Hermes Agent — Open Source Project Introduction

Repository: https://github.com/NousResearch/hermes-agent

1. Project Overview

Hermes Agent is a self-improving, open-source AI agent built by Nous Research that autonomously creates and refines skills from experience, maintains persistent memory across sessions, and runs on any infrastructure — from a $5 VPS to a GPU cluster — accessible via terminal or popular messaging platforms.

2. Project Background & Positioning

Origin & Core Mission

Most AI agents are stateless — they forget every conversation, repeat the same mistakes, and lock you into a single model or cloud provider. Hermes was built to fix that. Its core mission is to be the agent that grows with you: one that accumulates procedural knowledge (skills), builds a persistent model of the user over time, and continuously improves itself without requiring manual intervention.

How It Differs from Alternatives

Dimension	Hermes Agent	Typical Alternatives
Learning loop	Autonomous skill creation & self-improvement	None (stateless)
Model lock-in	Any provider via OpenRouter, OpenAI, Anthropic, etc.	Often single-provider
Infrastructure	Local, Docker, SSH, Modal, Daytona, Singularity	Usually local only
Interfaces	CLI + Telegram, Discord, Slack, WhatsApp, Signal	CLI or web UI
Memory	FTS5 session search + LLM summarization + user modeling	Simple context window
RL integration	Atropos RL environment for training tool-calling models	Not applicable

3. Feature Categories

🧠 Learning & Memory System

~5 core subsystems

Autonomous skill creation — agent builds reusable skills after completing complex tasks
Skill self-improvement — skills refine themselves during use
FTS5 session search — full-text search across all past conversations with LLM summarization
Honcho dialectic user modeling — builds a deepening profile of the user across sessions
AgentSkills.io compatibility — open standard for sharing skills across agents

Purpose: Give the agent genuine long-term memory and procedural knowledge.

🛠 Tools & Toolsets

40+ tools available

File system operations, web browsing, code execution
Voice memo transcription
Subagent spawning for parallel workstreams
Python RPC scripts that invoke tools directly
Configurable toolset distributions per use case

Purpose: Equip the agent to handle real-world tasks autonomously.

💬 Messaging Gateway

6+ platform integrations

Telegram, Discord, Slack, WhatsApp, Signal, Email
Single gateway process serves all platforms simultaneously
Cross-platform conversation continuity
Voice memo transcription across platforms

Purpose: Talk to your agent from anywhere, not just the terminal.

⏰ Cron Scheduling

Natural language task scheduling

Daily reports, nightly backups, weekly audits
Delivery to any connected platform
Runs unattended on remote infrastructure

Purpose: Automate recurring tasks without writing cron syntax.

🖥 Terminal Backends

6 execution environments

Local, Docker, SSH, Daytona, Modal, Singularity
Modal and Daytona offer serverless persistence — hibernates when idle
Android/Termux supported

Purpose: Run the agent anywhere, at any scale, with minimal idle cost.

🔬 Research & RL Integration

Batch trajectory generation
Atropos RL environment integration
Trajectory compression for training next-generation tool-calling models

Purpose: Enable AI researchers to use Hermes as a data engine for model training.

4. Core Highlights

Closed learning loop — The only open-source agent that autonomously creates, stores, and self-improves skills based on its own task history.
Zero model lock-in — Switch providers with a single command (hermes model). Supports Nous Portal, OpenRouter (200+ models), OpenAI, Anthropic, MiniMax, Kimi, GLM, and custom endpoints.
Runs where you're not — Deploy on a remote VPS or serverless infrastructure; interact via Telegram while the agent executes on the cloud.
MCP-native — Connect any Model Context Protocol server for extended, standardized tool capabilities.
Research-grade trajectory tooling — Built-in batch runner, RL environment adapter, and trajectory compressor for training data generation.
Full TUI — Multiline editing, slash-command autocomplete, streaming tool output, interrupt-and-redirect — not just a chatbot wrapper.

5. Use Cases by Role

👨‍💻 General Developers

Use Hermes as a persistent coding assistant that remembers your project context, coding style, and past decisions. Run it alongside your editor via the CLI or connect it to Slack for team-wide access.

⚙️ DevOps / Infrastructure Engineers

Schedule automated health checks, log summaries, and deployment reports via the cron system. Deploy on Modal or Daytona for serverless, near-zero-cost background agents.

🔒 Security Engineers

Leverage isolated Docker or Singularity backends for sandboxed code execution. Use the command approval system to enforce allowlists before any shell command runs.

📊 Data / AI Researchers

Use the batch trajectory runner and Atropos RL integration to generate training datasets for tool-calling models. The trajectory compressor reduces token cost while preserving signal.

🗂 Project Managers

Set up Telegram or Discord bots that provide daily project status reports, summarize discussion threads, and send scheduled reminders — all in natural language.

6. Quick Start

🔍 How to Find Resources

Official documentation: https://hermes-agent.nousresearch.com/docs/
Skills Hub (community skills): https://agentskills.io
GitHub Discussions: https://github.com/NousResearch/hermes-agent/discussions

🛠 How to Install & Use

# Install (Linux, macOS, WSL2, Termux)
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

# Reload shell
source ~/.bashrc   # or source ~/.zshrc

# Start chatting
hermes

# Key commands
hermes model       # Select LLM provider and model
hermes tools       # Configure enabled tools
hermes setup       # Full interactive setup wizard
hermes gateway     # Start messaging gateway (Telegram, Discord, etc.)
hermes doctor      # Diagnose configuration issues
hermes update      # Update to latest version

🤝 How to Contribute

# Clone and set up development environment
git clone https://github.com/NousResearch/hermes-agent.git
cd hermes-agent

# Install uv (fast Python package manager)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Create virtual environment and install dependencies
uv venv venv --python 3.11
source venv/bin/activate
uv pip install -e ".[all,dev]"

# Run tests
python -m pytest tests/ -q

See the full contributing guide: https://hermes-agent.nousresearch.com/docs/developer-guide/contributing

7. Project Structure

hermes-agent/
├── agent/                  # Core agent loop and orchestration logic
├── skills/                 # Built-in skills (procedural memory units)
├── optional-skills/        # Community and optional skill packs
├── tools/                  # Tool implementations (40+ tools)
├── gateway/                # Messaging platform gateway (Telegram, Discord, etc.)
├── cron/                   # Cron/scheduler subsystem
├── plugins/                # Plugin system
├── environments/           # Terminal backend adapters (Docker, SSH, Modal, etc.)
├── acp_adapter/            # Agent Communication Protocol adapter
├── acp_registry/           # ACP service registry
├── hermes_cli/             # CLI entry point and TUI
├── plans/                  # Agent planning subsystem
├── docker/                 # Docker configuration
├── nix/                    # Nix flake for reproducible builds
├── tinker-atropos/         # RL environment submodule (Atropos integration)
├── docs/                   # Documentation source
├── tests/                  # Test suite
├── cli.py                  # Main CLI entry point
├── run_agent.py            # Agent runner
├── mcp_serve.py            # MCP server entry point
├── toolsets.py             # Toolset definitions and distributions
├── hermes_state.py         # Global agent state management
├── trajectory_compressor.py # RL training data compression utility
├── batch_runner.py         # Batch trajectory generation
└── Dockerfile              # Container build definition

8. Related Ecosystem

Upstream Dependencies & Platforms

OpenRouter — Multi-provider LLM routing (200+ models)
Nous Portal — Nous Research's own model endpoint
Model Context Protocol (MCP) — Standardized tool/server integration
Honcho (plastic-labs) — Dialectic user modeling backend
Modal / Daytona — Serverless and cloud dev environment platforms
Atropos — RL training environment framework

Complementary Projects

agentskills.io — Open standard and hub for sharing agent skills
Tinker-Atropos — RL environment submodule for tool-calling model training
OpenClaw — Predecessor project; Hermes provides a migration path

9. License

MIT License

✅ Free to use commercially
✅ Free to modify and distribute
✅ Free to use in private projects
✅ Free to sublicense
❌ Authors provide no warranty
❌ Cannot hold authors liable for damages
ℹ️ License and copyright notice must be included in distributions
ℹ️ Contributions are welcome under the same MIT terms

10. Frequently Asked Questions

Q: Which LLM providers are supported?
A: Any OpenAI-compatible endpoint, plus native support for Nous Portal, OpenRouter (200+ models), Anthropic, OpenAI, MiniMax, Kimi/Moonshot, and z.ai/GLM. Switch with hermes model.

Q: Does Hermes work on Windows?
A: Native Windows is not supported. Install WSL2 and use the standard Linux install command. Android is supported via Termux.

Q: How does the skills system work?
A: After completing complex tasks, the agent autonomously writes a reusable skill (a structured procedure). Skills are stored in ~/.hermes/skills/, improve during subsequent uses, and can be shared via agentskills.io.

Q: Can I run Hermes without keeping my laptop on?
A: Yes. Deploy on any Linux VPS, or use Modal/Daytona for serverless environments that hibernate when idle and wake on demand — costing nearly nothing between sessions.

Q: How do I migrate from OpenClaw?
A: Run hermes claw migrate after installation. It imports memories, skills, API keys, messaging settings, and persona files. Use --dry-run to preview first.

11. Quick Links

Resource	URL
GitHub Repository	https://github.com/NousResearch/hermes-agent
Official Documentation	https://hermes-agent.nousresearch.com/docs/
Quickstart Guide	https://hermes-agent.nousresearch.com/docs/getting-started/quickstart
Contributing Guide	https://hermes-agent.nousresearch.com/docs/developer-guide/contributing
Discord Community	https://discord.gg/NousResearch
Skills Hub	https://agentskills.io
GitHub Issues	https://github.com/NousResearch/hermes-agent/issues
GitHub Discussions	https://github.com/NousResearch/hermes-agent/discussions

12. Summary

Hermes Agent stands apart from the growing field of AI agents by solving a fundamental problem: agents that never learn. Its closed learning loop — autonomous skill creation, skill self-improvement, persistent cross-session memory, and user modeling — makes it genuinely more capable the longer you use it. With 61k+ GitHub stars, support for 200+ LLM models, 40+ tools, six execution backends, and integration with every major messaging platform, it is one of the most complete open-source agent frameworks available today. It is best suited for developers, researchers, and power users who want a persistent, self-improving AI agent they can deploy once and rely on indefinitely — without vendor lock-in.