p:: AI f:: Claude Code, OpenClaw
Self-hosted personal AI agent from Nous Research (MIT). One process gives you a terminal UI plus a messaging gateway (Telegram, Discord, Slack, WhatsApp, Signal, Email), and the agent keeps its own memory and skills across sessions. Model-agnostic: Nous Portal, OpenRouter, OpenAI, Anthropic or any OpenAI-compatible endpoint.
Install
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash # Linux, macOS, WSL2
source ~/.zshrc
hermesInstalls under ~/.hermes.
Commands
| Command | What it does |
|---|---|
hermes | Interactive TUI |
hermes setup | Full setup wizard (--portal = log in to Nous Portal for models + tool gateway) |
hermes model | Pick provider and model |
hermes tools | Enable/disable toolsets |
hermes config get / set | Read or write single config values |
hermes gateway setup / start | Configure and run the messaging gateway |
hermes update | Update to latest |
hermes doctor | Diagnose problems |
Slash commands
Same in the TUI and in chat apps.
| Command | What it does |
|---|---|
/new, /reset | Fresh conversation |
/model [provider:model] | Switch model mid-session |
/retry, /undo | Redo or drop the last turn |
/compress, /usage, /insights | Shrink context, check token use |
/<skill-name> | Run a skill directly |
/stop (chat) / Ctrl+C (CLI) | Interrupt; sending a new message also redirects |
/sethome, /status | Set the chat’s home channel, gateway status |
What sets it apart
- Learning loop. It writes skills after hard tasks and patches them as it uses them. It also keeps a small curated memory and can search past sessions. Skills follow the agentskills.io format.
- Cron. Scheduled jobs written in plain language that deliver to any connected platform.
- Subagents. Spawns isolated child agents for parallel work. It can also run Python that calls tools over RPC, so a multi-step pipeline takes one turn.
- Terminal backends. Local, Docker, SSH, Singularity, Modal, Daytona, Vercel Sandbox. Modal and Daytona hibernate when idle.
- MCP. Connects to any MCP server.