Quiet and frugal
Apple silicon idles at a few watts and is nearly silent, so leaving it on all day, every day is cheap and unobtrusive.
Your AI agents, always on.
A practical, step-by-step guide to turning a Mac mini into a quiet 24/7 server for AI assistants and coding agents — one that stays awake, comes back by itself after a power cut, and that you can reach from anywhere.
Read the transcript
0:01 Turn a Mac mini into an always-on personal AI server: a home for your AI assistants and coding agents, running around the clock.
0:08 Start with the hardware. Any Apple silicon Mac mini works. Get plenty of memory and storage, and use wired Ethernet.
0:14 Finish the initial macOS setup, install every update, and create a dedicated admin account.
0:20 Next, keep it awake. The setup-power script uses pmset to disable sleep and restart automatically after a power failure. It's a dry run by default, so you see every change first.
0:31 For unattended reboots, know the tradeoff: automatic login brings your agents back on its own, but it requires turning FileVault off.
0:38 For remote access, add Tailscale for a private network, and turn on Screen Sharing and Remote Login.
0:44 Login items and LaunchAgents start your background jobs and keep them alive. A template is included.
0:50 Sync files with iCloud, Google Drive, or OneDrive, so agents work on the same documents you do.
0:56 Then install your tools: Homebrew, Node, and Python, plus Cursor, Claude, and a Cursor private worker.
1:03 Harden it: firewall on, strong passwords, automatic security updates, and only the sharing you need.
1:09 Back it up with Time Machine, and test a restore.
1:12 After every reboot, run the health-check script to confirm power, network, sync, and agents are all healthy.
1:18 Always on, recoverable, and secure. Grab the full guide and scripts to set up yours.
Why a Mac mini?
Section titled “Why a Mac mini?”Runs the tools you already use
Cursor, Claude, Homebrew, Node, Python, Xcode, iOS simulators, a real browser — everything your agents need, on the same OS you work on.
Always reachable
With Tailscale, Screen Sharing and SSH you can check on long-running agents from your laptop or phone — without opening a single port on your router.
Self-healing
Power settings, LaunchAgents and a health check mean jobs restart on their own and you know within a minute of a reboot whether everything came back.
The ten steps
Section titled “The ten steps”- 1Choose your hardwareMemory, storage, Ethernet and a small UPS.
- 2Initial macOS setupAccounts, name, updates and sensible defaults.
- 3Stay awake & auto-recoverpmset, power-failure restart, FileVault vs auto-login.
- 4Remote accessTailscale, SSH and Screen Sharing — no open ports.
- 5Login items & LaunchAgentsKeep apps and headless jobs running.
- 6Cloud storage synciCloud, Google Drive and OneDrive without surprises.
- 7Install AI toolsHomebrew, Node, Python, Cursor, a Cursor worker, Claude.
- 8Security hardeningFirewall, keys, secrets and agent permissions.
- 9BackupsTime Machine plus an off-site copy you have tested.
- 10TroubleshootingWhat to check when something didn’t come back.
Scripts that do the fiddly parts
Section titled “Scripts that do the fiddly parts”Every script is dry-run by default, safe to run twice, and prints exactly what it would change before
you add --apply.
setup-power.sh
Never sleep, restart after a power failure or freeze, optional daily power-on. Read more
health-check.sh
After a reboot: what’s running, what isn’t, and why. PASS / WARN / FAIL. Read more
install-launchagent.sh
Turn any command — like a Cursor worker — into a job launchd keeps alive. Read more