An OpenClaw setup for Dad

Hi Dad. This is how to build your own AI assistant that lives on the Mac mini and answers you from your phone, anywhere. It's the open-source cousin of the setup I run in Chicago: always on, text it like a person, and you fully own it. Three pieces — OpenClaw (the assistant), Google Gemini (its brain, free to start), and Tailscale (a private tunnel so your devices can reach the mini safely). Take it one step at a time, and call me whenever.

(Not my dad? Somebody else's dad, or just setting up your first harness? Welcome — this works the same for you.)

It runs in its own account — the safe part

We'll run everything under a separate macOS user account on the mini, not your everyday one. macOS keeps accounts walled off from each other, so the assistant has its own home folder and cannot see your personal account — your documents, photos, Mail, Messages, and passwords all live somewhere it has no key to. If anything ever misbehaves it's contained, and you can delete that account later without touching a thing of your own. Do everything on the mini while logged into that account.

We set up Tailscale first so your MacBook can reach the mini over SSH — then you do the rest from your laptop instead of hunched over the mini.

1. Set up Tailscale (at the Mac mini)

Tailscale builds a private network — a tailnet — between your own devices. Think of it as a private group chat, but for devices: once they're signed in, your phone, laptop, and the mini can talk directly and securely from anywhere, as if they were on the same home WiFi. Nothing is exposed to the public internet.

On the mini, install Tailscale from tailscale.com and sign in. Give it an easy name when asked (like dads-mac-mini).
Install Tailscale on your phone and MacBook too, signed into the same account.
Turn on Tailscale SSH so your MacBook can log in without passwords or keys:
```
tailscale up --ssh
```
(If tailscale isn't found, call me — it's a one-time toggle.)

2. Reach the mini from your MacBook

SSH ("Secure Shell") lets you open a Terminal on the mini from another computer — over your tailnet, so it's private and works from anywhere. Let's make it one word. On the MacBook, run open -e ~/.zshrc and paste these two aliases at the bottom (already filled in with this mini's account and name — swap magerbot@Mac if yours differs):

alias macmini='ssh magerbot@Mac'
alias macminidash='ssh -N -L 18789:localhost:18789 magerbot@Mac'

Save, then run source ~/.zshrc (or open a new Terminal window). Now:

macmini drops you into a Terminal on the mini. This is where you'll run the rest of the setup. (exit leaves.)
macminidash opens a secure tunnel to OpenClaw's control panel: while it's running, open http://localhost:18789 in your browser to see the dashboard — chat, settings, logs. (Ctrl-C closes it.) The dashboard is admin-only, so we reach it privately through the tunnel rather than putting it on the internet.

3. Install OpenClaw (on the mini)

Type macmini to hop onto the mini and run OpenClaw's installer:

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

That one command installs everything it needs and launches the guided setup. When it asks for a provider, choose Gemini — and when it asks how to connect, pick Sign in with Google. That's the easy path: no API keys to create or copy, just your normal Google login in the browser. (If asked for a model, go with google/gemini-3.1-pro-preview.)

If onboarding doesn't offer the Google sign-in, do it afterward with this — it opens a browser to log in:

openclaw models auth login --provider google-gemini-cli --set-default

The installer also sets up a background service so the assistant keeps running on its own, and lets you connect a channel (like Telegram) so you can text it.

When it's done, confirm it's alive:

openclaw gateway status
openclaw models list --provider google

Then try the macminidash tunnel and text your assistant. That's it — it's running.

4. Keep the mini running

An always-on assistant needs an always-on computer:

Don't let it sleep. System Settings → Lock Screen / Energy → never sleep.
Keep the account logged in. Turn on Fast User Switching (System Settings → Control Center), log into the assistant's account once, then switch back to your own — it keeps running in the background.
Survive reboots. To have it come back on its own after a power blip, see how I keep mine alive across restarts: Keeping an always-on agent alive across reboots.

Cheat sheet

Every command, by where you type it.

On your MacBook:

macmini          # log into the mini over SSH
macminidash      # open the tunnel, then visit http://localhost:18789
exit             # leave the mini
open -e ~/.zshrc # edit your aliases
source ~/.zshrc  # reload aliases

On the mini — setup:

tailscale up --ssh                          # turn on Tailscale SSH
curl -fsSL https://openclaw.ai/install.sh | bash   # install OpenClaw (+ onboarding)
openclaw models auth login --provider google-gemini-cli --set-default   # sign into Gemini with Google

On the mini — day to day:

openclaw gateway status                  # is it running?
openclaw gateway restart                 # restart after a change
openclaw logs --follow                   # watch live logs (Ctrl-C to stop)
openclaw dashboard                       # get a one-click link to the panel
openclaw models list --provider google   # confirm Gemini is connected

If anything looks off or a command doesn't work, take a screenshot and call me.