How to install OpenClaw on a Mac mini

What OpenClaw is

OpenClaw is an open source AI assistant that runs on your own computer. You talk to it from chat apps you already use. Its README lists Discord, iMessage, Slack, Teams, Telegram and WhatsApp, plus more than 20 others (github.com/openclaw/openclaw, October 2026).

The core piece is the Gateway. It is a long running process that holds your chat connections and runs the agent. If the Gateway stops, the assistant stops answering. That is why the machine it lives on matters more than the install.

Requirements

• Node 24.16 or newer, or Node 26.1 or newer. The installer adds Node 26 on macOS if Node is missing (docs.openclaw.ai/install, October 2026).
• An AI login. The getting started page says it can reuse an existing Claude Code or Codex CLI login, or take a provider API key.
• Any Apple silicon Mac. The docs do not set a memory minimum. With a cloud model, 16 GB is plenty. A local model is a different story, covered in our local LLM guide.

Install it

Run the official installer in Terminal.

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

It walks you into onboarding when it finishes. To install first and set up later, the docs give a flag that skips onboarding. Then run onboarding yourself with the service option.

curl -fsSL https://openclaw.ai/install.sh | bash -s -- --no-onboard
openclaw onboard --install-daemon

We ran the installer on a fresh GitHub hosted macOS 26 machine on October 2, 2026. Node was already present, so it took about 25 seconds. This is what the version check printed.

$ openclaw --version
OpenClaw 2026.9.7 (c074824)

Then check the install and the service with the documented commands.

openclaw doctor
openclaw gateway status

On our test machine, doctor flagged three things before onboarding. The Gateway mode was unset. Gateway auth had no token. And no command owner was set. Run doctor again after onboarding and clear what is left. Take the owner seriously. Doctor describes it as the account allowed to change config, restart the Gateway and approve commands.

Keep it running 24/7

The service option installs a LaunchAgent. The docs give its label as ai.openclaw.gateway and its file as ai.openclaw.gateway.plist in the LaunchAgents folder of your home Library. launchd starts it at login and restarts it if it crashes. Logs go to one file.

openclaw gateway status --deep
openclaw gateway restart
openclaw health
tail -f ~/Library/Logs/openclaw/gateway.log

You can also install the service on its own, without onboarding. We did that on a second test machine. We set the Gateway mode to local, let doctor generate a token, and ran the install command.

openclaw config set gateway.mode local
openclaw doctor --fix --generate-gateway-token
openclaw gateway install

About 15 seconds later, the status command showed the service up.

$ openclaw gateway status
Service file: ~/Library/LaunchAgents/ai.openclaw.gateway.plist
Gateway: bind=loopback (127.0.0.1), port=18789 (service args)
Dashboard: http://127.0.0.1:18789/
Gateway version: 2026.9.7
Runtime: running (pid 16165, state active)
Connectivity probe: ok
Listening: 127.0.0.1:18789

A LaunchAgent only runs while a user is logged in. After a reboot, nothing starts until someone logs in. On an unattended Mac, turn on automatic login for the account that runs OpenClaw. Then stop the machine from sleeping, since sleep freezes the Gateway and drops every chat connection.

sudo pmset -a sleep 0
sudo pmset -a disksleep 0
sudo pmset -a displaysleep 10

The reasons and the laptop lid rules are in what happens when a Mac sleeps. The full checklist, including power cuts, is in the always-on Mac guide.

Security notes

• On a normal install the Gateway listens on loopback only. Leave it that way. To reach the dashboard from your laptop, the docs use an SSH tunnel.
• Unknown senders in direct messages get a pairing code, not an answer. You approve them with the pairing command below.
• Tools run on the host by default unless you set up sandboxing. The agent can touch whatever that macOS user can touch.
• Secrets live on disk in the OpenClaw state folder, with config in openclaw.json inside it. Treat that folder like a password file.
• Run the built-in audit after every config change. On our fresh test install it reported 0 critical, 1 warning and 1 info item.

ssh -N -L 18789:127.0.0.1:18789 user@gateway-host
openclaw pairing approve <channel> <code>
openclaw security audit

The simplest protection is physical. Run OpenClaw on a Mac that holds nothing else of yours. No personal Apple ID, no password manager, no browser full of logins. If the agent does something odd, the damage stops at that machine.

Running it on a MacRun Mac

Our agent plan is a dedicated Mac mini that never sleeps. Its login session comes back on its own after a reboot, so LaunchAgents start again. It comes with Claude Code and Codex installed. OpenClaw is not preinstalled. You install it yourself with the commands above, because you have full admin rights. If you have signed Claude Code in with your own plan, OpenClaw onboarding can pick that login up. Check your provider's terms for use inside other tools before you rely on it.

Who should not use us for this. MacRun is for developer work under our acceptable use policy. An OpenClaw that writes code, watches CI or triages issues fits. One that runs your personal email and family calendar does not. Also, OpenClaw runs on Linux. If you need nothing from macOS, a small Linux server costs less than any Mac.

Ready to try it on a machine that stays up? Compare the plans on the agent Mac page.