How to Set Up OpenClaw (Clawdbot): The Complete Beginners Guide (2026)

I’ve set up OpenClaw (Clawdbot) from scratch multiple times now. The first time took me most of a day because I made every mistake possible. Now I can get a fully working setup running in about 2 hours. This guide is everything I wish someone had told me before I started.

By the end, you’ll have a working OpenClaw instance with messaging connected, a custom personality, and your first scheduled tasks running. No fluff, just the actual steps that work in 2026.

Not sure if OpenClaw is right for you? Read my honest review after 6 months of daily use first. If you want to know what it’ll cost, here’s the full pricing breakdown.

Before you start: what you actually need

I see people overcomplicate this. Here’s what you need and nothing else.

Hardware (pick one):

• Mac Mini M4 (recommended for 24/7 use, starts at $499) – I wrote a full Mac Mini setup guide if you go this route
• Any Mac, Linux machine, or Windows PC with WSL2
• A Raspberry Pi 5 works for light use but struggles with heavier workloads

Software requirements:

• Node.js 22 or newer
• A terminal you’re comfortable with (Terminal on Mac, any Linux terminal, Windows Terminal with WSL2)
• A text editor (VS Code, Nano, or even TextEdit)

Accounts you’ll need:

• Anthropic account for Claude API access (console.anthropic.com)
• Telegram account (if using Telegram as your messaging channel)

Budget: The software is free. Expect $40-150/month for Claude API costs depending on usage. I break this down in detail in my cost breakdown post.

Step 1: Install Node.js

If you already have Node.js 22+, skip this. Check with:

node --version

If it says v22 or higher, you’re good. If not, here’s the install for each platform:

Mac:

brew install node

Don’t have Homebrew? Install it first from brew.sh. It’s a one-line command and you’ll use it for other things too.

Linux (Ubuntu/Debian):

curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash -
sudo apt-get install -y nodejs

Windows: Install WSL2 first (Microsoft has a good guide for this), then follow the Linux instructions inside WSL. Native Windows works too, but WSL gives you a much smoother experience with OpenClaw.

Common issue: If you have an older Node.js version installed through a different method, you might get version conflicts. The cleanest solution is to use nvm (Node Version Manager) to manage multiple versions. Install nvm, then run nvm install 22 and nvm use 22.

Step 2: Install OpenClaw

One command:

npm install -g clawdbot@latest

This installs Clawdbot globally. Verify it worked:

clawdbot --version

If you get a “permission denied” error on Linux/Mac, either use sudo npm install -g clawdbot@latest or (better) fix your npm permissions. The npm docs explain how to set up a global install directory in your home folder so you never need sudo for npm again.

Step 3: Run the onboarding wizard

This is where it gets real:

clawdbot onboard --install-daemon

The wizard walks you through:

1. Setting up your workspace directory
2. Configuring your Anthropic API key (or OAuth login)
3. Choosing your default AI model
4. Basic channel setup
5. Installing the background daemon so OpenClaw runs automatically

Follow each prompt carefully. The wizard handles most of the tricky configuration for you. If you mess something up, you can always re-run clawdbot onboard to redo any step.

Pro tip: When the wizard asks for your Anthropic API key, make sure you’re copying it from console.anthropic.com and not from a Claude Pro subscription. They’re different things. The API key starts with “sk-ant-” and you need credits loaded on your Anthropic account.

Step 4: Choose your AI model

OpenClaw works with Claude models from Anthropic. Your main choices:

• Claude Sonnet – Best for daily tasks, fast responses, moderate cost. This is what I use 90% of the time.
• Claude Opus – Best for complex reasoning and long tasks. Higher cost but noticeably smarter for difficult problems.
• Claude Haiku – Fastest and cheapest. Good for simple tasks where you don’t need heavy thinking.

My recommendation: start with Sonnet as your default. It handles the vast majority of tasks well and keeps costs reasonable. Switch to Opus for complex projects when you need the extra horsepower. I have a detailed comparison in my OpenClaw vs ChatGPT breakdown that covers model selection.

You can change models anytime in your gateway config or per-session. There’s no lock-in.

Step 5: Connect a messaging channel

This is what turns OpenClaw from a terminal tool into a real assistant you can talk to from your phone. I recommend starting with Telegram because it’s the fastest to set up and has the best bot API.

Telegram setup (10 minutes):

1. Open Telegram and search for @BotFather
2. Send /newbot and follow the prompts to create a bot
3. Copy the API token BotFather gives you
4. Add the token to your OpenClaw config with clawdbot gateway config.patch

The config section looks like this:

channels:
  telegram:
    token: "YOUR_BOT_TOKEN"
    allowedUsers:
      - YOUR_TELEGRAM_USER_ID

Then restart the gateway:

clawdbot gateway restart

Open Telegram, find your bot, and send it a message. If it responds, you’re live.

Finding your Telegram user ID: Message @userinfobot on Telegram. It replies instantly with your numeric user ID.

I wrote a more detailed walkthrough for all three messaging platforms (Telegram, Discord, and WhatsApp) in my channel connection guide.

Step 6: Write your AGENTS.md file

This is the most important step and the one most people rush through. Your AGENTS.md file tells OpenClaw who it is, who you are, and how to behave. A generic AGENTS.md gives you a generic bot. A detailed one gives you something that actually feels like your assistant.

Create the file in your OpenClaw workspace:

nano ~/clawd/AGENTS.md

Here’s a starter template:

# [Bot Name] - [Your Name]'s AI Assistant

## Who I Work For
[Your name] - [brief description of what you do]

## Core Rules
- Be concise and direct
- Ask before taking external actions (sending emails, posting content)
- Save important information to workspace files
- Be honest when you don't know something

## About [Your Name]
- Timezone: [your timezone]
- Work: [what you do]
- Communication style: [how you like to be talked to]

## How We Work Together
- Primary communication: [Telegram/Discord/etc.]
- I'm here to [your main use case]

The more detail you add over time, the better your bot gets. My AGENTS.md is over 100 lines and covers my businesses, my team, my preferences, and specific rules for how Jarvis handles different situations.

I wrote a full guide with 10 ready-to-use AGENTS.md templates if you want to go deeper. There’s also a complete configuration guide that covers every option.

Step 7: Install your first skills

Skills extend what OpenClaw can do. Think of them as plugins. Start with a few essentials:

clawdbot skill install weather
clawdbot skill install summarize

Check what’s available:

clawdbot skill list

My top recommendations for beginners:

• weather – Get forecasts without opening an app
• summarize – Summarize URLs, PDFs, YouTube videos
• remind-me – Natural language reminders (“remind me to call the accountant at 3pm”)
• timer – Set timers and alarms

Don’t install everything at once. Each skill adds context that your bot has to manage. Start with 2-3, learn them well, then expand. I cover the 15 best skills to install in a separate post.

Step 8: Set up your first cron job

Cron jobs are scheduled tasks that run automatically. This is where OpenClaw starts feeling less like a chatbot and more like a real assistant that works for you even when you’re not actively talking to it.

Here’s a simple first cron job: a daily morning briefing. Your bot sends you a good morning message with weather and your priorities every day at 7 AM. Set this up through the gateway config or by asking your bot to create a cron job.

The simplest way: just message your bot “Create a cron job that sends me a morning briefing at 7am every day with the weather and my top priorities.” It’ll set it up for you.

For more advanced automation ideas, check out my post on automating your entire morning routine with cron jobs.

Step 9: Test everything

Before you consider setup “done,” test these five things:

1. Send a message on Telegram – Does it respond within a few seconds?
2. Ask it something about you – Does it reference your AGENTS.md correctly?
3. Ask for the weather – Does the skill work?
4. Wait for your cron job to fire – Does the scheduled task run on time?
5. Ask it to remember something – Check if memory persists after a gateway restart

If any of these fail, check these common fixes first:

• Restart the gateway (clawdbot gateway restart)
• Check your API key is valid at console.anthropic.com
• Verify your Telegram bot token and user ID match
• Check the logs (clawdbot gateway logs) for specific error messages

Troubleshooting the most common setup problems

After helping dozens of people get OpenClaw running, these are the issues I see over and over.

“Bot doesn’t respond to Telegram messages” – This is almost always one of three things: wrong bot token (copy it again from BotFather, make sure there are no extra spaces), wrong user ID in allowedUsers (double-check with @userinfobot), or the gateway isn’t running (run clawdbot gateway status).

“API key not working” – Make sure you’re using an API key from console.anthropic.com, not your Claude Pro login. Also check that you have credits loaded. A new Anthropic account starts with $0 balance and you need to add funds before API calls work.

“Gateway keeps crashing” – Check your Node.js version. OpenClaw needs v22+. Also check available disk space and RAM. If you’re on a Raspberry Pi, this might be a resource issue.

“Skills not found after installing” – Restart the gateway after installing new skills. The gateway loads skills at startup, so new ones won’t be available until the next restart.

“Memory doesn’t persist” – Make sure your workspace directory has write permissions. Also check that your MEMORY.md and memory/ folder exist in the workspace root. The bot creates these on first use, but permission issues can prevent it.

Common mistakes to avoid

Don’t skip the AGENTS.md. Seriously. It’s tempting to just install and go, but without a good personality file your bot is just a worse version of ChatGPT. Spend 30 minutes on this. Future you will be grateful.

Don’t install 20 skills on day one. Each skill adds complexity and context. Start with 2-3, learn them well, then expand as you find things you actually need.

Don’t ignore memory structure. Create a MEMORY.md file early and start building your bot’s knowledge base. This is what makes OpenClaw compound over time instead of staying static. After a few weeks of accumulated context, the difference is night and day.

Don’t forget security. Always set allowedUsers in your channel config so only you can message your bot. An open bot is a security risk and a financial risk (anyone could rack up your API charges).

Don’t pick the most expensive model for everything. Use Sonnet for daily tasks and only switch to Opus when you need serious reasoning power. Your wallet will thank you.

Pro tips from 6 months of daily use

Run it on dedicated hardware. I tried running OpenClaw on my daily-use MacBook first. It worked, but every time I closed the lid or restarted for updates, my assistant went offline. Missing a morning briefing because my laptop was asleep is a terrible experience. A dedicated Mac Mini running 24/7 changed everything.

Create a USER.md file too. Beyond AGENTS.md, create a USER.md with detailed info about yourself. Birthday, timezone, family members, business details, food preferences, whatever matters. The more context your bot has, the more useful and personal the interactions become. My bot knows my coffee order.

Back up your workspace regularly. Your AGENTS.md, MEMORY.md, and memory folder become irreplaceable after months of use. Set up a simple backup – even just a weekly rsync to a backup drive. I’ve seen people lose months of accumulated context because they didn’t have backups.

Start a content pipeline early. If you create any kind of content, OpenClaw is incredibly good at research and drafting. I have an automated content pipeline that handles research through publishing with minimal input from me.

FAQ

How long does the full setup take?

If you follow this guide step by step, about 2-3 hours for a solid working setup. The AGENTS.md refinement is ongoing – you’ll keep improving it for weeks as you learn what works.

Can I use OpenClaw without a Mac Mini?

Yes. It runs on any Mac, Linux machine, or Windows PC with WSL2. A Mac Mini is just the most popular choice for 24/7 operation because it’s small, quiet, and energy-efficient.

What’s the difference between Clawdbot and OpenClaw?

Same software, different names. The project started as Clawdbot, was briefly called Moltbot, and is now officially OpenClaw. Most of the community still says Clawdbot.

Do I need to know how to code?

Not to set up and use it. Basic terminal comfort helps – navigating folders, editing text files, running commands. You don’t need to write any code unless you want to build custom skills.

Can I run OpenClaw on a VPS or cloud server?

Yes. Any Linux VPS with 2GB+ RAM and Node.js 22 works. DigitalOcean, Linode, Hetzner all work well. You lose macOS-specific features but gain the ability to run it without owning hardware. Monthly cost for a suitable VPS is $10-30.

What if I mess up a step?

Almost everything is reversible. You can re-run the onboarding wizard, regenerate bot tokens, edit config files, and restart the gateway as many times as you need. The only thing you can’t easily undo is deleting your AGENTS.md or memory files, so don’t do that.