Index / PicoClaw · updated Oct 24, 2026
PicoClaw
sipeed/picoclaw · healthy · rank 10 of 50 by stars
An ultra-lightweight Go rewrite of the personal AI assistant concept, built from scratch by Sipeed to run on $10 hardware with under 10MB of RAM. It targets extreme portability across x86, ARM, MIPS, RISC-V, and LoongArch with millisecond boot times.
Facts
Repository
Runtime
Posture
Community
Security breakdown
Composite 72 / 100 · how these are scored
higher is safer
higher is safer
higher is safer
higher is safer
higher is riskier
Model access
Read from the repository, not written by a model · 91 files examined
6 providers · gateway support
Pinned models
glm-4.7 released 2025-12-22 — 10 months old (from the public model catalogue)
Pin last edited 4 months ago
Evidence
Decision
Why choose PicoClaw over OpenClaw?
Why choose this
- Runs in under 10MB RAM versus OpenClaw's much heavier footprint
- Millisecond boot time on $10 hardware
- Single Go binary deployable across many CPU architectures
Tradeoffs
- Far fewer integrations, skills, and UI features than OpenClaw
- No true OS-level sandboxing for agent tool execution
- Smaller community and thinner documentation depth
Best fit
- Running an AI assistant on cheap SBCs or embedded boards
- Users who want minimal RAM footprint and instant startup
- Self-hosters on RISC-V, MIPS, or other exotic architectures
Avoid if
- You need a rich plugin ecosystem or mature integrations
- You want OS-level sandboxing for tool execution
- You rely on niche chat channels that may lag upstream API changes
Strong evidence from README, license, and active commit history for core claims; sentiment data is thin (no Reddit matches) and mostly from ecosystem digest bots, so community perception is uncertain.
AI layer reviewed Oct 5, 2026 · how this is written
Star activity
30,075 stars today
Overview
PicoClaw is an independent open-source personal AI assistant written entirely in Go by Sipeed, inspired by NanoBot but rebuilt from scratch through a self-bootstrapping process where the AI agent itself drove the architecture migration. Its headline claim is extreme efficiency: it runs on $10 hardware with under 10MB of RAM and millisecond boot times, roughly 99% less memory than OpenClaw, and ships as a single binary supporting x86_64, ARM64, MIPS, RISC-V, and LoongArch.
Architecturally, PicoClaw focuses on a lean core with configurable model provider fallback chains, multi-provider support (Anthropic, Bedrock, OpenAI-compatible), and a tool system with guardrails — for example, recent work steers write_file away from destructive overwrites toward append/edit tools when available. Recent commits show active maintenance including security-driven dependency updates (govulncheck fixes) and provider bug fixes for tool-call history handling.
Compared to OpenClaw, PicoClaw trades feature breadth for footprint: it lacks the full Control UI, extensive skills library, and mature channel integrations, but it is the clear choice for constrained edge devices and users who want an auditable, minimal Go codebase. It is best understood as an edge-first alternative rather than a feature-parity clone.