Skip to content
Get packs

Guide · Open source

How to use PicoClaw: clone, set up and run the AI assistant

Updated · written with AI from the project's own documentation and checked against our catalogue data

PicoClaw is an open-source personal AI assistant written in Go. The project says it is designed to use very little memory and run on low-cost boards, and it is inspired by NanoBot. This guide walks through getting the source code, what the project supports, and what to watch out for before you rely on it.

PicoClawsipeed/picoclaw · ★ 30.1k · MIT

Quick start

  1. 1Check the project status. Read the README warnings on security and early development before you start. Do not plan a production use before v1.0.
  2. 2Clone the repository. Run the git clone command from the README to download the PicoClaw source code to your machine.
  3. 3Read the official docs. Open docs.picoclaw.io for setup details, since the README source here lists no prerequisites or build steps.
  4. 4Pick suitable hardware. Check the hardware compatibility list if you plan to run on a low-cost board such as a RISC-V or ARM single-board computer.
  5. 5Test a non-critical workflow. Try a simple task first, such as a personal automation, and watch memory use since recent builds may use 10-20MB.

What is PicoClaw and who is it for?

PicoClaw is an ultra-lightweight personal AI assistant that aims to automate everyday tasks through an AI agent. The README describes it as an independent project initiated by Sipeed, written in Go, and it is licensed under MIT.

It suits hobbyists who want an assistant on a low-cost Linux board, developers experimenting with lightweight agents in Go, and people with limited hardware such as RISC-V or ARM single-board computers.

  • Best for: hobbyists on low-cost Linux boards
  • Best for: developers testing lightweight AI agents in Go
  • Best for: users who want to connect MCP servers to an assistant
  • Not for: production deployments (the README says not to deploy before v1.0)
  • Not for: non-technical users wanting a polished consumer app

Get the PicoClaw source code

The README's install instruction is to clone the repository. The project lists no prerequisites in this source, so check the official documentation at docs.picoclaw.io for build and setup details beyond what is shown here.

The official website is picoclaw.io. The README warns that other domains using .ai, .org, .com or .net may be registered by third parties, so use only the official site.

git clone https://github.com/sipeed/picoclaw

Know the memory and stability limits

The README claims a core memory footprint under 10MB, but it also notes that recent builds may use 10-20MB of RAM because of rapid pull request merges. Resource optimisation is planned after features stabilise, so do not assume the lower figure for every build.

The README also says PicoClaw is in early rapid development and that there may be unresolved security issues. Treat it as experimental.

  • Memory: the README claims <10MB core, but recent builds may use 10-20MB
  • Status: early development; do not deploy to production before v1.0
  • Security: the README says unresolved security issues may remain

Understand the main features

The README lists several features you can plan around. MCP support lets you connect any Model Context Protocol server to extend the agent's capabilities. The vision pipeline accepts images and files and automatically base64-encodes them for multimodal LLMs.

Smart routing is rule-based: simple queries can go to lightweight models, which the README says saves API costs.

  • MCP support: connect Model Context Protocol servers to extend the agent
  • Vision pipeline: send images and files to the agent
  • Smart routing: rule-based model routing to lighter models for simple queries
  • Portable: the README describes a single binary across RISC-V, ARM, MIPS and x86

Compare PicoClaw with NanoBot and OpenClaw

The README compares PicoClaw with OpenClaw and NanoBot. PicoClaw is written in Go, while OpenClaw is TypeScript and NanoBot is Python. The README's figures are its own claims and are based on its stated test conditions, so verify them yourself if they matter to you.

The project also links a Hardware Compatibility List in docs/guides/hardware-compatibility.md for tested boards.

What people use it for

  • Running a personal assistant on a low-cost Linux single-board computer
  • Connecting MCP servers to extend what the agent can do
  • Sending images and files to a multimodal model through the vision pipeline
  • Experimenting with lightweight AI agent design in Go
  • Routing simple queries to lighter models to reduce API costs

Frequently asked questions

Is PicoClaw free to use?

PicoClaw is open source and licensed under MIT, according to the README. The README also warns that the project has not issued any official tokens or cryptocurrency, and that such claims are scams.

What do I need to install PicoClaw?

The source lists no prerequisites. The README's install instruction is to clone the repository from GitHub. Check docs.picoclaw.io for build and setup requirements.

Is PicoClaw safe to use in production?

No, according to the README. It says not to deploy to production before v1.0 and that there may be unresolved security issues.

How does PicoClaw compare to NanoBot?

The README says PicoClaw is inspired by NanoBot but written from scratch in Go, while NanoBot is Python. The README reports lower memory use and faster boot for PicoClaw, based on its own benchmarks.

Why does PicoClaw use more memory than advertised?

The README says recent builds may use 10-20MB of RAM due to rapid pull request merges. It plans resource optimisation after features stabilise.

Which hardware can run PicoClaw?

The README says it runs on low-cost boards and mentions RISC-V, ARM, MIPS and x86 architectures. It links a Hardware Compatibility List in docs/guides/hardware-compatibility.md for tested boards.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

PicoClaw suits hobbyists and developers who want to experiment with a tiny Go AI agent on cheap hardware. Pick something else if you need a production-ready system, guaranteed low memory use, or a polished consumer app with simple setup.

Install options & downloads