Skip to content
CLAWDBOOK
Popular searches
Private, static site search Open
Getting startedGuides
Browse the OpenClaw handbook

Install OpenClaw

Install OpenClaw on macOS, Linux, WSL2, or Windows, complete onboarding, and verify that the CLI and Gateway are healthy.

Last verified
August 9, 2026
Reviewed against
OpenClaw 2026.7.1-2 documentation
Difficulty
Beginner
Time
~8 minutes
Verified against the official documentation listed below

The recommended installer detects your operating system, provisions a supported Node.js version when needed, installs OpenClaw, and starts onboarding.

Prerequisites

OpenClaw currently supports macOS, Linux, and Windows. Supported Node lines start at 22.22.3, 24.15, or 25.9; the hosted installer currently recommends Node 26. You do not need pnpm unless you are building OpenClaw from source.

Choose an environment you control. An agent can read files, call tools, and send messages with the permissions you give it, so a personal machine or isolated server is a better first setup than a shared production host.

macOS, Linux, or WSL2

Run the official installer:

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

The installer launches onboarding by default. If you need an unattended install, add --no-onboard and run openclaw onboard later.

Windows PowerShell

Open PowerShell and run:

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

Windows users can also choose the native Windows Hub app or a WSL2-based Gateway. Pick one primary Gateway path to avoid debugging two services at once.

Complete onboarding

Onboarding connects a model provider, creates the initial configuration, and can install the background Gateway service:

openclaw onboard --install-daemon

Start with one model and one channel. Confirm that each layer works before adding more providers, plugins, or Skills.

Verify the installation

Run these checks in order:

openclaw --version
openclaw doctor
openclaw gateway status

Expected result

  • openclaw --version prints a version instead of a shell error.
  • openclaw doctor reports no blocking configuration or service errors.
  • Gateway status reports a running runtime and a successful connectivity probe.

If the first command fails, go straight to the command-not-found fix. If the CLI works but the service does not, use Gateway not running.

Next steps

Connect a model before a chat channel, then test a direct turn. Once the model works, add Telegram, Discord, or WhatsApp and confirm the channel with openclaw channels status --probe.

Source notes

Technical facts on this page were checked against primary sources.

Related guides