PicoClaw

PicoClaw is an ultra-lightweight personal AI assistant written from scratch in Go, initiated by Sipeed and inspired by NanoBot (it is not a fork of OpenClaw or NanoBot). Its goal is to bring AI agents to very constrained and inexpensive hardware: it targets a core memory footprint under 10MB and can run on boards costing around $10, booting in under a second even on a sub-1GHz single-core processor. A single binary runs across RISC-V, ARM, MIPS, x86 and LoongArch architectures.

The project is notable for being largely AI-bootstrapped — most of the core code was generated by an agent and refined through human review. Capabilities include native Model Context Protocol (MCP) support for connecting external tool servers, a vision pipeline that forwards images and files to multimodal LLMs, rule-based model routing that sends simpler queries to lighter models to save cost, multiple messaging channels (Matrix, IRC, Discord, WeChat/WeCom and others), a Web UI, and a system-tray interface. It runs on Linux, Windows and Android. The maintainers note it is in early, rapid development, that recent builds may use 10–20MB of RAM, and that it should not be deployed to production before v1.0.