How to Set Up OpenClaw: Your Own AI Assistant That Actually Does Things

Most AI chatbots live in a browser tab. You type, they respond, and that is about it. OpenClaw is different. It runs on your machine, connects to your messaging apps, and can actually interact with your system — running commands, managing files, and automating repetitive tasks.

With over 149,000 stars on GitHub, OpenClaw (formerly known as Clawdbot) has become one of the most popular open-source AI agent projects out there. And the best part? You can set it up in about 20 minutes.

This guide walks you through the entire process — from installation to connecting your first messaging channel.

What Makes OpenClaw Worth Your Time

Before we get into the setup, here is why OpenClaw stands out from the crowd of AI tools:

• It runs locally on your machine, so your conversations stay private
• It connects to WhatsApp, Telegram, Discord, Slack, and 15+ other messaging platforms
• It supports multiple AI models — Claude, GPT-4, DeepSeek, or free local models via Ollama
• It has a skill system with over 10,000 community-built plugins
• It can actually do things on your computer — not just chat

Think of it as having a personal AI assistant that lives in your favorite messaging app instead of a separate website.

What You Need Before Starting

The requirements are pretty minimal:

• Node.js 22 or higher — grab it from nodejs.org if you do not have it
• 2GB RAM minimum (4GB recommended, 8GB+ if you want to run local models)
• An API key from Anthropic (Claude), OpenAI (GPT-4), or you can use Ollama for free local models
• About 500MB of disk space

Quick check — open your terminal and run:

node --version

If you see v22.x.x or higher, you are good to go.

Step 1: Install OpenClaw

You have got three options here. Pick whichever fits your setup.

Option A: npm (works everywhere)

This is the simplest approach and works on Windows, macOS, and Linux:

npm install -g openclaw@latest
openclaw onboard

The onboard command launches an interactive wizard that walks you through everything — model selection, API key, and your first channel connection.

Option B: Docker (most secure)

If you prefer isolation, Docker keeps OpenClaw sandboxed from the rest of your system:

docker run -d \
  --name openclaw \
  -v ~/.openclaw:/root/.openclaw \
  -p 18789:18789 \
  ghcr.io/steipete/openclaw:latest

Option C: One-line install script

On macOS or Linux, you can also use the install script:

curl -fsSL https://openclaw.ai/install.sh | bash

This handles Node.js installation automatically if you do not have it.

Step 2: Start the Gateway

The Gateway is the core process that routes messages between your channels, the AI model, and your system. It needs to stay running in a terminal window.

openclaw gateway start

Verify it is running:

openclaw gateway status

You should see something like Gateway running on port 18789. If it fails, make sure nothing else is using that port.

Step 3: Connect an AI Model

OpenClaw does not include its own AI — it is an orchestration layer that connects to the model you choose. Here are your options:

Provider	Cost	Best For
Anthropic (Claude)	~$20-40/month	Best reasoning, recommended for most users
OpenAI (GPT-4)	~$25-50/month	Broad ecosystem, familiar API
DeepSeek	~$5-15/month	Budget-friendly with solid performance
Ollama (local)	Free	Maximum privacy, needs 8GB+ RAM

Your config lives at ~/.openclaw/openclaw.json. Here is what a Claude setup looks like:

{
  "ai": {
    "provider": "anthropic",
    "apiKey": "sk-ant-your-key-here",
    "model": "claude-sonnet-4-6"
  }
}

If you want to try it for free first, Ollama lets you run models locally:

ollama pull llama3.1:8b

Then set your config to:

{
  "ai": {
    "provider": "ollama",
    "model": "llama3.1:8b",
    "endpoint": "http://localhost:11434"
  }
}

Step 4: Connect a Messaging Channel

This is where OpenClaw gets interesting. Instead of chatting in yet another app, you talk to your AI through the messaging platforms you already use.

Telegram (easiest to start with)

openclaw channel add telegram

You will need to create a bot via @BotFather on Telegram (takes about 2 minutes), then paste the token when prompted. After that, just message your bot and it routes straight to your local OpenClaw agent.

Discord

openclaw channel add discord

WhatsApp

openclaw channel add whatsapp

This uses WhatsApp's linked devices feature — scan a QR code and you are connected. One heads up though: WhatsApp's terms do not officially allow automated bots on personal accounts, so consider using a secondary number for testing.

Step 5: Install Skills

Skills are plugins that give OpenClaw specific abilities. There are over 10,000 available on ClawHub. Browse and install them like this:

openclaw skills browse
openclaw skills install google-calendar github-assistant daily-briefing

Some skills worth checking out:

• google-calendar — schedule and manage events through chat
• github-assistant — handle PRs, issues, and code review
• daily-briefing — morning summary with calendar, weather, and news
• file-manager — organize, rename, and convert files
• web-scraper — extract data from websites

Step 6: Personalize Your Agent

OpenClaw uses two special files to define your agent's behavior:

SOUL.md — defines the personality and communication style. Think of it as the agent's character sheet.

AGENTS.md — defines specific capabilities and rules. This is where you tell the agent what it can and cannot do.

Create these in your OpenClaw directory (~/.openclaw/) and the agent will follow them in every conversation.

Things to Watch Out For

A few gotchas that trip up newcomers:

• Daily reset at 4 AM: By default, OpenClaw clears its short-term memory every night. If you need conversations to persist, you will need to configure permanent memory in the settings.
• Gateway must stay running: If you close the terminal window running the gateway, your agent goes offline. Consider using a process manager like PM2 or running it as a system service.
• API costs add up: If you are using Claude or GPT-4, keep an eye on your usage. Set spending limits in your provider's dashboard. Ollama is free but requires more RAM.

Is It Worth It?

If you are comfortable with a terminal and want an AI assistant that goes beyond just answering questions, OpenClaw is one of the most capable options available right now. The self-hosted nature means your data stays on your machine, and the skill ecosystem means you can extend it to do almost anything.

The setup is straightforward — most people get it running in under 20 minutes. And if you hit a wall, the community Discord is active and helpful.

For those who do not want to deal with self-hosting, managed alternatives like ClawTrust handle the infrastructure for you, though you trade some control for convenience.

Either way, OpenClaw represents a shift in how we interact with AI — from passive chatbots to active agents that can actually get things done on your behalf.