How to install OpenClaw: the three commands, and what comes after
Installing OpenClaw takes minutes with npm. What you need first, what each command does, and the setup decisions that actually matter.
Three commands: install the package with npm, run the onboarding wizard, open the dashboard. You’ll need Node.js (22.22.3+, 24.15+, or 25.9+) and an API key from a model provider. Budget fifteen minutes for the commands and a little longer for the two decisions that matter: which model provider, and which machine this should really live on.
Before you start: two prerequisites
- Node.js. OpenClaw currently requires Node 22.22.3+, 24.15+, or 25.9+ (Node 26 is the recommended default as of mid-2026).
node --versiontells you where you stand. The version requirement has its own guide, including the npm--allow-scriptswrinkle and updating Node without breaking other projects. macOS, Linux, and Windows are all supported. Windows users should know the project itself calls native Windows the trickier path and points to WSL2 as the smoother one. - A model API key. OpenClaw brings the agent; the thinking is a model you connect yourself. What that means and where to get one has its own guide. Create the key and set a spending cap before you start, and onboarding becomes a copy-paste.
The three commands
npm install -g openclaw@latest
openclaw onboard --install-daemon
openclaw dashboardWhat each actually does:
npm install -g openclaw@latestinstalls the OpenClaw CLI globally: the-gflag makesopenclawavailable as a command anywhere, rather than inside one project folder.openclaw onboard --install-daemonruns the setup wizard: model provider and key, initial preferences, and, because of the--install-daemonflag, it registers the Gateway as a background service so your agent starts with the machine and keeps running when you close the terminal.openclaw dashboardopens the Control UI in your browser: the web interface where you meet your agent, connect channels, and manage settings.
From there, the natural next steps are connecting a chat channel (Telegram is the easiest first one) and saying hello.
What the installer just decided for you
Worth knowing on day one, so nothing later is a surprise:
- Everything lives in
~/.openclaw. Config, sessions, and your agent’s workspace and memory: one folder to protect and back up. - The Gateway binds to loopback. Only your own machine can reach it, a deliberately safe default. Keep it that way; remote access has right answers that aren’t “open the port.”
- Unknown chat senders get pairing challenges, not conversations (the default
pairingpolicy). You approve who can talk to your agent.
The decision bigger than the install
The commands above put OpenClaw on whatever machine you’re sitting at, right for a first evening. But an agent is only as available as its computer, and a laptop that sleeps takes the agent with it. Once you know you’re keeping it, the real decision is where it should permanently live: a server you run (the official docs ship Docker and VPS guides), or a managed home like an Everpod pod where the install, the security, and the always-on part arrive done. Verify anything here against the official install docs if you’re reading this long after mid-2026. OpenClaw moves fast.