OpenClaw

OpenClaw is a self-hosted, single-user AI gateway that bridges your messaging apps — WhatsApp, Telegram, Discord, Signal, iMessage, Slack, Matrix, Microsoft Teams, and 20+ more — to coding-capable AI agents running on your own hardware. You install one Gateway process (locally on macOS, Linux, Windows-WSL2, or in Docker), pair your devices and channels, and from that moment every DM you send on any connected platform routes to your private AI assistant. No hosted service sees your data; your workspace, your sessions, and your memory files live on your machine. OpenClaw is built for developers and power users who want agentic AI — full tool use, file access, sandbox execution, voice, and multi-agent orchestration — delivered through the messaging surfaces they already live in.

Showcase

Real-world skills and integrations built by the OpenClaw community, collected on the showcase gallery.

PR review skill running in Telegram

A coding-agent skill surfacing GitHub PR review comments directly inside a Telegram group.

Agents UI dashboard

The gateway's web Control UI showing active agent sessions, routing state, and connected nodes.

Oura health tracker integration

An Oura ring health-data skill queried directly through chat — sleep scores, readiness, and activity pulled on demand.

Codex monitor skill

A Codex monitor skill showing live coding-agent run status and approval prompts inside the chat UI.

Live examples

• Get started — install the Gateway in minutes
• Showcase gallery of real-world skills and integrations
• Step-by-step onboarding wizard (openclaw onboard)
• Channels index — every supported messaging platform
• Skills catalogue — bundled and community skills
• Gateway architecture and configuration reference
• Multi-agent routing concepts
• FAQ and troubleshooting

Features

• 25+ messaging channels — WhatsApp, Telegram, Discord, Signal, Slack, iMessage (via BlueBubbles), Matrix, Microsoft Teams, Google Chat, Feishu/Lark, LINE, Mattermost, Nextcloud Talk, Nostr, QQ Bot, Synology Chat, Tlon (Urbit), Twitch, WebChat, WeChat, Zalo, Zalo Personal, IRC, and more via plugins
• Multi-agent routing — route channels, groups, or individual senders to isolated agents each with their own workspace, session, model, and tool set
• Agent Sandbox — per-agent or per-session Docker containers provide a hard execution boundary for untrusted or multi-tenant tool runs
• ACP conversation bindings — pin a chat thread to a persistent ACP (Agent Client Protocol) workspace for durable coding-agent sessions
• Skills system — inject on-demand SKILL.md guides (apple-notes, gh-issues, 1password, camsnap, canvas, coding-agent, gifgrep, and hundreds more) into the agent's context without bloating the base prompt
• Voice support — STT/TTS pipeline on macOS, iOS, Android, and Discord voice channels; speech-to-text transcription and voice-note delivery
• Canvas UI — agent-controlled rendering surface pushed to mobile nodes for rich visual output
• Mobile nodes (iOS + Android) — pair native apps to the Gateway over WebSocket for camera, voice, and Canvas workflows
• Web Control UI — browser dashboard for chat, config management, session inspection, and device pairing
• Streaming previews — live-edit draft replies in Telegram, Discord, Matrix, and Feishu as the model generates tokens
• Persistent memory — per-agent MEMORY.md files with active memory surfacing, Dreaming (background consolidation), and QMD semantic search
• Exec approvals — interactive button-based approval flow for shell commands across Telegram, Discord, and Slack
• Session isolation — per-agent session keys, history, workspace, and context buffer kept fully separate even in broadcast-group setups
• Containerized Gateway — Docker and Podman support with Docker Compose, health-check endpoints, and VPS deployment guides
• Profile rotation and model failover — automatic provider failover across OpenAI, Anthropic, Gemini, and OpenRouter
• DM pairing — explicit owner-approval step for every new sender; allowlist, open, or disabled policies per channel
• Broadcast groups — fan out one inbound message to multiple agents running in parallel with isolated sessions
• CLI toolkit — openclaw onboard, openclaw gateway, openclaw message send, openclaw agent, openclaw doctor, openclaw pairing approve

Documentation

• Getting started — install and first run
• Onboarding wizard — guided channel and agent setup
• Channels — per-channel setup guides
• Skills — writing and installing skills
• Gateway configuration reference
• Agent runtimes and sandboxing
• Multi-agent routing
• Memory and dreaming

Quick start

npm install -g openclaw@latest
openclaw onboard --install-daemon
openclaw gateway --port 18789 --verbose
openclaw message send --channel telegram --target @you --message "Hello from OpenClaw"
openclaw agent --message "Review the open PRs" --thinking high
openclaw doctor