To install OpenClaw, use the official installer for your operating system, complete onboarding, then verify the Gateway and send a message. You need working model access as well as an installed CLI; those are separate parts of setup.

Independent guide. Installation references checked on October 2, 2026 against the official quickstart and installation overview. This page does not claim a fresh test on every operating system.

Before installation

Check your runtime:

node --version

At this review, the quickstart lists Node.js 24.16+ or 26.1+ and recommends Node 26. Consult the Node setup guide before installing a new runtime; requirements can change between OpenClaw releases. Node 18/20 examples from older tutorials are unsuitable for a current install.

Have a provider API key or supported existing model login available. Onboarding checks the route you choose. A subscription used by another product does not automatically authorize every provider integration, so follow that provider's current instructions.

Install on macOS or Linux

Use the command published by OpenClaw:

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

The installer starts guided onboarding. Choose Quick start to use detected access or Custom setup for more options. If setup was interrupted, run:

openclaw onboard

Keep the installation under your ordinary user account. After completion:

openclaw --version

If the command is unavailable, open a new terminal and check the installer's PATH guidance before reinstalling. On a VPS, continue with our server deployment guide for unattended service operation.

Install on Windows

The current official quickstart supports a native Windows installation. Its PowerShell command is:

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

OpenClaw also documents Windows Hub and a WSL2 Gateway route in the Windows guide. Choose one environment for the installation. If using WSL2, run Linux installation and service commands inside that WSL environment; Windows and WSL can have different home directories, runtimes, and active configs.

A tutorial that says Windows always requires WSL2, or that directs you to install a Python package to obtain the Gateway, does not describe all current official options.

Install a background Gateway service

Quick start can keep the Gateway running in the onboarding terminal. If it is still foregrounded, press Ctrl+C before installing a service:

openclaw gateway install
openclaw gateway status
openclaw gateway health

OpenClaw uses the operating system's service manager. The Gateway service reference describes installation and recovery. Do not launch a second manual Gateway while the service already owns the listener port.

Open the dashboard and send a message

openclaw dashboard

The usual local Control UI address is http://127.0.0.1:18789/. Authenticate using your Gateway credentials and send a short message. A successful reply checks more than the dashboard loading: it also exercises your selected model access.

If you installed on a remote server, use an SSH tunnel or the documented remote-access method. Your laptop's localhost address alone cannot reach a VPS.

Alternative installation methods

Method Use it when Reference
Official platform installer You want the guided operator setup Install overview
Docker Compose You already manage containers Our Docker guide
Repository source checkout You are developing OpenClaw itself Official repository

Do not combine an arbitrary package from PyPI with the official Gateway installation. OpenClaw's operator setup is not a Python SDK scaffold. Likewise, cloning the source repository is unnecessary just to run a personal assistant.

Troubleshoot the first installation

Run these before deleting configuration:

openclaw config validate
openclaw doctor
openclaw logs --follow
Failure Check first
Unsupported Node version Runtime listed by the current official install guide
Command not found User PATH and the terminal environment
Port already in use Existing Gateway service or another listener
Config validation error The named field and installed schema
Dashboard opens but no reply Provider credentials, model access, and quota

For symptom-specific fixes, see installation errors and fixes. Avoid changing model, channel, service, and config settings together; it hides which correction helped.

FAQ

Can I try OpenClaw without a global installation?

The current quickstart documents npx openclaw@latest. Follow that page for its temporary foreground behavior and the transition to a managed service.

Should I run openclaw init or create a project?

For the assistant workflow covered here, use openclaw onboard. The earlier project-scaffolding examples on this site have been replaced with the documented onboarding and Gateway commands.

Is it working when the dashboard opens?

The UI loading confirms one part of setup. Send a real message and check that the provider returns a reply. Then add channels or scheduled work gradually using the local setup guide.