How to Install OpenClaw: A Beginner's Guide

You have heard about OpenClaw, want to try it out and are now sitting in front of your terminal. Wondering:

Where do I start?

No surprise. OpenClaw has become the fastest-growing open-source project in history within just a few months (over 280,000 GitHub stars as of March 2026). Everyone is talking about it. But a proper step-by-step guide? That is harder to find than you would think.

I have installed, configured, broken and repaired OpenClaw on multiple machines since the early days (back when it was still called Clawdbot). My MacBook swallowed the onboarding without issues on the first try. My Linux server did not. The installation is not rocket science. But there are a few pitfalls nobody tells you about when you simply copy the one-liner from the documentation.

In this guide, I will show you step by step how to install OpenClaw, set it up and get your first AI agent running. Whether you use macOS, Linux or Windows.

What Is OpenClaw Actually?

A quick overview in case you are not exactly sure what this is about:

OpenClaw is an open-source AI agent with an MIT license. The project was founded in November 2025 by the Austrian Peter Steinberger (who has since moved to OpenAI; the project is now continued as a foundation).

And no:

OpenClaw is not just another coding tool like Claude Code or GitHub Copilot. OpenClaw is a universal AI agent. That means you can use it through WhatsApp, Telegram, Discord, Signal or Slack, not just through the command line.

You ask it via message to do something, and it does it. Research, summaries, data analysis, automations. If you are primarily looking for coding assistance, a comparison of AI coding tools might be more helpful. But OpenClaw can do significantly more.

Essentially, it is a personal AI assistant that you can connect anywhere.

What You Need for the OpenClaw Installation

Check these points quickly before you start:

Node.js 22 or newer. This is the most important prerequisite. If you are unsure which version you have, open your terminal and type node --version. Does it show something below v22? Then you need to update. The easiest way is with nvm (Node Version Manager):

nvm install 22

If you do not have nvm yet, you can find the installation instructions on the official GitHub page.

Your operating system. OpenClaw runs on:

• macOS 13 (Ventura) or newer
• Linux with glibc 2.31+
• Windows only through WSL2 (more on that shortly)

An API key (optional but recommended). You need an API key from an AI provider, e.g. Anthropic (Claude), OpenAI (GPT) or Google (Gemini). Alternatively, you can use Ollama for local models. That is completely free. More on that later.

That is all you need. Let's go!

Installing OpenClaw on macOS and Linux

The fastest method is the official one-liner. Open your terminal and paste the following command:

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

The script downloads OpenClaw, installs it and sets up the gateway service. The gateway then runs as a background process and handles communication with the AI models.

If you (understandably) prefer not to blindly run a script from the internet, there is an alternative via npm:

npm install -g openclaw@latest
openclaw onboard --install-daemon

Both paths lead to the same result. The npm method just gives you more control over what happens.

Once the installation is complete, the setup wizard starts automatically:

openclaw onboard

If it does not start on its own, simply type the command manually. The wizard guides you through the basic configuration. API keys, model selection and so on.

Installing OpenClaw on Windows (WSL2)

This gets a bit trickier. If you primarily work with Windows, you need to take a small detour.

OpenClaw does not run directly on Windows. You need WSL2 (Windows Subsystem for Linux). If you have not set up WSL2 yet, open PowerShell as administrator and type:

wsl --install

After that, restart your computer and open the WSL terminal (Ubuntu by default). From here, the installation is identical to Linux:

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

Setting Up OpenClaw (API Keys and Model Selection)

After installation, you have two things to do. Set up an API key and select a model.

Setting Up an API Key

The openclaw onboard wizard asks you directly. But you can also edit the configuration file afterwards (all options can be found in the official documentation):

~/.openclaw/openclaw.json

OpenClaw now supports quite a few providers:

• Anthropic (Claude Opus 4.5, Sonnet 4.5)
• OpenAI (GPT models)
• Google (Gemini)
• Ollama (local models, completely free)
• OpenRouter, Mistral, DeepSeek, Groq and more

Choosing a Model

My recommendation for getting started:

Start with Claude Sonnet 4.5 if you want quality. This model delivers the best results for most tasks and is reasonably priced (about $3 per million input tokens).

If you do not want to spend money yet, install Ollama and use a local open-source model. Honestly, you notice the quality difference compared to Claude or GPT with more complex tasks. But for trying things out and simpler tasks, it is sufficient.

Connecting Messaging Channels (Optional)

This is exactly what sets OpenClaw apart from other AI agents:

You can connect it directly to WhatsApp, Telegram, Discord, Signal or Slack. You do this either through the setup wizard or through the dashboard:

openclaw dashboard

This command opens an interface in the browser where you can connect and configure channels. It looks a bit overwhelming at first, but I had the Telegram connection set up in under 5 minutes.

Getting Started with OpenClaw

Everything set up? Then try it out.

If you connected a messaging channel, simply send your OpenClaw bot a message. For example through Telegram:

"Summarize the most important AI news from the past week."

OpenClaw then uses the configured model, researches (if you enabled web access) and responds directly in the chat.

You can also use OpenClaw through the command line. Simply start a session:

openclaw chat

Start with something simple. Ask for a summary, have it rewrite a text or request an analysis of a PDF. This way you get a feel for what works and where the limits are.

You will notice the limits quickly. OpenClaw hallucinates just like any other AI tool when you feed it vague requests. The more specific your instructions, the better the results. If you want a broader overview, check out my overview of the best AI tools.

Common Installation Errors (and How to Fix Them)

I have made each of these mistakes at least once myself. Here are the most common problems and what you can do about them:

"openclaw: command not found"

The classic (happened to me too). After installation, your terminal cannot find the command. This is almost always a PATH issue. Close the terminal and reopen it. If that does not help, manually add the installation path to your .bashrc or .zshrc.

Node.js Version Too Old

If you get an error message related to Node.js, check your version with node --version. Anything below v22 is too old:

nvm install 22

Build Tools Missing (Linux)

On Linux, you may need additional packages. If the build fails, install:

sudo apt install build-essential git python3

Gateway Port Is in Use

If the gateway service does not start, something else is probably already running on the same port. You can change the port in openclaw.json or check with openclaw gateway status what is going on.

Configuration Is Broken

Did you mess up openclaw.json so badly that nothing works anymore? Been there. The solution:

openclaw doctor --fix

This command analyzes your configuration and fixes what can be fixed.

WSL2 Is Slow

As mentioned. Work in native WSL directories, not in /mnt/c/.... It makes an enormous difference.

Security Notes You Should Know

Quick reality check.

OpenClaw is a strong project with an active community. But it is also still young. And that brings security risks.

In February 2026, a critical security vulnerability was discovered (CVE-2026-25253) that allowed attackers to execute malicious code with a single click. The vulnerability was patched, but it shows why you should always keep OpenClaw up to date.

openclaw update --channel stable

Additionally, a study found that about 36% of skills on ClawHub (the marketplace for OpenClaw extensions) contain prompt injections. Cisco, Microsoft and Kaspersky have also published security warnings. According to security researchers, over 135,000 OpenClaw instances are accessible unprotected on the internet.

My security rules for OpenClaw:

1. Always use the latest version (sounds basic, but CVE-2026-25253 showed why this matters)
2. Do not give sensitive data (passwords, customer data, internal documents) to the agent
3. Check skills from ClawHub before installing, do not install blindly
4. For automated agents, set a spending limit with your API provider

What Does OpenClaw Cost?

The software itself costs nothing. OpenClaw is completely open source under the MIT license.

The actual costs come from the AI models you use in the background. Those are in your control.

Typical costs with normal usage run $5 to $30 per month for API calls (depending on model and intensity). That is comparable to a ChatGPT Plus subscription.

If you want to go completely free, install Ollama and use local models. Then you truly pay nothing (apart from the electricity for your computer).

Conclusion

That is it! You have installed OpenClaw, configured it and know what to watch out for.

The security warnings might sound intimidating, but they are not a reason to ignore OpenClaw. Update regularly, do not feed it sensitive data and do not blindly install every extension from ClawHub. Then you are on the safe side.

If you want to go deeper, check out skill development next. Building custom skills was a real aha moment for me, because that is when you realize how flexible the system is. If you are not sure whether OpenClaw is the right tool for you, my comparison of the best OpenClaw alternatives can help with the decision.