sipeed/picoclaw

PicoClaw is an AI assistant agent written in Go, designed to run on extremely low-cost and low-power hardware. While most AI agent tools require a powerful computer with gigabytes of RAM, PicoClaw's core footprint is under 10 megabytes of memory, which means it can run on devices that cost as little as ten dollars, such as small single-board computers using RISC-V, ARM, or MIPS processors, and starts up in under one second even on a slow processor. The program acts as a personal AI agent that you can connect to various large language model providers (such as those from AWS Bedrock, Azure, or others). It supports text and image inputs, can search the web, route different queries to different models based on rules you define (for example, sending simple questions to cheaper models), and maintains a memory store so it can remember context across sessions. It integrates with messaging platforms including Discord, Matrix, WeChat, and WeCom, and supports the Model Context Protocol, a standard for connecting AI agents to external tools and data sources. Because it compiles to a single binary that runs on multiple processor architectures, you can deploy the same executable to a tiny embedded Linux board, a Raspberry Pi, an Android phone, or a conventional server without changes. Someone who wants a lightweight, self-hosted AI assistant on cheap hardware, or who wants to automate tasks using AI without depending on cloud-heavy infrastructure, would use PicoClaw.