Plugin Marketplace

Full Spotify playback control — 7 tools for playback, devices, queue, search, playlists, albums, and library. Uses Spotify Web API + PKCE OAuth.

Join Google Meet calls, transcribe live captions, and speak in realtime. Three modes: transcribe-only, duplex audio, remote node.

Auto-tracks and cleans up ephemeral files (test scripts, temp outputs, cron logs) created during sessions. Runs via hooks — no agent action required.

Microsoft Teams meeting pipeline with durable runtime state. Graph-backed transcript-first meeting summaries with operator CLI flows.

Multi-profile work queue and orchestration board. Durable SQLite board for multi-agent collaboration with task lifecycle management.

Default web search backend. Excellent balance of speed and result quality. Supports direct API and Nous-hosted tool-gateway.

AI-optimized web search and content extraction. Good quality results tailored for agent use cases.

Self-hosted privacy-respecting metasearch engine. No API keys, no rate limits, no tracking. Aggregates multiple search sources.

Free web search via DuckDuckGo. No API key needed. Lower result depth than dedicated APIs.

Independent search index via Brave Data-for-Search API. 2,000 queries/month on free tier.

Exa web search and content extraction. Excellent for technical and academic content.

Parallel.ai web search with async SDK. Newer backend, less field-tested than alternatives.

Keenable search backend. Part of the keyless round-robin ring of web search backends that need no API key.

Perplexity search and content extraction. Requires a Perplexity API key.

xAI Web Search - searches the web via Grok agentic web_search tool (Responses API). Requires xAI Grok OAuth (via hermes auth) or XAI_API_KEY.

Default built-in memory. Local SQLite fact store with FTS5 search, trust scoring, and HRR-based compositional retrieval. Zero setup, zero cost.

AI-native cloud memory with dialectic Q&A, semantic search, and persistent conclusions across devices.

Server-side LLM fact extraction with semantic search, reranking, and auto-deduplication. Strong search but known perf issue with prompt caching.

Cloud memory provider. Community reports only ~30% of available API surface is utilized.

Cloud memory provider. Limited documentation available.

Cloud memory provider. Limited documentation available.

Cloud memory provider. Limited documentation available.

Cloud memory provider. Limited documentation available.

OpenAI image generation using GPT Image 2 and 2.5 (Flare/Sunburst) for generation and editing. Saves to ~/.hermes/cache/images/.

Image generation backed by ChatGPT/Codex OAuth. Same gpt-image-2 model, different billing path.

xAI image generation via Grok-Imagine. Text-to-image. Newer and less refined than OpenAI.

DeepInfra image generation backend (FLUX, Qwen-Image and similar) via the OpenAI-compatible /v1/images/generations endpoint. Catalog discovered live.

FAL.ai image generation backend (flux-2-klein, flux-2-pro, nano-banana, gpt-image-1.5, recraft-v3 and more).

Krea image generation backend (Krea 2 Large, Medium, and Medium Turbo foundation models). Direct KREA_API_KEY or managed Nous Subscription gateway.

OpenRouter and Nous Portal image generation via chat-completions image output with reference grounding. Text-to-image and image-to-image.

FAL.ai multi-model video generation. Supports Veo 3.1, Kling, Pixverse. Text-to-video and image-to-video.

Grok-Imagine video generation. Text-to-video, image-to-video, reference-image-guided, video edit, video extend.

DeepInfra video generation (text-to-video and image-to-video) via the OpenAI-compatible /v1/openai/videos endpoint. Catalog discovered live.

Microsoft Teams gateway adapter via Bot Framework. Supports Adaptive Card approval prompts.

Internet Relay Chat gateway adapter. Uses Python stdlib asyncio — no external dependencies.

Google Chat gateway via Cloud Pub/Sub pull subscription. No public URL required.

LINE Messaging API gateway with HMAC-SHA256 signature verification and reply token optimization.

Agent-to-Agent protocol adapter (A2A v1.0). Lets Hermes agents exchange messages with other A2A-compatible agents.

Buzz adapter - Block open-source human+agent collaboration platform built on the Nostr protocol. Relays Buzz channels and DMs to the agent; outbound traffic shells out to the buzz CLI, inbound uses a Nostr WebSocket subscription.

ntfy push-notification channel - simple HTTP pub-sub notification service. Works with the free public ntfy.sh server or any self-hosted instance; subscribe from the ntfy mobile app to talk to the agent.

Photon iMessage integration - a managed service that handles Apple line allocation and abuse prevention, so no self-hosted Mac relay is needed. Supports voice, images, files, and reactions. Distinct from BlueBubbles.

Raft adapter - connects Hermes to Raft as an external agent through a local wake-channel bridge. The adapter receives content-free wake hints on a loopback HTTP endpoint; the agent reads and sends messages through the Raft CLI.

SimpleX Chat adapter - private decentralised messaging with full media and voice support. Every contact is an opaque connection-time ID, so there are no persistent user IDs.

Matrix gateway adapter for federated, decentralised deployments. Full media support including images and files.

Self-hosted enterprise chat adapter using the Mattermost REST API and webhooks. Images, files, and threads.

Email gateway adapter over SMTP/IMAP with subject-based threading. Best for notification-heavy workflows and offline communication.

Bare-bones text-only channel via Twilio. Useful for alerts, 2FA codes, and offline notifications.

Home Assistant gateway adapter using the WebSocket API, so the agent can send and receive smart home notifications alongside device control.

Alibaba enterprise messaging adapter with OAuth 2.0 auth. Images, files, and reactions. Essential for enterprise deployments in China.

ByteDance enterprise platform adapter with card-style interactive messages and full media support.

Tencent enterprise communication adapter. Two modes: simple webhook bot, or full callback app with crypto support.

WhatsApp Web bridge via Baileys - QR code pairing, no Business API needed. Note the docs also ship an official WhatsApp Cloud API adapter as a separate platform.

Telegram gateway adapter via Bots API. Full feature support - voice, images, files, threads, typing, and streaming.

Discord gateway adapter with voice channel support, reactions, and permission controls.

Slack gateway adapter using Socket Mode for real-time WebSocket communication. Voice, images, files, threads, reactions, and typing.

Full observability — traces conversations, LLM calls, and tool usage. Hooks into pre/post API requests, LLM calls, and tool calls. Essential for debugging.

Gateway to 200+ models from every major provider. One API key gives access to everything. The recommended starting point.

Claude models (Sonnet, Haiku, Opus). Best-in-class for coding and complex reasoning.

OpenAI GPT models via the openai-api provider. Broad capabilities, excellent tool-use.

DeepSeek Chat and DeepSeek Reasoner. Excellent quality-to-price ratio. Strong for coding.

Gemini 3 Flash and newer models. Fast, cheap, large context window.

Local models via Ollama. Run Llama 3.2, Mistral, Phi-3, and hundreds more on your own hardware. Zero API cost.

Fireworks AI - fast inference for open and proprietary models. Uses native slash-form catalog model IDs.

Arcee AI - Trinity model family.

GMI Cloud - open and reasoning models via an OpenAI-compatible API.

Xiaomi MiMo models via the Xiaomi API.

Tencent TokenHub (Hy3 Preview and other Tencent-hosted models).

Kilo Code provider.

StepFun Step-series models via an OpenAI-compatible API.

Upstage Solar API provider.

NVIDIA NIM - Nemotron and other open models on build.nvidia.com, or a local NIM endpoint via NVIDIA_BASE_URL.

AWS Bedrock - Claude, Nova, DeepSeek, and Llama models via the Converse API. Uses the standard boto3 credential chain, so no Hermes API key is needed.

Google Vertex AI - Gemini models via the Vertex OpenAI-compatible endpoint. OAuth2 access tokens minted from a service-account JSON or Application Default Credentials.

Azure AI Foundry provider using an Azure OpenAI / Foundry endpoint and key.

LM Studio local server - any GGUF model served locally over an OpenAI-compatible API at http://127.0.0.1:1234/v1.

Ollama Cloud - cloud-hosted Ollama API for the same open-weight catalog without a GPU. Catalog fetched dynamically from ollama.com/v1/models.

Qwen Portal with browser-based OAuth login (PKCE). Routes to portal.qwen.ai/v1 and persists the refresh token in ~/.hermes/auth.json.

MiniMax OAuth - browser login for MiniMax-M2.7 and MiniMax-M2.7-highspeed. Uses the Anthropic Messages-compatible endpoint and ignores MINIMAX_API_KEY.

xAI Grok OAuth for SuperGrok and X Premium+ subscribers - browser login instead of an API key. The bearer token is reused by TTS, image gen, video gen, and transcription.

OpenCode Zen provider.

GitHub Copilot as an LLM provider using the Copilot backend models. Free for verified students, teachers, and OSS maintainers.

GitHub Copilot via ACP - spawns a local copilot --acp --stdio subprocess instead of using the Copilot API directly.

Custom / self-hosted OpenAI-compatible endpoint - point Hermes at any server implementing /v1/chat/completions. This is also the path used for local Ollama (base_url http://localhost:11434/v1, no key).

Alibaba DashScope (international) - Qwen Cloud models via an OpenAI-compatible API.

Alibaba Cloud Coding Plan - a separate billing SKU with its own endpoint at coding-intl.dashscope.aliyuncs.com. Reuses the same DASHSCOPE_API_KEY as the alibaba provider.

Moonshot Kimi Coding provider (global api.moonshot.ai and China api.moonshot.cn endpoints).

Moonshot Kimi provider against the China endpoint api.moonshot.cn.

MiniMax M-series models (global endpoint, China endpoint, and OAuth variants).

MiniMax against the China endpoint.

OpenCode Go provider.

Z.AI / ZhipuAI GLM models. Hermes auto-probes global, China, and coding endpoints to find one that accepts the key.

NovitaAI Model API - 200+ models plus Agent Sandbox and GPU Cloud.

Hugging Face Inference Providers - routes to 20+ open models through a unified OpenAI-compatible endpoint with automatic backend failover.

DeepInfra - 100+ open models, pay-per-use.

OpenAI Codex provider using ChatGPT/Codex OAuth and the Responses API.

Ramp Router provider.

security-guidance - a 25-rule dangerous-code pattern matcher that hooks write_file and patch to flag risky edits before they land.

cron_providers/chronos - managed scale-to-zero cron provider. Keeps scheduled jobs off your machine when idle.

Dashboard auth provider - username/password with no OAuth identity provider. Activates when dashboard.basic_auth.username plus a password or password_hash are configured.

Dashboard auth provider - non-interactive shared bearer secret for the gateway drain-control endpoint. First consumer of the generic token-auth capability.

Dashboard auth provider - OAuth 2.0 authorization-code plus PKCE against Nous Portal. Auto-activates when dashboard.oauth.client_id is configured.

Dashboard auth provider - generic self-hosted OpenID Connect (authorization-code plus PKCE, public client). Works with any conformant OIDC provider such as Keycloak, Authentik, Zitadel, or Okta.

browser/browser_use - Browser Use cloud browser backend for the browser toolset. Supports a direct BROWSER_USE_API_KEY or the managed Nous tool gateway.

browser/browserbase cloud browser backend. Supports stealth, proxies, and keep-alive sessions, and auto-falls back when paid features are unavailable.

browser/firecrawl cloud browser backend. Distinct from the Firecrawl web search and extract plugin, though the two share an API key and operate on different endpoints.

Proposed external memory provider for augmented recall. Most-wanted plugin feature request.

Achievement/badge tracking system. Originally community-built by @PCinkusz, later brought into the main repo.

Proposed gateway platform plugin. Feature request on GitHub.