Self-hosted ยท MIT License ยท Node 22+

OpenClaw ๐Ÿฆž

The AI agent gateway that connects all your chat apps to powerful AI โ€” on your own hardware.

WhatsApp, Telegram, Discord, iMessage, Mattermost โ€” one Gateway, infinite agents.

5+
Chat Channels
1
Gateway Process
5min
Setup Time
MIT
License

Architecture

One Gateway. All Your Chats. Any Agent.

A single process bridges every chat app to your AI agents โ€” no vendor lock-in, no cloud dependency, runs on your machine.

๐Ÿ“ฑ WhatsApp โœˆ๏ธ Telegram ๐ŸŽฎ Discord ๐Ÿ’ฌ iMessage ยทยทยท More ๐Ÿฆž Gateway Session routing ยท Auth ยท Streaming ๐Ÿค– Agent Loop Context โ†’ Model inference โ†’ Tool exec Claude ยท GPT ยท Ollama ยท Custom โš™๏ธ Shell ๐ŸŒ Browser ๐Ÿ” Search ๐Ÿ’พ Files ๐Ÿง  Memory โฐ Cron ๐Ÿ“ฑ Mobile Nodes iOS ยท Android ยท Camera ๐Ÿ”Œ Plugins Skills ยท Hooks ยท Custom
๐Ÿ“ก

Multi-Channel Ingress

All your chat apps connect to a single Gateway. No separate bots, no duplicate configs โ€” one process handles everything.

๐Ÿฆž

Gateway Core

Routes messages to the correct session, handles auth tokens, and streams responses back to the right channel in real time.

๐Ÿค–

Agent Loop

Assembles context from memory and skills, calls your AI model, orchestrates tool execution, and streams the reply.

๐Ÿ”ง

Pluggable Tools

Shell, browser, search, files, memory, cron โ€” attach exactly the tools your agent needs. Extend with custom plugins.

Message In. Magic Out.

The agent loop turns a simple chat message into a full AI-powered response with tool use โ€” in seconds.

1

You send a message on WhatsApp, Telegram, or Discord

Your message hits the Gateway's webhook or polling loop instantly. Any channel, same experience.

Multi-channel
2

Gateway routes to your session

Sessions are isolated per sender. Your conversation history, memory, and workspace are loaded for context continuity.

Sessions & Memory
3

Agent assembles context โ€” system prompt, memory, active skills

Long-term memory, workspace files, and pluggable skill prompts combine into a rich context window.

Context Assembly
4

Model thinks and decides what tools to invoke

Claude, GPT, Ollama โ€” whichever model you configure. Tool calls are parsed from the model output automatically.

Model Inference
5

Tools execute โ€” search web, run code, read files, browse

Tool results feed back into the loop. The agent can chain multiple tools until the task is complete.

Tool Execution
6

Response streams back to your chat app

Streaming tokens appear in real time in your chat. Markdown, code blocks, rich formatting โ€” Gateway handles it all.

Streaming Reply

Capabilities

Everything Your Agent Needs

10 core capabilities built into every OpenClaw installation โ€” no plugins required for the essentials.

๐ŸŒ

Multi-Channel Gateway

One process serves WhatsApp, Telegram, Discord, iMessage, Mattermost simultaneously. Add more channels without running new bots.

๐Ÿ”„

Agent Loop

Full message โ†’ context โ†’ inference โ†’ tool execution โ†’ streaming reply pipeline. Handles multi-step tasks end-to-end.

๐Ÿง 

Sessions & Memory

Per-sender sessions with workspace files and long-term memory. Your agent remembers context across conversations.

๐Ÿ”ง

Tool Use

Shell execution, file operations, web search, browser control, cron scheduling โ€” attach the right tools for any task.

๐Ÿ“ฑ

Mobile Nodes

Pair iOS and Android devices to unlock camera, location, notifications, and hardware sensors as agent tools.

๐Ÿ•ธ๏ธ

Multi-Agent Routing

Route to isolated agents by session, workspace, or sender. Run specialist coding agents, research agents, and more in parallel.

๐ŸŽฏ

Skills System

Pluggable skill modules โ€” GitHub integration, weather, coding assistants, custom APIs. Skills snap in like LEGO bricks.

โฐ

Heartbeats & Cron

Schedule background tasks and proactive automations. Your agent checks in, sends reminders, and runs jobs without prompting.

๐Ÿ”’

Security Controls

Allowlists, auth tokens, command sandboxing, and safety filters. Self-hosted means your data never leaves your infrastructure.

๐Ÿ”Œ

Plugin System

Hook into every part of the pipeline โ€” message ingress, tool execution, response formatting. Extend anything, override anything.

Up in 5 Minutes

Node 22+, an API key, and a couple of commands. That's all it takes to deploy your first AI agent gateway.

1

Install OpenClaw globally via npm. Requires Node 22 or newer.

2

Run the onboard wizard to configure your first channel and AI provider. Interactive, takes ~2 minutes.

3

Start the Gateway and connect your first chat app. OpenClaw prints the webhook URL and QR code.

4

Send a message in WhatsApp, Telegram, or Discord. Your AI agent replies in seconds.

โœ“

You're live. Add more channels, skills, and tools from the config file anytime.

Terminal
# Install globally $ npm install -g openclaw@latest # Run the setup wizard $ openclaw onboard # Start your gateway $ openclaw start ๐Ÿฆž OpenClaw Gateway v1.0.0 Listening on :3000 Sessions: active Tools: ready Channels: WhatsApp โœ“ Telegram โœ“ # Add a channel later $ openclaw channel add discord # Fire a system event $ openclaw system event --text "Hello from the terminal" --mode now

Use Cases

What Will You Build?

From personal assistants to team bots โ€” OpenClaw is the foundation for any AI-powered workflow.

๐Ÿค–

Personal AI Assistant

Your own always-on assistant in WhatsApp or Telegram. Search the web, manage files, set reminders, answer questions โ€” on your hardware.

๐Ÿ‘ฅ

Team Bot

Deploy a shared agent in your Discord or Mattermost. Answer team questions, run CI, fetch metrics, automate standups.

๐Ÿ’ป

Coding Agent

Connect Claude Code or GPT-4 to your chat app. Review PRs, write code, run tests, and commit โ€” all from a chat message.

๐Ÿ 

Home Automation

Pair with mobile nodes for camera, location, and sensors. Control smart home devices and get proactive notifications via chat.

Free. Open. Yours.

OpenClaw is MIT-licensed and community-driven. Self-host with confidence โ€” your data never touches our servers.