OpenClaw Quickstart: Your First Self-Hosted AI Assistant in 15 Minutes

Install the gateway, connect a channel, and talk to your own assistant.

Prerequisites

• Node.js (24.15+ recommended; 22.22.3+ or 25.9+ also supported).
• An API key or subscription for an LLM provider (OpenClaw is provider-agnostic).
• A machine you can leave running — OpenClaw installs a background gateway service.
• About 5 minutes to install, and 10 to get your first reply.

Make sure you are installing the right OpenClaw. It is a Node.js CLI on npm ('npm install -g openclaw'). Ignore any guide that says 'pip install openclaw' or 'from openclaw import Agent' — that refers to an unrelated PyPI package, not this project.

Step 1 — Install and Onboard

Install the CLI globally, then run the onboarding wizard. The wizard walks you through configuration and installs the gateway as a background user service.

npm install -g openclaw@latest
# or: pnpm add -g openclaw@latest
 
# Interactive setup; installs the gateway daemon (launchd/systemd)
openclaw onboard --install-daemon

If anything looks off during setup, the built-in diagnostics command is the first place to look:

openclaw doctor

Step 2 — Confirm the Gateway Is Running

The gateway is the control plane for sessions, channels, tools and events. Check its status; for debugging you can also run it in the foreground with verbose logging.

openclaw gateway status
 
# Foreground / debug mode (stop the daemon first):
openclaw gateway stop
openclaw gateway --port 18789 --verbose

Step 3 — Point It at a Model

OpenClaw stores configuration in a JSON file at ~/.openclaw/openclaw.json. At minimum, set the agent's model as a "<provider>/<model-id>" string. Consult the project's configuration reference for the exact provider identifiers and how to supply credentials for your chosen provider.

// ~/.openclaw/openclaw.json
{
  agent: {
    model: "<provider>/<model-id>",
  },
}

Step 4 — Talk to Your Assistant

You do not need a chat app to test it. The agent command sends a one-shot message straight from the terminal. The --thinking flag lets you ask for more deliberate reasoning on harder tasks.

openclaw agent --message "What is the square root of 144?"
 
# A task that uses more reasoning:
openclaw agent --message "Draft a short shipping checklist for a web app" --thinking high

Step 5 — Connect a Messaging Channel

The real value of OpenClaw is talking to your assistant from apps you already use — WhatsApp, Telegram, Slack, Discord, Signal and more. You connect a channel during or after onboarding. By default, OpenClaw uses a pairing policy: a message from an unknown sender is not processed until you approve it.

# When a new sender messages your assistant, they get a pairing code.
# Approve them explicitly:
openclaw pairing approve <channel> <code>

Do not disable the pairing policy for a channel that strangers can reach. An assistant with tool access that processes messages from anyone is a serious security risk. Keep command-running tools sandboxed (OpenClaw supports Docker, SSH and OpenShell backends).

Step 6 — Extend It With Skills

OpenClaw's capabilities are extended through 'skills' — reusable instructions and tools defined as SKILL.md files. Your workspace lives at ~/.openclaw/workspace, with skills under ~/.openclaw/workspace/skills/<skill>/SKILL.md. Community skills are shared through the ClawHub registry.

Handy Commands to Remember

Next Steps

• Read the configuration reference to wire up your specific LLM provider and channel credentials.
• Browse ClawHub for community skills before writing your own.
• Keep tool execution sandboxed and review which senders and channels are paired.
• Pin your OpenClaw version if you need stability — releases ship frequently.