Nanobot

Nanobot is an MCP-native agent runtime. It reads its model configuration from ~/.nanobot/config.json, where you register a provider, name a model preset, and point the agent defaults at it. Router9 plugs in as a custom OpenAI-compatible provider.

Install

python -m pip install nanobot-ai
nanobot onboard --wizard

nanobot onboard creates ~/.nanobot/config.json if it doesn't exist yet.

Setup

Add Router9 as a custom provider, define a model preset that points at it, and select that preset in your agent defaults. Merge this into ~/.nanobot/config.json:

{
  "providers": {
    "router9": {
      "apiBase": "https://api.router9.com/v1",
      "apiKey": "${ROUTER9_API_KEY}"
    }
  },
  "modelPresets": {
    "primary": {
      "label": "Router9",
      "provider": "router9",
      "model": "gpt-4o",
      "maxTokens": 8192,
      "contextWindowTokens": 128000,
      "temperature": 0.1
    }
  },
  "agents": {
    "defaults": {
      "modelPreset": "primary"
    }
  }
}

A custom provider key like router9 is treated as a direct OpenAI-compatible provider, so apiBase is required — include the /v1 version path. Choose a name that doesn't collide with a built-in provider (openai, anthropic, ollama, etc.), and leave apiType unset.

Environment Variables

Set your Router9 key so the ${ROUTER9_API_KEY} reference resolves:

export ROUTER9_API_KEY=sk-r9k-your-key-here

Environment variables set this way apply only to the current terminal. For long-running services (systemd, Docker, a remote shell), set the variable in that service's environment before starting nanobot.

Verify

nanobot status
nanobot agent -m "Hello!"

If the CLI reply works, you can go on to connect the WebUI, gateway, or chat apps.

Choosing Models

The model field inside a preset is the id Router9 routes on. Add more presets to expose several models, and switch between them at runtime with /model:

{
  "modelPresets": {
    "fast": {
      "label": "Fast",
      "provider": "router9",
      "model": "gpt-4o",
      "maxTokens": 8192,
      "contextWindowTokens": 128000
    },
    "deep": {
      "label": "Deep",
      "provider": "router9",
      "model": "claude-sonnet-4-20250514",
      "maxTokens": 8192,
      "contextWindowTokens": 200000
    }
  },
  "agents": {
    "defaults": {
      "modelPreset": "fast"
    }
  }
}

Swap the model id for any model Router9 supports. In a chat surface, /model deep switches presets for the next turn without editing the config.

Fallback Models

Named presets double as fallback targets. List them under agents.defaults.fallbackModels; nanobot tries the active preset first, then each fallback in order:

{
  "agents": {
    "defaults": {
      "modelPreset": "fast",
      "fallbackModels": ["deep"]
    }
  }
}

Tips

• Nanobot is MCP-native — expose Router9 Skills as tools by adding Router9's hosted MCP server.
• Router9's flat monthly pricing suits Nanobot's long-running, tool-heavy sessions.
• Keep your sk-r9k- key in the environment rather than committing it to config.json.