PicoClaw

PicoClaw is a personal AI assistant written entirely in Go, built to run on very cheap, very constrained hardware. It targets developers, hobbyists, and hardware tinkerers who want to run an LLM-backed agent on a Raspberry Pi Zero, an old Android phone, a RISC-V board, or similar low-power devices without paying for a beefy server. If you've looked at other agent frameworks and found them too heavy for a $10 board, this is the alternative it's designed to replace.

The project positions itself against OpenClaw (TypeScript) and NanoBot (Python), claiming a much smaller memory footprint and faster boot. It ships as a single static binary across x86_64, ARM, ARM64, MIPS, RISC-V, and LoongArch, so deployment is copy-and-run rather than dealing with runtimes or interpreters.

Key features

• Small memory footprint: core process reportedly uses under 10MB RAM (recent builds may run 10-20MB due to active development), which is a large drop from typical agent runtimes that need hundreds of MB or more.
• Fast boot: starts in under a second even on sub-1GHz single-core CPUs, useful for devices that get power-cycled often.
• Single-binary portability: one Go binary per architecture, no dependency install step on the target device.
• MCP support: connects to Model Context Protocol servers to add tools and capabilities to the agent.
• Vision pipeline: send images and files to the agent with automatic base64 encoding for multimodal models.
• Smart model routing: rule-based routing sends simple queries to lightweight models to cut API costs.
• Multiple channels: integrations for Telegram, Matrix, IRC, WeCom, Discord, and WeChat, plus a browser-based Web UI launcher.
• Multi-provider LLM support: works with several LLM providers including AWS Bedrock, Azure, and others listed in its provider list.
• Cross-platform deployment: runs on desktop OSes, Docker, and Android (via APK or Termux).

Ideal use cases

PicoClaw fits scenarios where you want a persistent, chat-driven AI agent running on hardware that costs less than a nice dinner: a $10 RISC-V board acting as a home assistant, a NanoKVM doing automated server operations, an old Android phone repurposed as a smart assistant, or a Raspberry Pi Zero 2 W running unattended in a closet. It also suits developers who want a scriptable agent runtime they can extend with MCP tools and route across multiple LLM providers without a heavyweight backend.

It's a weaker fit if you need a mature, security-audited system for production workloads right now. The README explicitly flags the project as being in early rapid development with possibly unresolved security issues and advises against production deployment before a v1.0 release. If you need guaranteed low memory use today, note that recent builds may temporarily run 10-20MB rather than the advertised under-10MB figure, since resource optimization is planned after feature work settles. Teams needing long-term API stability or formal support contracts should also wait for the project to mature.

Installation

The simplest path is the official site, which auto-detects your platform:

Visit picoclaw.io for a one-click download, or grab a precompiled binary from the GitHub Releases page.

To build from source, you need Go 1.25+ and, for Web UI builds, Node.js 22+ with pnpm 10.33.0+:

git clone https://github.com/sipeed/picoclaw.git
cd picoclaw
make deps

# Install frontend dependencies
(cd web/frontend && pnpm install --frozen-lockfile)

# Build the core binary for the current platform
make build

# Build the Web UI Launcher (required for WebUI mode)
make build-launcher

# Build core binaries for all Makefile-managed platforms
make build-all

# Build for Raspberry Pi Zero 2 W
make build-pi-zero

# Build and install
make install

For the WebUI Launcher (recommended for desktop), run the launcher binary and open http://localhost:18800 in a browser:

picoclaw-launcher
# For remote access, Docker, or VM setups, bind to all interfaces:
picoclaw-launcher -public

A Docker Compose setup is also available: clone the repo, run the launcher profile once to auto-generate docker/data/config.json, edit that file to add API keys, then bring the stack up again in detached mode. macOS users may need to explicitly allow the launcher binary through Gatekeeper on first run since it isn't notarized. Android users can install a prebuilt APK directly or set it up through Termux for a terminal-based launcher.