# Hermes
Self-improving personal AI assistant (Nous Research) that connects to Telegram and routes queries to your chosen LLM provider. Docs: [hermes-agent.nousresearch.com](https://hermes-agent.nousresearch.com/docs)
- **URL**: no public URL — outbound polling only, no inbound port
- **Data**: `~/.hermes/` on the VPS
## Deploy
```bash
auberge deploy hermes
```
Dependency layers (hardening, infrastructure) resolved automatically.
## Required config
| `hermes_llm_provider` | Provider slug: `anthropic`, `deepseek`, `gemini`, `kimi`, `openrouter`, `xai`, `zai` |
| `hermes_llm_api_key` | API key for the chosen provider |
| `hermes_telegram_bot_token` | Bot token from [@BotFather](https://t.me/BotFather) |
Optional: `hermes_exa_api_key` (web search), `hermes_telegram_allowed_users` (comma-separated user IDs — restricts bot access; without this any user knowing the token can interact).
## Notes
?> `hermes_llm_provider` must match the provider set in `~/.config/hermes/config.yaml`, which is synced to the VPS on deploy.
After deploy, verify the service and send a test message to your bot:
```bash
systemctl --user status hermes-gateway
journalctl --user -u hermes-gateway -f
```
Secrets stored in `~/.hermes/.env` (mode `0600`). No public ports exposed.