Index / Picobot · updated Sep 3, 2026
Picobot
louisho5/picobot · healthy · rank 25 of 43 by stars
An ultra-lightweight OpenClaw-inspired AI agent that ships as a single ~9MB Go binary with persistent memory, tool calling, and Telegram/Discord/Slack integration. Built to run on a $5 VPS, Raspberry Pi, or even an old Android phone via Termux.
Facts
Repository
Runtime
Posture
Community
Security breakdown
Composite 62 / 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 · 27 files examined
1 provider · gateway support
Pinned models
gemini-2.5-flash released 2025-06-17 — 15 months old (from the public model catalogue)
Pin last edited 6 months ago
Evidence
Decision
Why choose Picobot over OpenClaw?
Why choose this
- ~9MB single binary vs heavyweight runtime stacks
- ~10MB idle RAM and near-instant cold start
- Runs on extremely constrained hardware (Pi, Termux, $5 VPS)
Tradeoffs
- Much smaller ecosystem of skills and integrations
- Younger project with limited production track record
- Fewer guardrails and sandboxing features documented
Best fit
- Self-hosters wanting a minimal agent on cheap VPS or Raspberry Pi
- Users who want Telegram/Discord/Slack bots without Python/Node bloat
- Tinkerers who value a single-binary, zero-dependency deployment
Avoid if
- You need a mature plugin/skill ecosystem or large community
- You require strong sandboxing for untrusted tool execution
- You want a battle-tested project with a long release history
Evidence is solid on architecture and footprint from the README and releases, but community sentiment data is thin and partly off-topic (several Reddit matches are unrelated). Security and multi-user capabilities are inferred, not documented.
AI layer reviewed Aug 24, 2026 · how this is written
Star activity
1,344 stars today
Overview
Picobot is a Go-based, single-binary AI agent explicitly positioned as a lightweight alternative to OpenClaw. It bundles persistent memory, tool calling, skills, and channel integrations (Telegram, Discord, Slack) into a ~9MB binary with ~10MB idle RAM usage, targeting deployments on $5 VPS instances, Raspberry Pis, or Android phones via Termux. Configuration, memory, and skills persist in a simple host-mounted data directory, and setup is a single docker run or go build away.
Recent development is active: the v0.2.0 release added MCP server integration, a tool activity indicator, and hardened JSON error handling, suggesting the project is maturing its tool-calling surface. Compared to OpenClaw, Picobot trades ecosystem breadth and guardrails for radical simplicity — no Python, no Node, no heavy containers — making it one of the most resource-efficient entries in the OpenClaw-alternative landscape.