Getting Started
OpenClaw installs in one command. The setup has three steps: install the runtime, connect an LLM, connect a messaging channel. Once those are done, you send a message from your phone and your computer responds.
One-line installer on Mac, Windows, or Linux
Add your Claude, OpenAI, or Ollama endpoint
Link WhatsApp, Telegram, Discord, or another app
Message it from your phone - it acts on your computer
Step 1 - Install
The fastest way on macOS, Linux, or Windows (with WSL) is the one-liner:
Alternative install methods:
- npm:
npm install -g openclaw - From source: clone
github.com/openclaw/openclaw, thennpm install && npm start - macOS companion app: a native menu-bar app (beta) available at
openclaw.ai
After install, OpenClaw runs as a background process. Check it's running with openclaw status.
Step 2 - Connect a Model
Edit ~/.openclaw/config.json (created on first run) and add your model:
Claude (Anthropic)
"model": "claude-sonnet-4-6",
"anthropic_api_key": "sk-ant-..."
OpenAI
"model": "gpt-4o",
"openai_api_key": "sk-..."
Local model via Ollama (no API key needed)
"model": "ollama/llama3.3",
"ollama_base_url": "http://localhost:11434"
Step 3 - Connect a Messaging Channel
Run openclaw connect and choose your channel. OpenClaw will walk you through the channel-specific auth flow:
| Channel | How to connect |
|---|---|
| Telegram | Create a bot via @BotFather, paste the token - fastest setup |
| Discord | Create a Discord application, add bot to your server |
| Scan a QR code with WhatsApp on your phone (uses WhatsApp Web protocol) | |
| Slack | Create a Slack App, install to workspace, add bot token |
| iMessage | macOS only - grant Accessibility permissions to OpenClaw |
Start with Telegram - it has the simplest setup and no phone number scanning required.
Step 4 - Your First Task
Once connected, send a message from your channel. Start with something simple to verify the setup is working:
what time is it- verifies the agent is running and respondinglist files in my Desktop folder- verifies file system accessrun echo hello world in shell- verifies shell accessopen google.com in browser- verifies browser (Playwright) access
If a skill is restricted, OpenClaw will tell you what permission to grant in the config. The first run is the hardest - once the four layers are confirmed working, everything else builds on that.