Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Local AI Agents & Tool Use

PicoClaw Review: Open-Source AI Agent for Small Devices

·5 min readBy Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

PicoClaw (version 0.3.1, released on 3 July 2026) is a free, MIT-licensed personal AI agent from Sipeed that runs as a single Go program on small boards, Android phones and desktops and passes your requests to a language model you configure. It does not run a model itself, so a cloud API key or a local Ollama or vLLM server is required.

Key Takeaways

  • Maintenance: the latest stable release, version 0.3.1, is from 3 July 2026 and the newest commit on the main branch is from 19 August 2026, so no stale-project warning applies.
  • Model: none is included; Ollama, vLLM and many cloud providers are supported.
  • Cost: the software is free under the MIT licence, and you pay only your own model provider, per the site FAQ.
  • Maturity: the README calls it early-stage software and advises against production use before version 1.0.
  • Maker: Sipeed, a hardware company; the README says PicoClaw is not a fork of OpenClaw or NanoBot.

📍 In One Sentence

PicoClaw is a small open-source AI assistant from Sipeed, written in Go, that connects chat apps and tools to a language model of your choice.

💬 In Plain Terms

You install one program on a board, phone or computer, point it at a cloud model or your own Ollama server, and then talk to it from a chat app or a browser page.

📌Note: This review is based on picoclaw.io, the GitHub repository and the documentation site, checked on 1 November 2026. PromptQuorum has not installed, run or benchmarked PicoClaw, so speed, memory use and answer quality are not assessed.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

PicoClaw is a free, MIT-licensed personal AI agent written in Go by the hardware company Sipeed. It is built to run on low-cost Linux boards, old Android phones and ordinary desktops, and it forwards your requests to a language model you choose: a cloud provider, or your own Ollama or vLLM server. This review is based on picoclaw.io, the GitHub repository and the project documentation, checked on 1 November 2026; PromptQuorum has not tested PicoClaw hands-on.

PicoClaw Review: Open-Source AI Agent for Small Devices

What Is PicoClaw?

PicoClaw is a personal AI assistant program written in Go by Sipeed, built to start fast and use little memory on cheap hardware. The README calls it an independent open-source project written from scratch and inspired by NanoBot. Sipeed also makes the RISC-V boards that the project advertises.

The program connects a language model to chat channels and tools. Per the site, it supports more than 16 chat channels, web search, scheduled tasks, long-term memory, MCP and REST API integration, and a gateway mode. The README adds that about 95 per cent of the core code was generated by an AI agent and refined through human review; this is the maker's statement and was not checked.

Get It

PicoClaw is a free download for Linux, macOS, Windows and Android, with Docker and package files as alternatives.

Option
How to get it
Windows, macOS, LinuxDownload from picoclaw.io/download, then open the picoclaw-launcher program to use the browser-based WebUI on your own computer
AndroidAPK from picoclaw.io/download, no Termux needed per the README; a Termux setup is the alternative
DockerDocker Compose files in the GitHub repository
Packages and binariesGitHub Releases: archives for Linux, macOS, Windows and FreeBSD, plus .deb and .rpm packages
Websitepicoclaw.io
Documentationdocs.picoclaw.io
Privacy policyNone found on picoclaw.io or in the repository
Source codegithub.com/sipeed/picoclaw, MIT licence

Version reviewed: 0.3.1, released on 3 July 2026 per GitHub Releases. The macOS build is not notarized by Apple, so macOS asks you to allow it under Privacy & Security on first launch. The download page lists platforms only through JavaScript, so exact installer file names were not seen; the file types above come from GitHub Releases.

Requirements

The site gives 64 MB of RAM as the minimum and 512 MB as the recommended amount, plus an internet connection.

Memory

What the sources say:
The README gives under 10 MB for the core program and notes that recent builds may use 10 to 20 MB

Processors

What the sources say:
x86_64, ARM64, ARMv7, ARMv6, RISC-V, MIPS and LoongArch are named

Operating system

What the sources say:
Linux, macOS, Windows, Android and FreeBSD; the minimum Android version is not stated

Language model

What the sources say:
A provider you configure: a cloud API with a key, or an Ollama or vLLM server without one

Building from source

What the sources say:
Go 1.25 or later, plus Node.js 22 and pnpm for the WebUI launcher

Which models run well on which board is not stated; that depends on the model server, not on PicoClaw.

How to Get Started

The steps follow the README and the site; PromptQuorum has not run them.

  1. 1
    Download PicoClaw
    Why it matters: Get the build for your system from picoclaw.io, or the APK on Android; the sources mention no account.
  2. 2
    Open the launcher or run onboarding
    Why it matters: On a desktop, start picoclaw-launcher and use the WebUI in your browser; in a terminal, run picoclaw onboard to create the workspace.
  3. 3
    Choose a model provider
    Why it matters: Enter an API key for a cloud provider, or point PicoClaw at your Ollama server (for example http://localhost:11434/v1) or your vLLM server.
  4. 4
    Connect a chat channel
    Why it matters: Add Telegram, Discord or another channel in the configuration and message the agent; MCP servers and scheduled tasks are further options in the documentation.

Features Confirmed by the Sources

Every item below comes from the website, the README or the documentation; none has been tested here.

  • Chat channels. More than 16 are listed by the site, including Telegram, Discord, Slack and QQ; the documentation also names DingTalk, Feishu, WeCom and LINE.
  • Providers. Many cloud providers plus Ollama and vLLM, with rule-based model routing that sends simple queries to lighter models.
  • Tools. Web search, cron-based scheduling, MCP servers managed from the command line and the WebUI, a REST API, and a vision pipeline for images and files.
  • Memory. Long-term memory and a JSONL memory store, per the site and the release notes.
  • Portability. One binary for x86_64, ARM, RISC-V, MIPS and LoongArch, plus an Android build.

Data Flow and Privacy

The site FAQ says conversation history and configuration stay on your device, while requests to the language model still go to the provider you configure.

With a cloud provider, your prompts therefore reach that provider under its own policy; with Ollama or vLLM on your own network they stay there. Messages sent through a chat channel such as Telegram also pass through that platform.

📌Note: These are the maker's statements, not audit results. The README warns of possible unresolved security issues, and PromptQuorum has not inspected network traffic or the code.

Who Should Use It

  • Raspberry Pi and RISC-V owners who want a small agent on a board they already have.
  • Ollama or vLLM users who want chat-app access to a model on their own network.
  • Tinkerers with an old Android phone who want to try an always-on assistant without extra hardware.

What We Could Not Verify

  • Production readiness. The README says PicoClaw is in early rapid development, may have unresolved security issues and should not be deployed to production before version 1.0.
  • Memory claims. The headline figure of under 10 MB and the comparison with OpenClaw and NanoBot are the maker's own; none was measured here.
  • Android details. The minimum Android version, APK signing and which features work on a phone are not stated.
  • Data handling. No privacy policy or data-handling document was found, and network traffic was not inspected.
  • Local-model quality. How well small boards and local models handle agent tasks is untested.

Competitors and Alternatives

nanobot

Platforms:
Windows, macOS, Linux
Price / license:
Free / MIT
Key difference:
Self-hosted agent framework with memory, MCP tools and chat channels
Articles about nanobot (4)

OpenClaw

Platforms:
Windows, macOS, Linux
Price / license:
Free / MIT
Key difference:
Orchestration framework for agent workflows
Articles about OpenClaw (6)

Also mentioned in:

+4 more not shown

Confirm each app's current price, license and platforms on its own listing before choosing.

Compare in directory

Frequently Asked Questions

Is PicoClaw a fork of OpenClaw?

No. The README calls it an independent project written from scratch in Go, inspired by NanoBot, and says it is not a fork of OpenClaw, NanoBot or any other project.

Which local models does PicoClaw work with?

The README shows configuration examples for Ollama (such as llama3.1:8b) and vLLM through their OpenAI-compatible addresses, and the site says any OpenAI-compatible API works.

Is there an official PicoClaw cryptocurrency?

No. The README says PicoClaw has issued no tokens or cryptocurrency, that picoclaw.io is the only official website, and that many lookalike domains were registered by third parties.

Verdict

PicoClaw is an MIT-licensed way to put a small AI agent on cheap boards, old phones and desktops and connect it to your own Ollama or vLLM server or a cloud model. Against that, the maker itself calls it early-stage, data handling is described only in a short FAQ answer, and PromptQuorum has not tested speed, memory use or Android behaviour. It suits tinkerers who can try it on spare hardware; it should not carry anything that must be reliable or secure until the project says it is ready.

Sources

  • PicoClaw website — product overview, FAQ, hardware figures and footer links, checked 1 November 2026.
  • PicoClaw download page — download sources and the onboarding command, checked 1 November 2026.
  • PicoClaw on GitHub — README, MIT licence, security notice and commit history, checked 1 November 2026.
  • PicoClaw releases — version 0.3.1 of 3 July 2026 and its file list, checked 1 November 2026.
  • PicoClaw documentation — chat channels and configuration guides, checked 1 November 2026.

← Back to Power Local LLM

MCP