OpenClaw
★ 381.4k · GitHub →Self-hosted personal AI assistant that lives in your messaging apps.
What is OpenClaw?
OpenClaw is a different animal from the coding agents it shares star charts with: it is a self-hosted personal AI assistant that you talk to through the messaging apps you already use. A local Gateway process — run as a daemon on your own machine — acts as the control plane, bridging an AI agent to WhatsApp, Telegram, Slack, Discord, Signal, iMessage, WeChat, and over a dozen other channels, with a multi-channel inbox and per-channel routing.
Born from Peter Steinberger's viral "Clawdbot/Moltbot" project (the lobster mascot survives the renames), it exploded into one of the fastest-growing open-source projects of the era. It is MIT-licensed and free; you bring your own model — the docs recommend a current flagship model from a provider you trust — plus whatever platform credentials you connect. Voice on mobile and a live Canvas for agent-driven visual output extend it beyond plain chat.
The honest caveat is operational: you are wiring an agent with real capabilities into accounts that can message real people, so security posture matters. OpenClaw ships sane defaults — inbound DMs are treated as untrusted, unknown senders require pairing approval, and non-main sessions can be sandboxed in Docker — but running it well is closer to administering a small server than installing an app. Expect Node 24, config files, and periodic maintenance.
Key capabilities
- ▸ Bridges one assistant to 20+ messaging platforms: WhatsApp, Telegram, Slack, Discord, Signal, iMessage, WeChat, and more
- ▸ Local-first Gateway you run as a daemon — your conversations and credentials stay on your hardware
- ▸ Model-agnostic: bring a flagship model from whichever provider you trust
- ▸ Voice on iOS/Android/macOS plus a live Canvas for agent-driven visual output
- ▸ Security defaults for the messaging context: untrusted-DM handling, sender pairing approval, Docker sandboxing
When to use it
Pick OpenClaw if you want an always-on personal assistant reachable from your phone's chat apps — one that remembers context, runs on your hardware, and isn't tied to any single AI vendor. It rewards people comfortable self-hosting and thinking about security. Don't pick it as a coding agent — Claude Code, Gemini CLI, or Aider are built for that — and skip it if you won't maintain a daemon with access to your messaging accounts; a hosted assistant is safer for set-and-forget users.
Alternatives
Frequently asked questions
What is OpenClaw and how is it different from ChatGPT or Claude apps?
OpenClaw is a self-hosted assistant you run on your own machine and reach through WhatsApp, Telegram, Slack, and other chat apps you already use. Unlike vendor apps, it is MIT-licensed open source, works with any model provider, keeps conversations and credentials on your hardware, and can take actions across channels — at the cost of setting it up and maintaining it yourself.
Is OpenClaw safe to connect to my WhatsApp or Slack?
It ships reasonable defaults: inbound DMs are treated as untrusted input, unknown senders need pairing approval before the agent processes them, and sessions can be sandboxed in Docker. But you are still granting an AI agent access to real messaging accounts, so review permissions, restrict who can trigger it, and keep the gateway patched — treat it like any internet-facing service.
Was OpenClaw previously called Clawdbot or Moltbot?
Yes. The project started as Clawdbot, was briefly renamed Moltbot, and settled on OpenClaw — the lobster mascot ("Molty") carried through all three names. It was created by Peter Steinberger and grew through community contributions into one of the fastest-growing open-source projects, which is why you will find older tutorials under the previous names.
Star count updated 2026-07-02 · refreshed weekly from the GitHub API · browse all tools →