Skip to content
openclawOS
v2026.x · MIT licensed · Star us on GitHub →

Your AI agent. Every chat app. Your hardware.

openclawOS is a self-hosted gateway that connects Claude, GPT and Gemini to WhatsApp, Telegram, Discord, iMessage, Signal, Slack and 30+ more channels — running on your own machine. No cloud middleman. No data egress. Just one Gateway, every messenger.

Runs on macOS · Linux · Windows (WSL2) · iOS · Android

  • 30+

    Channels supported

  • 534k

    Lines of TypeScript

  • MIT

    Licensed, forever

  • 100+

    Contributors

Self-hosted

Runs on your Mac, your Linux box, your home server. Sessions, memory and credentials never leave your network.

Multi-channel

Send a message in Telegram. Get the reply on WhatsApp. Same identity, same context — Gateway routes it all.

Agent-native

Pi ships in the box: tool use, vector memory, vision, multi-turn sessions, branching, compaction.

Open source

MIT licensed. Fork it, audit it, run a private build. We push releases weekly.

Get started

From zero to messaging in 60 seconds.

Pick your installer. Run one command. Pair your favourite messenger. Talk to Pi.

npm install -g openclawos@latest
openclawos onboard --install-daemon

Recommended for macOS & Linux. Requires Node 22+.

1. Install

Pick npm, curl, Docker or Homebrew.

2. Pair a channel

openclaw apps install telegram

3. Say hi

Message your bot. Pi responds with tools, memory, context.

Architecture

The Gateway is the kernel. Channels are processes.

openclawOS borrows the OS metaphor: one daemon coordinates routing, sessions and memory; channels and agents are isolated apps you compose. Crash one, the rest keep humming.

Channels

WhatsApp app
Telegram app
Discord app
Slack app
iMessage app
Signal app
Gateway (kernel)

Routes messages, tracks sender identity, isolates sessions per user/group, persists context.

  • Sessions & memory
  • Bindings & routing
  • Auth & secrets
  • Hooks & cron
  • Vector store (sqlite-vec)
core

Agents & providers

Pi (default)
Custom agent
Claude (Anthropic)
GPT (OpenAI)
Gemini (Google)
Any OpenAI-compatible
← inbound channels routing & sessions LLM providers →

Bring your own LLM

  • Claude (Anthropic)

    Recommended: Opus 4.6+

  • GPT (OpenAI)

    GPT-4, Codex, Responses

  • Gemini (Google)

    Vision + long context

  • GLM

    Self-hosted OK

  • MiniMax

    Long context

  • Any OpenAI-compatible

    Ollama, vLLM, Together…

Welcome to the lobster tank.

Star the repo, join Discord, ship your first agent before lunch.