PicoClaw: Ultra‑Lightweight Go AI Assistant for <$10 Hardware

Introduction

If you’ve been hunting for a tiny, high‑performance AI assistant that fits inside a pocket‑sized board or even an old Android phone, PicoClaw is the project for you. Built entirely in Go, PicoClaw was designed from scratch to be both ultra‑efficient and fully autonomous. It boots in under one second, consumes less than 10 MB of RAM, and can run on any $10 board or even a repurposed Android phone.

What Makes PicoClaw Tick?

Quick Start (Linux or macOS)

# 1️⃣ Clone the repo
git clone https://github.com/sipeed/picoclaw.git
cd picoclaw

# 2️⃣ Install dependencies (Go required)
make deps

# 3️⃣ Build the binary
make build

# 4️⃣ Run the assistant
./cmd/picoclaw agent -m "Hello, PicoClaw"

Alternatively, download a pre‑compiled binary for your platform from the Releases page. Just chmod +x it and run picoclaw agent.

Deploying on an Old Android Phone

PicoClaw can be turned into a smart assistant on a decade‑old Android phone using Termux.

# Install Termux from F‑Droid or Play Store
# Download the ARM64 binary
wget https://github.com/sipeed/picoclaw/releases/download/vX.Y.Z/picoclaw-linux-arm64
chmod +x picoclaw-linux-arm64

# Install supporting packages
pkg install proot

# Run inside Termux
./picoclaw-linux-arm64 onboard  # Initializes workspace and config
./picoclaw-linux-arm64 agent -m "What’s the weather?"

PicoClaw also works on phones that run Linux via Termux‑Chroot or via pRoot. The same commands apply; you now have a pocket‑size AI.

Docker Compose Deployment

For developers who prefer containerization, Picoclaw supports Docker Compose:

# docker-compose.yml
version: "3"
services:
  picoclaw-gateway:
    image: ghcr.io/sipeed/picoclaw:latest
    profiles: [gateway]
    environment:
      - DISCORD_BOT_TOKEN=your-token
    volumes:
      - ./config/config.json:/app/config/config.json:ro
      - ./workspace:/app/workspace

1. Commit your config.json with API keys and channel credentials.
2. Run docker compose --profile gateway up -d.
3. The agent will now listen to Discord, Telegram, or any channel you configured.

Chat Apps Integration

PicoClaw natively exposes a gateway that can be plugged into standard messaging platforms. The same configuration file is used:

{
  "channels": {
    "telegram": {
      "enabled": true,
      "token": "YOUR_TELEGRAM_TOKEN",
      "allow_from": ["YOUR_USER_ID"]
    },
    "discord": {
      "enabled": true,
      "token": "YOUR_DISCORD_TOKEN",
      "allow_from": ["YOUR_USER_ID"]
    }
  }
}

Running picoclaw gateway will spawn a lightweight HTTP server that forwards messages to the AI bot, processes them, and returns replies. The security sandbox ensures the bot can only touch files inside its workspace.

Advanced Features

Periodic Tasks (Heartbeats)

Create a HEARTBEAT.md in your workspace:

# Periodic Tasks
- Report the current weather every hour
- Check email for promotions

PicoClaw’s heartbeat routine polls this file every 30 minutes (configurable) and executes the tasks asynchronously, publishing results directly to the channel.

Sub‑Agents and Spawn

Long‑running actions (e.g., web search) can be delegated to a sub‑agent using the spawn tool. The sub‑agent operates in its own context and can communicate back with the user via the message tool, enabling non‑blocking workflows.

Model‑Centric Configuration

PicoClaw’s model_list lets you plug in any OpenAI‑compatible or proprietary LLM without writing new code. For example:

{
  "model_list": [
    { "model_name": "gpt-4o-mini", "model": "openai/gpt-4o-mini", "api_key": "sk-…" },
    { "model_name": "llama3", "model": "ollama/llama3" }
  ]
}

The agent prefers the primary model and falls back to the next if it fails.

Security & Sandbox

PicoClaw enforces a workspace‑only policy by default:

{
  "agents": { "defaults": { "restrict_to_workspace": true } }
}

All file‑system and exec tools are sandboxed to ~/.picoclaw/workspace. Even when the sandbox is disabled, dangerous commands (rm -rf, shutdown, etc.) are explicitly blocked.

Community & Contributions

With more than 16 k stars, PicoClaw has become a community favorite for lightweight AI experiments. Contributions are welcome—especially around:

• New tool integrations (e.g., custom database connectors)
• Improved deployment scripts for niche SBCs
• Security audits and hardening
• Documentation in new languages

The project is managed through GitHub Discussions and a dedicated Discord channel. Contributions also help support the roadmap, which includes further memory optimizations, native support for other operating systems, and tighter integration with popular AI workflows.

Final Thoughts

PicoClaw proves that a powerful conversational AI is no longer confined to high‑end GPUs or expensive cloud servers. By combining a tiny Go binary, a sandboxed runtime, and a flexible configuration system, the project turns any inexpensive microcontroller or old phone into a fully functional personal assistant. Whether you’re looking to prototype micro‑AI on a $10 board, add an AI companion to a home server, or experiment with autonomous agents, PicoClaw gives you the tools to do it quickly, safely, and for free.

Check out the GitHub repository today and start building your own low‑cost AI assistant tomorrow.