OpenClaw Setup Guide: Configure Your AI Agent

Setup Overview

OpenClaw setup involves five key areas. You can configure them in any order, but this sequence gets you to a working agent fastest:

1. AI Model — Choose which LLM powers your agent
2. Chat Platform — Connect Telegram, Discord, WhatsApp, or Web
3. Gateway — Set up authentication and access control
4. Skills — Add capabilities like web search, coding, and automation
5. Memory — Configure how your agent remembers conversations

Step 1: Choose Your AI Model

The model is the brain of your agent. OpenClaw supports 100+ models from major providers. Set it in your openclaw.json:

{
  "models": {
    "providers": {
      "openrouter": {
        "apiKey": "sk-or-..."
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "openrouter/anthropic/claude-sonnet-4.6"
      }
    }
  }
}

Popular choices:

See our models page for the full list, or read individual model guides: Claude, OpenAI, Ollama, DeepSeek.

Step 2: Connect a Chat Platform

Your agent needs a way to talk to users. OpenClaw supports multiple platforms:

You can connect multiple platforms at once. Each platform is configured in the channels section of openclaw.json.

Step 3: Configure the Gateway

The gateway provides web-based access to your agent and handles authentication:

{
  "gateway": {
    "enabled": true,
    "auth": {
      "token": "your-secret-token"
    },
    "controlUi": {
      "allowInsecureAuth": true
    }
  }
}

The token protects your gateway from unauthorized access. For Telegram and Discord, also configure dmPolicy to control who can message your bot. Use "pairing" for Telegram (required — "open" allows anyone on the internet to use your bot). For Discord, "open" is safe since bots are invite-only, but you must also set allowFrom: ["*"] alongside it.

See the Gateway Pairing Guide if you encounter connection issues.

Step 4: Add Skills

Skills extend what your agent can do. Install them from ClawHub:

openclaw skills install @official/web-search
openclaw skills install @official/code-runner

Popular starter skills:

• Web Search — Let your agent search the internet
• Code Runner — Execute code in a sandbox
• File Manager — Read and write files
• Browser Automation — Control a headless browser

Browse all available skills on ClawHub or read the Skills Installation Guide.

Step 5: Configure Memory

Memory lets your agent remember past conversations and user preferences:

{
  "session": {
    "scope": "user"
  }
}

Key memory options:

• session.scope: "user" — Each user gets their own conversation history
• session.scope: "global" — Shared conversation across all users

For advanced memory with embeddings and long-term recall, see the Memory Guide. For the experimental dreaming memory consolidation system, see the Dreaming Guide.

Verify Your Setup

After configuration, verify everything works:

1. Start OpenClaw: openclaw start (or docker compose up)
2. Open the gateway URL in your browser
3. Send a test message to confirm your model responds
4. Test each connected platform (send a message on Telegram, Discord, etc.)
5. Try a skill command to verify skills are loaded

Common Setup Issues

What's Next?

• Agent Setup Guide — Advanced agent configuration and multi-agent setups
• Best Skills — Top-rated skills to supercharge your agent
• MCP Setup — Connect external tools via the Model Context Protocol
• Documentation — Full OpenClaw documentation reference