SideMini

Server & LLM box

How to turn a Mac, even an old MacBook, into an always-on LLM server

SideMini4 min read

Apple silicon Macs are quietly excellent local LLM machines: unified memory, low power draw, no fan noise. The hard part is not the model. It is keeping the Mac awake, reachable and able to recover when nobody is sitting in front of it. Here is the setup, from a spare MacBook in a drawer to a Mac mini in a cupboard.

Why a Mac makes a good LLM box

On Apple silicon the GPU shares the Mac’s memory, so a 16 GB machine runs 7–8B models comfortably and a 32 or 64 GB Mac runs much larger quantized models without a graphics card. Idle draw is a few watts, a busy Mac mini stays well under a light bulb, and the whole thing is silent.

macOS also brings the remote-access tools for free: Screen Sharing, Remote Login over SSH, Wake for network access. What it does not do by default is stay awake, come back after a power failure, or show anything useful over Screen Sharing when no monitor is attached. That is the part this guide fixes.

Which Mac: mini, Studio or an old MacBook?

Mac mini is the natural choice: no screen, no battery, tiny, and it is the machine SideMini was built on. Mac Studio is the same idea with more memory for bigger models. An old MacBook is fine too, and free if you already own it: keep it plugged in, put it on a stand, and leave the lid open. With the lid closed, macOS sleeps unless it sees power and an external display, so do not count on lid-closed use for a server.

Intel Macs work but are slow for anything beyond small models. If you are buying, a base M-series Mac mini with as much memory as you can afford beats everything else per watt and per dollar.

Install the model runner

  1. Ollama for a command-line server: install the app, then ollama run llama3.1 in Terminal. It downloads the model and starts serving on port 11434.
  2. LM Studio if you prefer a window: browse models, click download, turn on the local server from the Developer tab.
  3. Let other devices reach it. Ollama listens on localhost only by default; set OLLAMA_HOST=0.0.0.0 (in LM Studio, enable “Serve on local network”). Then test from your laptop or phone with the Mac’s IP address.
  4. Optional: a web front end such as Open WebUI, so the family or the team gets a chat page instead of a terminal.

Test everything with a monitor attached first. Once it works from a second device, unplug the monitor and it should still work; if it does not, the next two sections are why.

Keep it awake, three layers deep

1. Keep awake. In SideMini, Settings → Always On → “Keep this Mac awake: Always”. The Mac stays up as long as the app runs, which is from login onward.

2. Server mode. One click and your password: the Mac never sleeps, does not spin the disk down, starts up again after a power failure and wakes for network access. These are the same switches you would hunt for in System Settings → Energy, flipped together and reversible with one more click.

3. Clean power. A small UPS turns a flaky socket into a non-event. With Server mode on, even a full outage only means the Mac boots again by itself when power returns.

Reach it with no monitor attached

SSH is enough for Ollama, and you should turn on Remote Login regardless. For everything else, Screen Sharing is the tool, and this is where a headless Mac disappoints: with no display, the picture is blank or tiny. SideMini’s virtual display fixes that. Pick 1080p in Settings → Always On and Screen Sharing shows a normal, sharp desktop.

Outside your home network, use Tailscale or a similar private network rather than opening ports on your router. And if you ever want to sit down at the machine, put an iPad next to it: SideMini makes it the screen, and takes it away again when you leave.

Booting without you

Everything above happens after login, so a server needs to log in by itself. That means automatic login, which macOS only allows with FileVault off. SideMini’s iPad Only mode makes that change with your confirmation and can undo it. If the Mac holds sensitive data, keep FileVault and accept a manual password at cold boot; the FileVault post walks through the options.

The always-on checklist

  1. Install Ollama or LM Studio and run a model. Test it from a second device.
  2. SideMini → Always On → Turn On Server Mode. Set Keep awake to Always.
  3. Virtual display: 1080p. Turn on Screen Sharing and Remote Login in macOS.
  4. Decide on FileVault versus automatic login, deliberately.
  5. Unplug the monitor, restart, and confirm you can reach the Mac and the model. Then unplug the power for a minute and confirm it comes back on its own.

Sources

SideMini is the small menu bar app that makes your iPad the screen of a Mac mini and connects it at every boot. Try it free for three days; $9 once, no subscription.

Keep reading