OpenClaw Agent Tutorial 2026: Zero to Running Agent

You keep hearing about OpenClaw. People on X are posting screenshots of their agents booking meetings, writing newsletters, and managing entire businesses. You want in. This is the tutorial that gets you from zero to a working agent in under 30 minutes. No fluff. Just the steps.

What OpenClaw Actually Is

OpenClaw is an open-source personal AI assistant that runs on your own hardware. Your laptop, a Mac Mini, a $15/month VPS. It connects to AI models like Claude or GPT, then plugs into your real tools: Telegram, WhatsApp, Discord, Gmail, your calendar, your codebase.

The difference from ChatGPT or other chatbots: OpenClaw runs 24/7 as a background service. It has persistent memory across conversations. It can execute code, browse the web, manage files, and take actions on your behalf. It's not a chat window you open and close. It's an always-on employee.

The project is fully open source on GitHub with active development and a growing community. If you want to understand what OpenClaw can do before installing, check out our complete breakdown of what OpenClaw is.

Prerequisites (What You Need Before Starting)

Before you touch the terminal, make sure you have these ready:

• A computer that stays on. Mac, Linux, or Windows with WSL2. A Mac Mini is ideal. A VPS works too. Your laptop works for testing but your agent sleeps when you close the lid.
• Node.js v20 or higher. Check with node --version. If you need to install it, grab it from nodejs.org or use nvm.
• An AI API key. You need at least one: Anthropic (Claude), OpenAI (GPT), or OpenRouter (access to multiple models). Anthropic's Claude is the most popular choice. Keys cost based on usage, typically $5-20/month for personal use.

For a detailed walkthrough on getting your Anthropic API key configured, read our API key setup guide.

Step 1: Install OpenClaw

Open your terminal. The fastest way to install is with npm:

npm install -g openclaw

That's it. One command. If you prefer building from source (useful if you want to contribute or modify):

git clone https://github.com/openclaw/openclaw.git
cd openclaw
pnpm install && pnpm ui:build && pnpm build
pnpm link --global

Verify it worked:

openclaw --version

You should see a version number. If you get "command not found," make sure your npm global bin directory is in your PATH.

Step 2: Run the Onboarding Wizard

This is where the magic starts. Run:

openclaw onboard

The wizard walks you through everything:

1. AI Model Provider. Pick Anthropic, OpenAI, or OpenRouter. Paste your API key when prompted.
2. Default model. For Anthropic, Claude Sonnet is a solid default. Claude Opus for heavier reasoning tasks.
3. Gateway daemon. Say yes to installing the daemon. This is what makes OpenClaw run 24/7 as a background service instead of dying when you close the terminal.
4. Workspace directory. The wizard creates ~/.openclaw/workspace. This is your agent's home. All its files, memory, and configuration live here.

After onboarding, start the gateway:

openclaw gateway start

Check it's running:

openclaw gateway status

You should see the gateway running on its default port. Your agent is alive.

Step 3: Connect a Messaging Channel

An agent without a way to talk to you is useless. OpenClaw supports multiple channels:

• Telegram (most popular, easiest setup)
• WhatsApp (via WhatsApp Web bridge)
• Discord (great for team use)
• Web UI (built-in, accessible via browser)

For Telegram, the fastest option:

1. Open Telegram and message @BotFather
2. Send /newbot, give it a name and username
3. Copy the API token BotFather gives you
4. Add it to your OpenClaw config: openclaw config set telegram.token YOUR_TOKEN
5. Restart: openclaw gateway restart

Now message your bot on Telegram. It responds. That's your agent.

We have dedicated setup guides for each channel: Telegram setup, WhatsApp setup, and Discord bot setup.

Step 4: Set Up Memory

This is what separates OpenClaw from every chatbot you've used before. Memory means your agent remembers things between conversations. Not just within a session. Across days, weeks, months.

OpenClaw uses markdown files for memory. Go to your workspace:

cd ~/.openclaw/workspace

You'll see files like:

• AGENTS.md -- instructions and rules for your agent
• SOUL.md -- personality and tone
• USER.md -- information about you
• MEMORY.md -- long-term curated memory
• memory/ folder -- daily notes and logs

Start by editing USER.md. Tell your agent who you are: your name, timezone, what you do, what you care about. The more context you give, the more useful it becomes.

Then edit SOUL.md to shape its personality. Want it direct and concise? Say so. Want it to crack jokes? Tell it. This file is its identity.

For a deep dive into how the memory system works, read our complete guide to OpenClaw's memory architecture.

Step 5: Add Skills

Skills are what make your agent actually useful beyond conversation. A skill is a package of instructions and scripts that teach your agent how to do something specific: manage your calendar, post on social media, analyze data, generate images.

Browse available skills:

openclaw skills list

Install one:

openclaw skills install weather

Now your agent can check weather forecasts. Ask it "What's the weather in Bali?" and it actually knows how to look it up.

Some skills I use daily:

• GitHub -- manage issues, PRs, code review
• Weather -- quick forecasts without opening an app
• Summarize -- extract text from URLs, transcribe YouTube videos
• Apple Notes / Reminders -- sync with your Apple ecosystem

You can also build custom skills. Drop a SKILL.md file into your workspace's skills folder with instructions, and your agent picks it up automatically.

Check out the OpenClaw Lab community for skill templates and workflows shared by other users.

Step 6: Automate with Cron Jobs

This is where OpenClaw goes from "cool assistant" to "employee that never sleeps." Cron jobs let you schedule tasks that run automatically.

Examples of what you can automate:

• Morning briefing at 8 AM (weather, calendar, emails)
• Check your inbox every 2 hours and summarize unread messages
• Post to social media at optimal times
• Monitor your website for downtime
• Generate weekly reports every Friday

Set up a cron through your agent's config or just ask your agent to create one. "Set up a cron job that checks my email every morning at 8 AM and sends me a summary on Telegram." It handles the rest.

For the full breakdown on cron job setup and patterns, read our cron jobs automation guide.

Step 7: Run Multiple Agents

One agent is useful. Multiple agents working together is a system. OpenClaw supports multi-agent setups where each agent has its own personality, skills, and responsibilities.

You could have:

• A research agent that monitors trends and competitors
• A content agent that drafts posts and newsletters
• An ops agent that manages your calendar and inbox
• A coding agent that handles PRs and bug fixes

Each agent gets its own SOP file, its own memory, and its own cron schedule. They can hand off work to each other. The dispatcher pattern (one agent routes tasks to specialists) works especially well.

I run 13 agents on a single Mac Mini. Total cost: the hardware ($600 one-time) plus API usage ($100-200/month depending on activity). That replaces what would cost $3,000-5,000/month in virtual assistants.

For a complete walkthrough on running OpenClaw on dedicated hardware, check our Mac Mini AI server guide.

Common Mistakes (and How to Avoid Them)

After helping hundreds of people set up OpenClaw in the OpenClaw Lab community, these are the mistakes I see over and over:

Mistake	Fix
Not filling out USER.md	Your agent can't help you if it doesn't know who you are. Spend 5 minutes on this.
Using the cheapest model for everything	Free models produce free-quality output. Claude Sonnet is the sweet spot for most tasks.
Running too many crons too frequently	Start with 2-3 daily crons. Add more once you understand your API costs.
Not setting up the daemon	Without the gateway daemon, your agent dies when you close the terminal. Always run openclaw gateway start.
Ignoring security basics	Your agent has access to your tools. Set proper permissions. Don't expose the gateway port publicly without auth.
Giving vague instructions	"Help me with marketing" gets you nothing. "Draft 3 tweets about our new podcast episode with @guestname" gets results.

If you run into issues, our troubleshooting guide covers the most common problems and fixes.