Skip to content
IFAI

Open source

PicoClaw squeezes an always-on AI agent into 10MB of RAM

Published Reach 48

Open-source developers just launched PicoClaw, an ultra-lightweight AI assistant framework designed to replace bulky setups like OpenClaw. Built from scratch in Go, it uses under 10MB of RAM and boots in less than a second on $10 single-board computers. You can hook it up to Telegram or WhatsApp to handle tasks 24/7 without burning cash on hardware.

PicoClaw squeezes an always-on AI agent into 10MB of RAM
Image: github.com

PicoClaw is an ultra-lightweight open-source personal AI assistant framework that lets you run a 24/7 AI agent on almost any dirt-cheap hardware, taking orders through chat apps like Telegram and WhatsApp.

If you have ever tried running OpenClaw or similar agent projects, you already know the pain. They are often resource-heavy monsters. OpenClaw relies on TypeScript and can easily swallow over 1GB of RAM, forcing people to dedicate a whole Mac mini or rent pricey VPS instances just to keep an agent awake around the clock.

What it does

PicoClaw handles the exact same core job without the bloat. You leave it running in the background, send it a message on Telegram or WhatsApp, and let the AI browse the web, run tasks, or trigger local scripts for you.

Despite the tiny footprint, it does not strip away key agent capabilities. It features native Model Context Protocol support to link external tool servers, a vision pipeline that feeds images straight to multimodal LLMs, and smart model routing to redirect easy questions to cheaper models so your API bills stay low.

Recent community builds have also added support for Matrix, IRC, WebCom, a simple Web UI manager, and desktop system tray controls.

How it works and hardware requirements

PicoClaw is not a fork of OpenClaw. It was written from scratch in pure Go. In an interesting twist, Sipeed notes that 95% of its core codebase was generated by an AI agent through a self-bootstrapping process with human review.

The entire tool builds into a single standalone binary. It runs natively across x86, ARM, RISC-V, and MIPS boards, and there is even an Android APK if you prefer running it on a spare phone.

On a single-core 0.8GHz processor, PicoClaw boots in less than one second and keeps its core memory usage below 10MB. Compared to OpenClaw, that is a 99% reduction in memory and 400 times faster boot speed, cutting your hardware barrier from a $599 desktop down to a $10 Linux board.

You can grab the code from GitHub under sipeed/picoclaw or check documentation at picoclaw.io. Note that the team warns recent feature additions may push RAM into the 10MB to 20MB range temporarily, and because it is still pre-v1.0, you should not deploy it in production environments yet. Also, watch out for fake crypto scams, as the project has no official tokens.

Why it matters

If you have been looking for an always-on personal bot to run daily chores without running up a big electric or cloud bill, PicoClaw is the most compelling DIY tool right now. It drops the hardware floor straight into scrap-drawer territory, letting you revive dusty single-board computers or old smartphones.

We think shifting agent runtimes into pure Go single binaries is a brilliant move that exposes how unnecessarily bloated early agent software was. It is still in active early development, so you will run into rough edges and ongoing security tweaks, but for hobbyists looking for a snappy, self-hosted chat assistant, it is an absolute steal.

Sources (1)

  1. [1]
    github.com(opens in a new tab)Originalgithub.com

Best of AI, daily