERMES\_HOME/money-magnet/. ### [ProofGate](https://github.com/NousResearch/hermes-agent/pull/149185) Tools · by Whydiriki700 · merged · 11-05 No proof, no done. Verifies an agent's completion claims (file created, message sent, command ran, tests passed, figures quoted) against the tool evidence recorded in the same session, before the final reply is delivered. Unproven claims are flagged as unverified or withheld, fail closed, and every verdict is written to an append-only ledger stored locally in $HERMES\_HOME/proofgate/. ### [clack](https://github.com/NousResearch/hermes-agent/pull/149183) Tools · by adroidian · merged · 11-05 Encrypted agent-to-agent messaging with durable receipts and no plaintext fallback. ### [custody](https://github.com/NousResearch/hermes-agent/pull/149180) Tools · by Parker-Fawcett · open · 11-05 Resolve what a destructive command will actually touch before Hermes asks you to approve it, then escalate anything this session did not create. Approvals judge a command by its text, and the shell runs it by its semantics: in the PowerShell incident behind #147373, \`,\` bound tighter than \`+\`, so five written paths became eight array elements and three of them were profile roots; 188,303 delete records, 21.4 GiB. Custody computes the semantics first. It reads the command the way the shell does (tilde, allowlisted variables, quoting, PowerShell array precedence, \`find -delete\`, redirects, \`git clean\`, \`Remove-Item\`, \`osascript\`, \`killall\`/\`pkill\`/\`taskkill\`), resolves each target expression to concrete paths with file and byte counts, and attributes every path against a per-session provenance ledger: files the agent wrote, plus a baseline snapshot of the working tree taken at session start. Anything that predates the session belongs to the user, and it goes to Hermes's own approval gate with the real numbers and the reason attached. Globs, command substitution and unknown variables resolve to an unknown target, which escalates instead of guessing. A GUI-app command that supplies its own answer (\`close active presentation saving no\`, the #141346 shape) is blocked, because no approval prompt can be answered for the user's unsaved state. The command is never executed and nothing is auto-approved: custody adds a human step, it never removes one. ### [gogogoogle](https://github.com/NousResearch/hermes-agent/pull/149174) Tools · by josefacero · merged · 11-05 Hand bulk work to Google Antigravity's agy CLI as a worker agent on your Antigravity quota while Hermes keeps its own model, then have a fresh agy conversation verify the result against the real files and git diff. Roles do, research and review; background jobs; resume after timeout; a verifier chain that skips models out of quota until their reset; a git check that fails read-only runs that changed files; per-run stats. ### [basekit](https://github.com/NousResearch/hermes-agent/pull/149166) Tools · by D0xedDevi0 · merged · 11-05 Read-only live Base (chain 8453) intelligence — gas and real USD cost estimates, wallet portfolios, decoded transaction receipts, token facts with proxy detection, and windowed transfer/approval reads — plus a desktop pane, status-bar chip and a standard-library-only core with no PyPI dependencies. ### [SerpKite](https://github.com/NousResearch/hermes-agent/pull/149160) Web · by SerpKite · merged · 11-05 SerpKite web search and extract provider for Hermes Agent. Google web results through /v1/search and pages as Markdown through /v1/webpage. Requires SERPKITE\_API\_KEY (https://app.serpkite.com/keys). ### [Understudy](https://github.com/NousResearch/hermes-agent/pull/149159) Models · by stffinfcti · merged · 11-05 Audition cheaper models on your real Hermes history before you switch. Understudy replays a sample of your past decisions (replies and tool calls) on candidate models through Hermes's own credentials, has a blind judge compare each step with what your current model actually did, and reports how often each candidate could have taken over (with a 95% confidence interval), its critical misses, a per-category breakdown and projected monthly cost. Two tools, /understudy, hermes understudy, and a Desktop page with an inline scorecard. ### [thott.bot](https://github.com/NousResearch/hermes-agent/pull/149158) Tools · by thott-bot · merged · 11-04 thott.bot makes your AI an expert on World of Warcraft: Forever, from Blizzard's own game files for the build you're playing, current to the latest hotfix, with Blizzard's news and credited community data, drawn right in the chat. ### [Proxmox VE](https://github.com/NousResearch/hermes-agent/pull/149153) Tools · by cgray2222 · open · 11-04 Proxmox VE for Hermes Agent: six pve\_\* tools (cluster inventory, per-guest status and config, node health and storage, snapshot list/create/delete, recent backup job results, and QEMU guest-agent command execution) plus an @pve: context reference that pulls live cluster, guest or node state into the prompt on demand. Talks to the Proxmox HTTP API directly with a scoped API token; standard library only, no Python dependencies. Write actions (snapshot create/delete, guest exec) refuse unless the same call carries confirm=true. ### [Hermes Chronos](https://github.com/NousResearch/hermes-agent/pull/149151) Tools · by DevTarlow · merged · 11-04 Time-travel filesystem checkpoints, approval-gated rollback, and a scrubbable Desktop timeline for Hermes Agent. ### [Undo Effects](https://github.com/NousResearch/hermes-agent/pull/149148) Tools · by Moodtuner997 · merged · 11-04 Puts back files an agent deleted or overwrote: before write\_file, patch and destructive terminal commands (rm, mv, cp, redirects, tee, sed -i, find -delete) it saves the affected files to a local store and lets the tool run unchanged; /undo-effects lists the effects and restores them, refusing to overwrite later edits unless forced and keeping a safety copy so a restore can be undone. ### [CaMeL Guard](https://github.com/NousResearch/hermes-agent/pull/149147) Tools · by nativ3ai · merged · 11-04 Gate sensitive tools on trusted user capabilities after untrusted tool output. ### [agentmail-frontdoor](https://github.com/NousResearch/hermes-agent/pull/149142) Tools · by tylerjharden · open · 11-04 Claim, classify, and route AgentMail threads. Create drafts only. ### [fleet-memory](https://github.com/NousResearch/hermes-agent/pull/149142) Memory · by tylerjharden · open · 11-04 Recall local fleet ops notes, with an optional Notion search mirror. ### [odoo-prod-guard](https://github.com/NousResearch/hermes-agent/pull/149142) Tools · by tylerjharden · open · 11-04 Odoo JSON-2 tools with a Hermes approval gate on non-routine production writes. ### [Guardrails](https://github.com/NousResearch/hermes-agent/pull/149133) Tools · by eddiemessiah · open · 11-04 The policy layer every payment plugin is missing. Backend-neutral hard spend limits enforced in code before any payment tool runs: per-payment, daily and per-payee caps, Hermes approval for large or first-time payees (scoped to one payment), idempotency keys so retries never double-charge, a kill switch only a human can lift, and a SHA-256 hash-chained audit log. Can also guard other plugins' spending tools. Ships a zero-key mock backend and an optional Stripe test-mode backend. ### [wazoo](https://github.com/NousResearch/hermes-agent/pull/149129) Memory · by wazootech · merged · 11-04 Wazoo world-model memory provider — typed memories (decision, preference, person, project, finding, correction) in an RDF graph with retraction provenance, background recall prefetch, hybrid search, and SPARQL for structural questions. Works against the hosted Wazoo cloud (provisions a world automatically from a platform token) or a self-hosted worlds endpoint with a locally minted key. ### [Rigforge](https://github.com/NousResearch/hermes-agent/pull/149127) Desktop · by DECRUX9812 · open · 11-04 A build studio for robots, printed machines and controller boards, inside Hermes Desktop. Say a sentence — or click a card — and get the whole build: posed 3D geometry with real forward kinematics and an exploded view, joint torque sizing that picks real servos (or a belt stage when one cannot, and says so when nothing can), a per-rail power budget with wire gauge and a fuse that never exceeds its wire, an ESP32 pin map that refuses strapping and flash pins, firmware generated from that same map, and a priced BOM with SKUs, mass, print time and per-part STL packs split to fit your printer bed. The board track designs a controller board with placements, solder verdicts and a fab/assembly pack (BOM + centroid + spec). Every headline number opens into the formula that made it, any change can be compared before/after, and a build code carries the whole design in one line. Ten archetypes from a sensor node to a metre-scale gantry. Deterministic and fully offline — every number is computed, never narrated. Two agent tools (rigforge\_build, rigforge\_state) queue a build and read the spec back, plus rigforge\_prices for live supplier pricing; a ::rigforge{...} transcript directive turns a chat line into a "load into studio" card. Desktop page plus agent tools; no hooks, no middleware, no env vars. ### [neuralwatt](https://github.com/NousResearch/hermes-agent/pull/149124) Tools · by 0xGingi · merged · 11-04 Check NeuralWatt service health, available models, account balance, request costs, energy use, and session activity from Hermes. Includes seven read-only tools. Health and public model discovery need no credentials; account analytics use your NeuralWatt API key. ### [Zaprite](https://github.com/NousResearch/hermes-agent/pull/150568) Tools · by welliv · merged · 11-04 Get paid from a Hermes chat. One checkoutUrl for card, bank, Lightning, on-chain bitcoin, and stablecoins, then re-fetch until Paid. ### [Shortrizz](https://github.com/NousResearch/hermes-agent/pull/148687) Tools · by mojomast · closed · 11-04 Skill id shortrizz:shortrizz-clipper. Self-hosted AI clipper: turn a long video (yt-dlp URL or local file) into captioned 9:16 shorts. Transcribes on CPU with faster-whisper, picks moments with an LLM (or a built-in heuristic), reframes with face tracking and burns in face-aware word-by-word captions (Hormozi, MrBeast, Abdaal or custom styles). Optional edits: filler/restart cuts, cold-open hook, emoji pops, tight 20-35 s mode, skipping moments used by earlier runs, and a stock Kokoro narrator intro/CTA (weights download on first use; never clones voices). shortrizz\_make\_highlights builds one 16:9 highlight reel from the top viral-scoring segments (source framing kept, no vertical crop), with loudness normalised, a short crossfade or hard cuts, and optional lower-third captions. A highlights job writes chapters.txt and highlights.json. Writes MP4s, clips.json and an HTML gallery. Renders run as background jobs; poll shortrizz\_status. Publishing is a dry-run plan only; nothing is uploaded. Downloads the video you name with yt-dlp; the optional LLM call sends the transcript to the OpenAI-compatible endpoint you configure. Needs ffmpeg on PATH. Only clip content you have rights to. ### [Annotate for Hermes](https://github.com/NousResearch/hermes-agent/pull/146960) Desktop · by LarryLemonBot · open · 10-30 Highlight passages in a reply, add comments, and send them together. The annotate feature Hermes is missing. Works on macOS now; Windows and Linux are next. ## Navigation - [Home](https://aaddrick.com/) - [Blog](https://aaddrick.com/blog) - [Tags](https://aaddrick.com/tags) - [Archive](https://aaddrick.com/archive) - [Projects](https://aaddrick.com/projects) - [Feeds](https://aaddrick.com/feeds) - [Stats](https://aaddrick.com/stats) - [About](https://aaddrick.com/about) - [Help](https://aaddrick.com/help)