Hermes Agent — Ecosystem Guide
Hermes Agent — Complete Configuration & Ecosystem Guide
Source basis: Official Nous Research docs, canonical GitHub repo (
NousResearch/hermes-agent), community awesome list (0xNyk/awesome-hermes-agent), and theresearchwang13.space/hermes/tool reference (image-sourced, April 2026). All region-specific access notes removed — global supply chain assumed throughout.
What Is Hermes Agent?
Hermes is an open-source, self-improving AI agent built by Nous Research — the lab behind the Hermes, Nomos, and Psyche model families. MIT-licensed. github.com/NousResearch/hermes-agent (23k+ stars, April 2026).
The core proposition differs from typical agent wrappers: Hermes has a closed learning loop. It creates skills from experience, improves them during use, maintains persistent cross-session memory, and builds a progressively deeper model of its user. The longer it runs, the more capable it becomes on your specific workload — without manual fine-tuning.
It is not a coding copilot tied to an IDE, nor a chatbot wrapper around a single API. It is a persistent autonomous agent designed to live on infrastructure you control, reachable from any messaging platform, capable of running fully unattended.
Architecture at a Glance
┌─────────────────────────────────────────────────────────┐
│ Messaging Gateway │
│ Telegram · Discord · Slack · WhatsApp · Signal · │
│ Matrix · Mattermost · Email · SMS · DingTalk · │
│ Feishu · WeCom · BlueBubbles · Home Assistant │
└────────────────────────┬────────────────────────────────┘
│
┌────────────────────────▼────────────────────────────────┐
│ Core Agent Loop │
│ Prompt building → LLM call → tool dispatch → loop │
│ Context window management (summarisation + caching) │
└──────┬────────────────┬────────────────┬────────────────┘
│ │ │
┌──────▼──────┐ ┌──────▼──────┐ ┌─────▼───────────────┐
│ Tools (47) │ │ Memory │ │ Skills System │
│ web search │ │ MEMORY.md │ │ ~/.hermes/skills/ │
│ browser │ │ USER.md │ │ agentskills.io std │
│ file I/O │ │ FTS5 DB │ │ hub install + scan │
│ terminal │ │ Honcho │ │ │
│ delegation │ │ Mem0 etc │ │ │
└──────┬──────┘ └─────────────┘ └──────────────────────┘
│
┌──────▼──────────────────────────────────────────────────┐
│ Terminal Backends (6) │
│ Local · Docker · SSH · Daytona · Singularity · Modal │
└─────────────────────────────────────────────────────────┘
Installation
Requirements
The only hard prerequisite is Git. The installer manages everything else:
| Dependency | Version | Notes |
|---|---|---|
| Python | 3.11 | Installed via uv, no sudo required |
| Node.js | 22 | Browser automation and WhatsApp bridge |
| ripgrep | latest | Fast file search |
| ffmpeg | latest | Audio processing for TTS and voice mode |
Linux / macOS / WSL2
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
source ~/.bashrc # or ~/.zshrc
hermes
Android (Termux)
Same installer script — Termux is detected automatically. Uses pkg for system dependencies and a curated .[termux] pip extra. Browser and WhatsApp bootstrapping skipped by default.
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
Windows
Native Windows is not supported. Install WSL2 (wsl --install), then run the Linux installer from within WSL2.
Nix / NixOS
Dedicated Nix flake and declarative NixOS module available. See the Nix & NixOS Setup guide.
Post-install Configuration Commands
hermes model # Choose LLM provider and model
hermes tools # Enable / disable tool groups
hermes gateway setup # Connect messaging platforms
hermes config set # Set individual config values
hermes setup # Full interactive setup wizard
hermes doctor # Diagnose missing dependencies
LLM Providers
Hermes works with any OpenAI-compatible endpoint. Switch without code changes via hermes model. You pay only for API usage — Hermes itself is free.
Frontier / Cloud Providers
| Provider | Key Models | Notes |
|---|---|---|
| OpenRouter | 200+ models | One API key for everything — recommended starting point |
| Nous Portal | Nous-native models | Nous Research's own inference endpoint |
| OpenAI | GPT-4o, o1, o3 | Direct API |
| Anthropic | Claude 3.5 / 4 series | Via OpenRouter or compatible proxy |
| Gemini 2.0 / 2.5 series | Via OpenRouter or compatible proxy | |
| xAI | Grok-3 | Via OpenRouter |
| Mistral | Mistral Large, Codestral | Via OpenRouter |
| NVIDIA NIM | Nemotron, Llama-3.1-Nemotron | Direct NIM endpoint |
| Groq | Llama 3.3 70B | Ultra-low latency — good for fast subagent tasks |
| Together AI | Open model hosting | Budget-friendly inference |
| Fireworks AI | Open model hosting | Fast inference, generous free tier |
Local / Self-hosted
| Provider | Notes |
|---|---|
| Ollama | Point at http://127.0.0.1:11434/v1 — zero ongoing cost |
| LM Studio | GUI-based local model runner |
| vLLM | Production-grade local serving |
| Any OpenAI-compat | Custom endpoints fully supported |
Subagent Model Routing
Route subagents to a different model than your primary orchestrator:
# ~/.hermes/config.yaml
delegation:
model: "google/gemini-2-flash-preview"
provider: "openrouter"
Common pattern: powerful frontier model as orchestrator → fast/cheap model for subagent tasks (summarisation, formatting, classification).
Core Features
Closed Learning Loop
Four mechanisms working together:
- Autonomous skill creation — After completing a novel task, Hermes generates a reusable skill document. Next time, it loads the skill rather than reasoning from scratch.
- Skill self-improvement — Detects when a skill underperforms and refines it during use.
- FTS5 cross-session recall — Every conversation is indexed in a full-text-search SQLite database. Hermes retrieves and summarises relevant past exchanges on demand.
- Dialectic user modeling — Via Honcho, Hermes maintains
USER.md: a structured, continuously-updated model of your preferences, working patterns, and environment.
Persistent Memory
Two files persist across all sessions in ~/.hermes/:
MEMORY.md— Curated facts: tech stack, project state, recurring preferences.USER.md— Deeper model of you as a collaborator.
Context window management is automatic: at ~50% fill, oldest exchanges are replaced with a compact summary. Anthropic prompt caching is applied when Claude is the backend, reducing cost on long sessions.
Skills System
Skills are on-demand knowledge documents in ~/.hermes/skills/, following the agentskills.io open standard — portable across Hermes, Claude Code, Cursor, and Codex.
Progressive disclosure — only the section currently needed is loaded into context, minimising token cost.
Hub-installed skills pass a security scanner (data exfiltration, prompt injection, destructive command checks) before installation.
Tool System (47 built-in)
| Toolset | Capabilities |
|---|---|
| Web | Search, extract, browse, vision |
| Terminal | Shell execution, file read/write/edit |
| Memory | Read/write memory, search past conversations |
| Delegation | Spawn subagents, execute_code RPC |
| Media | Image generation, TTS, voice transcription |
| MCP | Connect any MCP server |
Subagent Delegation
delegate_task spawns child agent instances with isolated context, restricted toolsets, and their own terminal sessions. Default: 3 concurrent subagents (configurable). Use for parallel workstreams — research, writing, and verification running simultaneously.
execute_code lets Hermes write Python scripts that call Hermes tools programmatically via sandboxed RPC, collapsing multi-step pipelines into a single LLM inference call.
Scheduled Automations (Cron)
Natural language or standard cron expressions. Jobs can attach skills, deliver results to any connected platform, and support pause/resume/edit. Typical uses:
- Daily briefing delivered to Telegram
- Nightly server backup with status report
- Hourly monitoring with threshold alerts
Context Files
Auto-discovered and loaded at session start:
| File | Purpose |
|---|---|
.hermes.md |
Hermes-specific project instructions |
AGENTS.md |
Cross-agent shared instructions (Claude Code, Codex compatible) |
CLAUDE.md |
Claude Code compatibility |
SOUL.md |
Personality override for this project |
.cursorrules |
Cursor compatibility |
Context References
Inject content inline with @:
@src/main.py → file content
@./ → directory listing
@HEAD~3 → git diff
@https://example.com → page content
Checkpoints and Rollback
Working directory is snapshotted before every file change. /rollback restores the previous state. No manual git commits needed as a safety net.
Voice Mode
Full voice interaction across CLI and messaging platforms. Microphone input, spoken replies. Discord voice channel support for live group conversations.
Memory Providers
Beyond the built-in MEMORY.md system, external backends are pluggable:
| Provider | Characteristic |
|---|---|
| Honcho | Dialectic user modeling — bundled as default |
| Mem0 | Graph-based memory with entity relationships |
| OpenViking | Vector-search memory |
| Hindsight | Retrospective learning from outcomes |
| Holographic | Distributed memory representation |
| RetainDB | SQL-backed structured memory |
| ByteRover | Lightweight embedded memory |
Browser Automation
Multiple backends, configured via hermes tools:
| Backend | Type | Notes |
|---|---|---|
| Browserbase | Cloud | Managed headless browsers — most reliable for production |
| Browser Use | Cloud | Alternative managed option |
| Local Chrome (CDP) | Local | Attach to a running Chrome instance via DevTools Protocol |
| Local Chromium | Local | Hermes manages its own Chromium instance |
| vessel-browser | Community | AI-native Linux browser built for agents from the ground up |
| hermes-cloudflare | Plugin | Browser rendering through Cloudflare's infrastructure |
Messaging Gateway
15+ platforms from a single gateway process:
| Category | Platforms |
|---|---|
| Real-time messaging | Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Mattermost |
| Email / SMS | Email (SMTP/IMAP), SMS |
| Enterprise | DingTalk, Feishu (Lark), WeCom |
| Other | BlueBubbles (iMessage bridge), Home Assistant |
Start on one platform, pick up on another — unified context across all gateways.
Terminal Backends (6)
| Backend | Best for |
|---|---|
| Local | Development, testing |
| Docker | Isolated container execution with namespace hardening |
| SSH | Remote server execution |
| Daytona | Serverless — hibernates when idle, near-zero cost |
| Singularity | HPC / research clusters |
| Modal | Serverless GPU workloads |
Recommended production pattern: Daytona or Modal backend + Telegram gateway. Agent runs in the cloud; you talk to it from your phone.
Voice / TTS / STT
Five TTS providers, selectable via hermes tools:
| Provider | Cost | Notes |
|---|---|---|
| Edge TTS | Free | Microsoft-backed, good quality, zero cost |
| ElevenLabs | Paid | Highest quality, voice cloning support |
| OpenAI TTS | Paid | Consistent, well-integrated |
| MiniMax | Paid | Strong multilingual support |
| NeuTTS | Paid | Alternative option |
STT uses model-native transcription capabilities or platform audio processing.
Image Generation
Eight models via FAL.ai, configured with hermes tools:
| Model | Characteristic |
|---|---|
| FLUX 2 Klein | Fast, efficient |
| FLUX 2 Pro | High quality |
| GPT-Image 1.5 | OpenAI-aligned style |
| Nano Banana Pro | Speed-optimised |
| Ideogram V3 | Strong text-in-image |
| Recraft V4 Pro | Design-oriented output |
| Qwen | Multilingual context awareness |
| Z-Image Turbo | Ultra-fast generation |
Web Search Backends
Configurable via hermes tools. The community hermes-web-search-plus plugin adds intelligent multi-provider routing:
| Provider | Notes |
|---|---|
| Tavily | Recommended default — purpose-built for LLM integration |
| Serper | Google results via API |
| Exa | Neural search — strong for research queries |
| Brave Search | Privacy-respecting, no personalisation bias |
| SerpAPI | Broad engine support |
Advanced Integrations
MCP (Model Context Protocol)
Connect any MCP server via stdio or HTTP transport. Per-server tool filtering and sampling support included. Hermes gains access to GitHub, databases, file systems, and internal APIs without writing native tools.
API Server (OpenAI-compatible)
Expose Hermes as an OpenAI-compatible HTTP endpoint. Connect any frontend: Open WebUI, LobeChat, LibreChat, and others.
IDE Integration (ACP)
Use Hermes inside ACP-compatible editors: VS Code, Zed, JetBrains. Chat, tool activity, file diffs, and terminal commands render natively inside the editor.
RL Training (Research)
Trajectory data generation in ShareGPT format. Integrated with Atropos for RL and model fine-tuning. Batch processing runs the agent across thousands of prompts in parallel.
Provider Resilience
- Fallback providers — Automatic failover on error, including independent fallback for vision and compression.
- Credential pools — Rotate across multiple API keys per provider; automatic rotation on rate limit or failure.
- Provider routing — Cost/speed/quality sorting, whitelists, blacklists, priority ordering.
Customisation
Personality (SOUL.md)
SOUL.md is the first content in the system prompt — defines Hermes's default voice. Swap presets per session with /personality, or place a custom SOUL.md in any project directory.
Plugins
Three types, managed via hermes plugins:
- General plugins — Custom tools and lifecycle hooks.
- Memory providers — Alternative cross-session knowledge backends.
- Context engines — Alternative context management strategies.
Event Hooks
- Gateway hooks — Logging, alerts, webhooks on message events.
- Plugin hooks — Tool interception, metrics, guardrails.
Skins / Themes
CLI visual customisation: banner colors, spinner animations, response-box labels, branding text.
Tips and Productivity Patterns
| Pattern | How |
|---|---|
| Resume last session | hermes -c |
| Resume by name | hermes -r "my research project" |
| Interrupt and redirect mid-response | Ctrl+C once |
| Force exit | Ctrl+C twice within 2 seconds |
| Paste clipboard image | Ctrl+V in CLI — agent uses vision directly |
| Tab-complete slash commands | / then Tab |
| Inject file into message | @path/to/file |
| Persist project instructions | AGENTS.md in project root — auto-loaded every session |
Ecosystem: Notable Community Projects
GUIs and Dashboards
| Project | Stars | Description |
|---|---|---|
| hermes-workspace | 500+ | Native web workspace: chat, terminal, memory browser, skills manager |
| mission-control | 3.7k+ | Agent orchestration dashboard: fleet management, task dispatch, cost tracking |
| hermes-webui | — | Lightweight process monitoring and config dashboard |
| hermes-neurovision | — | Terminal neurovisualizer with 42 animated themes |
Skills Libraries
| Project | Stars | Description |
|---|---|---|
| wondelai/skills | 380+ | Cross-platform skills (Hermes, Claude Code, agentskills.io-compatible) |
| Anthropic-Cybersecurity-Skills | 3.6k+ | 734+ cybersecurity skills mapped to MITRE ATT&CK |
| chainlink-agent-skills | — | Chainlink blockchain oracle skills on agentskills.io spec |
| black-forest-labs/skills | — | FLUX image generation skills — by the FLUX authors |
| hermes-agent-skills | — | 28 general-purpose skills |
| cognify-skills | — | 19 business operations skills: CRM, invoicing, project management |
| traction-skills | — | 19 growth channel playbooks as agent skills |
| hermes-incident-commander | — | Autonomous SRE agent for production incident detection and healing |
| hermes-life-os | — | Personal OS agent: detects patterns, learns routines |
Plugins
| Project | Description |
|---|---|
| hermes-plugins | Goal management, inter-agent bridge, model selection, cost control |
| hermes-web-search-plus | Multi-provider search routing: Serper, Tavily, Exa, and more |
| hermes-weather-plugin | NWS-grade model imagery, NEXRAD radar, meteorological calculations |
| hermes-cloudflare | Headless browsing via Cloudflare infrastructure |
| hermes-plugin-chrome-profiles | Switch browser tools between Chrome profiles via CDP |
| hermes-payguard | USDC and x402 payment integration |
| evey-bridge-plugin | Claude Code + Hermes side-by-side with shared context |
Multi-Agent and Orchestration
| Project | Description |
|---|---|
| hermes-agent-acp-skill | Multi-agent delegation bridging Hermes, Codex, and Claude Code |
| opencode-hermes-multiagent | 17 specialised agents with defined communication interfaces |
| Ankh.md | TAW Agent × Hermes multi-agent swarm framework |
| zouroboros-swarm-executors | Local executor bridge for Claude Code + Hermes handoff |
| hermes-council | Adversarial multi-perspective MCP council — structured debate before committing |
Deployment
| Project | Description |
|---|---|
| hermes-agent-docker | Minimal Docker sandbox image |
| portainer-stack-hermes | Docker Compose + Portainer + browser-accessible ttyd terminal |
| hermes-autonomous-server | systemd + cron headless Linux server deployment |
| nix-hermes-agent | Nix package and NixOS module for reproducible deployments |
| hermes-alpha | Cloud-deployed Hermes with pre-configured infrastructure templates |
| evey-setup | Opinionated quickstart: free models + 29 plugins, running in minutes |
Domain Applications
| Project | Domain |
|---|---|
| hermes-embodied | Robotics — self-improving via VLA model fine-tuning |
| hermescraft | Minecraft — embodied AI companion with persistent memory |
| hermes-legal | Contract risk analysis (English + Turkish) |
| hermes-startup-architect | Startup to investor-ready kit: market analysis, pitch deck, financials |
| hermes-ai-infrastructure-monitoring-toolkit | Infrastructure monitoring, cost forecasting, headless DevOps |
| hermes-research-agent | Autonomous end-to-end LLM research loops |
| job-scout-agent | Autonomous job hunting and application tracking |
| mercury | Multi-chain blockchain cash flow analysis with WebGL dashboard |
Migration
| Project | Description |
|---|---|
| openclaw-to-hermes | Full migration from OpenClaw: settings, memories, skills, API keys, messaging configs |
Key Links
| Resource | URL |
|---|---|
| GitHub | https://github.com/NousResearch/hermes-agent |
| Official Docs | https://hermes-agent.nousresearch.com/docs/ |
| Skills Hub | https://agentskills.io |
| Community Discord | https://discord.gg/NousResearch |
| Awesome List | https://github.com/0xNyk/awesome-hermes-agent |
| OpenRouter | https://openrouter.ai |
| Nous Portal | https://portal.nousresearch.com |
| Ollama Integration | https://docs.ollama.com/integrations/hermes |
Last verified: April 2026. The ecosystem moves fast — treat star counts and feature lists as approximate. Always check the official docs and repo for the current state.