Install OpenClaw: Windows, Mac, and Linux tutorial

To install OpenClaw, run the official script for your operating system, complete the setup wizard, then check the Gateway and get an initial response from the model. On Windows, you can use the native PowerShell installation or WSL2; on Mac and Linux, the terminal setup starts with the same command. The commands and requirements below were checked against the official documentation on October 24, 2026. To know whether a self-installed agent suits you against a hosted offer, read open source or proprietary AI agent.

1. Prepare your computer and model access

You will need a terminal, an internet connection, and working access to your chosen model provider. An API key is an authentication secret: never paste it into a public conversation or include it in a screenshot you plan to share with support.

For Node.js, the dedicated documentation requires Node 24.22.1 or later in the 24 release line, or Node 26.12.1 or later in the 26 release line. It recommends the 26 release line. These requirements, checked on October 24, 2026, are stricter than those still shown on some general documentation pages. Follow the dedicated Node.js documentation.

If Node is already installed, check its version:

node -v

The official script can install a runtime if one is missing. If your computer already has several Node installations, note the version displayed before you begin. This can help explain any difference between the environment in your terminal and the one used by the service.

Choose a working folder without sensitive documents for your initial tests. Your first goal is to confirm that the whole setup works, from your request to the model’s response, before connecting your accounts. Our OpenClaw reference guide covers models, skills, and messaging integrations in more detail.

2. Choose the installation path for your system

The table summarizes the terminal installation paths covered here. Follow only the one that matches your environment.

Criterion	Native Windows	Windows with WSL2	Mac	Linux
Terminal	PowerShell	PowerShell, then a Linux terminal	Terminal	Terminal
OpenClaw installation	PowerShell script	Linux script inside WSL2	Shell script	Shell script
Managed service	Scheduled task, with a fallback	systemd user service	LaunchAgent	systemd user service
What to watch	Permission to create tasks	WSL environment and startup	User session	Availability of systemd user services

The official installation page describes these mechanisms and the desktop applications. On Windows, it recommends Windows Hub for a graphical setup. This tutorial covers the commands so you can see what to check at each stage.

Start with a single installation path. Installing a desktop application, a native CLI, and another CLI inside WSL2 makes troubleshooting harder. You may end up checking one environment while the Gateway is running in another.

Keep a brief record of your operating system, the terminal you used, and your installation method. Those details will be useful when updating the software or asking for help. To see where this tool fits among other types of agents, browse our AI agent directory.

3. Install OpenClaw on Mac or Linux

Open your usual terminal and copy the official command:

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

This downloads a script and runs it on your computer. Check the domain before executing it, and use your regular user account. Read any additional permission prompts instead of automatically granting administrator access throughout the process.

Let the installer finish. If it reports an error, save the full message and note which step failed before running anything again. A download interruption and a permission error require different fixes.

The installer normally launches the setup wizard. If it is already open, continue directly to the model connection step. Avoid starting another wizard in a separate window while the first is waiting for your input.

After installation, check that the command is available:

openclaw --version

Seeing a version number confirms only that this step worked. It does not prove that the Gateway is running or that your model provider accepts your credentials.

On a managed work computer, follow the security policies already in place. If a policy blocks the download or execution, identify exactly what is being blocked before looking for another installation method.

4. Install OpenClaw on Windows, natively or with WSL2

Native PowerShell installation

Open PowerShell and run:

iwr -useb https://openclaw.ai/install.ps1 | iex

This official command downloads and runs the installer. Follow the prompts, then continue through the wizard. Do not weaken system protections across the board to work around an error whose cause you have not identified.

WSL2 installation

The official Windows documentation, checked on October 24, 2026, describes WSL2 as the Gateway runtime with the best compatibility with the Linux environment. If you choose this path, start in PowerShell:

wsl --install

Complete the requested steps, then open your Linux distribution. The documentation provides this command to enable systemd:

sudo tee /etc/wsl.conf >/dev/null <<'EOF'
[boot]
systemd=true
EOF

Warning: this replaces the contents of the target file. If you have already customized that file, back it up and merge the setting into your existing configuration.

Return to PowerShell to shut down WSL, after saving your work in any open distributions:

wsl --shutdown

Then reopen the Linux terminal and install OpenClaw:

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

For the rest of the WSL2 setup, run OpenClaw commands in that Linux terminal. Keep using this environment throughout the process rather than switching between it and a native Windows installation.

5. Configure and verify the model connection

If the wizard did not start, launch it:

openclaw onboard

According to the CLI wizard documentation, Quick start detects available access options, asks you to choose a connection, and verifies it by generating an actual response. Custom setup walks you through more settings.

Check the selected provider and access method before continuing. A detected connection may not be the one you intend to use, especially on a shared computer or a machine with several accounts.

The documentation distinguishes between two results: Local setup confirms that local preparation is complete, while Inference ready means the selected connection passed its check. If you skipped the model connection, the installation is not yet fully operational.

For your first launch, choose a simple configuration and leave optional integrations for later. Record the provider and model you selected, but keep secrets out of your setup notes.

Also make sure you can monitor spending with your provider. Installing the software and paying for model generations are separate matters. The LLM pricing calculator can help you estimate a budget from the rates and your expected usage. Avoid running a lengthy automation until you have checked the cost of a limited test.

6. Install the service and complete your first exchange

In the documented Quick start flow, the Gateway initially runs in the foreground in your terminal. To run it in the background, stop that process with Ctrl+C, then run:

openclaw gateway install

Next, check the service:

openclaw gateway status

Then open the dashboard:

openclaw dashboard

These commands follow the official getting started guide. Check that the status reports a running service and a successful connection, then send a short request from the interface.

For an initial test, ask it to rephrase a paragraph containing no personal information. This lets you check the model’s response without relying on a messaging service, an automated browser, or access to your files.

Next, try opening the dashboard again from another terminal. The aim is to confirm that you are using the installed service, rather than a process still attached to the original terminal window.

Before testing an action on files, prepare a test folder with disposable copies and give the agent a narrowly defined task. The distinction between answering and taking action is explained in AI agent vs. chatbot: what’s the difference?.

7. Review permissions before adding integrations

Before connecting accounts or installing extensions, run:

openclaw security audit

The official security documentation describes this audit as a configuration check. It also explains that the Gateway is a trust boundary: it is not designed to isolate mutually hostile users who share the same environment.

Read the results before changing any settings. For each capability you plan to enable, identify which resources it can access and which actions it allows. Reading a folder, modifying a file, and sending a message have different consequences.

Apply the principle of least privilege: grant the access needed for your current task. Keep the Gateway accessible locally for this initial setup. Do not open a port on your router to solve an interface problem.

Then add integrations individually. After each addition, send another simple request and review the permissions you granted. This makes it easier to trace a failure to the latest change, instead of troubleshooting several connections at once.

An audit with no warnings does not guarantee that an action proposed by the model is correct. The guide to securing an AI agent expands on these checks with guidance on managing secrets, isolation, and human approval.

Common problems

The OpenClaw command cannot be found

Close and reopen your terminal, then check that you are in the environment where you installed the software. With WSL2, use the Linux terminal. A PATH issue can prevent the terminal from finding a program that is already installed. Check this before adding more installations.

The dashboard opens, but no response arrives

Check the interface, the Gateway, and model access separately. Run the documented diagnostics:

openclaw status
openclaw gateway status
openclaw doctor
openclaw logs --follow

The official troubleshooting procedure uses these checks to investigate authentication errors, stopped services, and connection problems. Read the specific error before replacing a key or changing network settings.

The service will not start

An “address already in use” message may mean that another Gateway is already using the port. Check whether the process you launched in the foreground is still running. Changing the port arbitrarily can hide a duplicate process without resolving it.

Behavior changes after an update

Keep a verified backup before making a major change. The update documentation distinguishes between CLI installations and packages managed by an application. Use the update mechanism that matches your installation.

When asking for help, include your operating system, installation method, and the error message, after removing secrets and private data. Our article on AI agent security risks explains the precautions to take with logs and access permissions.

Final checklist

The installation is working when the program, the service, and the model connection all work together. Before trusting it with data you need, check the following:

You have identified your installation path and environment.
The runtime meets the current requirements in the dedicated documentation.
The OpenClaw command is available in the correct terminal.
The model connection has passed an actual generation test.
The Gateway is running, and the dashboard receives a response.
You have reviewed permissions and the audit results.
You can monitor spending with your provider.
Your initial tests use nonsensitive data.

If you are now deciding whether this is the right software for your needs, read the OpenClaw vs. Hermes Agent comparison. Keep this installation simple until you have confirmed that it can handle your first practical use case.

Frequently asked questions

Can I install OpenClaw without knowing how to code?

You can follow the installation process without writing a program. You still need to be comfortable opening a terminal, recognizing an error, and understanding the permissions requested before connecting files or accounts.

Does my computer need to stay on to use OpenClaw?

The Gateway must be running to process your requests. If you host it on your computer, shutting down the machine or putting it to sleep can interrupt the service. Remote hosting requires a different setup.

Does an open OpenClaw window mean the installation works?

No. The interface may be accessible even when the connection to your model provider fails. Also check that a simple request receives a response and that the diagnostics report no errors that prevent operation.

Does installing OpenClaw on my PC keep all my data local?

Where the software is installed does not determine where processing happens. If you choose a remote provider, review what information is sent to it and its terms before submitting confidential documents.

Install OpenClaw step by step (Windows, Mac, Linux)