NanoBot

NanoBot is an ultra-lightweight Python AI agent sandbox. Where other sandboxes focus on desktop and coding workflows, NanoBot is built for channel-based AI — connecting to Telegram, Discord, Slack, WhatsApp, and other messaging platforms with minimal overhead.

Port: 18790 | Token type: Gateway Token | RAM: 2 GB | Best for: Channel bots, Telegram/Discord agents, MCP-connected assistants

What starts on boot

1. NanoBot onboard — creates ~/.nanobot/config.json with defaults on first boot.
2. Proxy injection — OpenRouter and Anthropic providers are automatically configured to route through the AgentStead LLM proxy using your proxy token. No API key setup needed.
3. Health server — a lightweight HTTP server starts on port 18790 at /healthz.
4. NanoBot gateway — connects to any channels you've configured (Telegram, Discord, etc.) and starts the agent loop.

Data directory

NanoBot stores config, workspace, and cron jobs in ~/.nanobot/ inside your persistent volume.

~/.nanobot/
├── config.json        # provider keys, channel tokens, model settings
├── workspace/
│   ├── AGENTS.md      # agent persona / instructions
│   ├── HEARTBEAT.md   # scheduled task definitions
│   ├── SOUL.md        # agent values and constraints
│   └── memory/        # persistent memory files
└── cron/              # scheduled job definitions

Configuring channels

Edit ~/.nanobot/config.json from the desktop terminal to enable channels. The gateway picks up changes on next restart.

Telegram

{
  "channels": {
    "telegram": {
      "enabled": true,
      "token": "YOUR_BOT_TOKEN",
      "allowFrom": ["YOUR_TELEGRAM_USER_ID"]
    }
  }
}

Get a bot token from @BotFather on Telegram, then restart the gateway:

pkill -f "nanobot gateway" && nanobot gateway &

Discord

{
  "channels": {
    "discord": {
      "enabled": true,
      "token": "YOUR_BOT_TOKEN",
      "allowFrom": ["YOUR_DISCORD_USER_ID"],
      "groupPolicy": "mention"
    }
  }
}

Other channels

NanoBot supports WhatsApp, Slack, Email, DingTalk, Feishu, QQ, WeCom, and Matrix. See the NanoBot channel docs for setup instructions for each.

LLM provider setup

The AgentStead proxy is injected automatically on boot — OpenRouter and Anthropic are pre-configured. The default model is anthropic/claude-sonnet-4-6 via OpenRouter.

To change the model, update config.json:

{
  "agents": {
    "defaults": {
      "model": "anthropic/claude-opus-4-6",
      "provider": "openrouter"
    }
  }
}

You can also add a direct Anthropic API key in Secrets and point the provider at it instead of the proxy.

MCP tools

NanoBot supports MCP (Model Context Protocol) servers, giving your agent access to external tools and data sources.

{
  "tools": {
    "mcpServers": [
      {
        "name": "filesystem",
        "command": "npx",
        "args": ["-y", "@modelcontextprotocol/server-filesystem", "/home/kasm-user"]
      }
    ]
  }
}

ClawHub skills

NanoBot integrates with ClawHub — a directory of installable agent skills. Ask the agent to install a skill:

Read https://clawhub.ai/skills/some-skill/skill.md and install it.

Useful commands

nanobot --version          # check version
nanobot status             # show config and provider status
nanobot agent -m "Hello"   # send a one-off message via CLI
cat ~/.nanobot/config.json # view current config

When to use NanoBot

NanoBot is the right choice when:

• You want a Telegram, Discord, or Slack bot powered by Claude
• You need lightweight resource usage — NanoBot uses half the RAM of other sandboxes
• You're building scheduled automations via NanoBot's cron and heartbeat system
• You want MCP tool integrations with a simple Python-based sandbox