Can a Mac mini run local AI?

Yes. A Mac mini with Apple silicon runs local models with Ollama or LM Studio. Its memory decides the size: about 3B parameter models on 8 GB, 7B on 16 GB, 13B on 32 GB.

Can a Mac mini be a home AI server?

Yes. LM Studio can serve models to other machines on your network, so a Mac mini can answer for laptops and other apps in the house.

Does Grux OS need a screen on the Mac mini?

No. With `~/.grux/HEADLESS` present it runs with no window and takes no focus, and `~/.grux/SILENT` turns its sound off.

How do I run local AI on a Mac mini?

Run a model server such as Ollama or LM Studio on the Mac mini, size the model to its memory, and either let it serve your other devices or run an agent on it. A Mac mini suits the job because on Apple silicon the memory, not a separate graphics card, decides how large a model it can hold: by Jan's published minimums, 8 GB handles models of about 3B parameters, 16 GB about 7B and 32 GB about 13B. LM Studio can serve models to other machines on your network, and Ollama answers on a local API. Grux OS (MIT) can run on the Mac mini as the agent. Its headless mode shows no window and takes no focus, its silent mode makes no sound, its command line reads with the app closed, and the optional phone companion pairs over your local network as a remote microphone and control surface, encrypted, with nothing leaving your LAN.

Grux OS 3.0.0 · last checked 2026-09-30 · generated from the shipping release

Step by step

Install a model server. Ollama for a runtime other apps share, or LM Studio for a desktop app that can also serve models across your network.
Pick a model that fits the memory. See the table below, or the which LLM can my Mac run page.
Serve other devices, or run an agent on the mini. For an agent, install Grux OS and turn on headless and silent mode if nobody uses the mini's screen.

brew install ollama
ollama run llama3.2:3b
brew install --cask dotcomjack/tap/grux
# no window, no focus, no sound
mkdir -p ~/.grux && touch ~/.grux/HEADLESS ~/.grux/SILENT

How big a model a Mac mini can run

Memory	Model size, roughly
8 GB	About 3B parameters
16 GB	About 7B parameters
32 GB	About 13B parameters

These are the minimums Jan publishes for macOS. Quantized models fit in less, and more memory runs larger ones.

Running Grux OS on a Mac mini

Headless. With ~/.grux/HEADLESS present it shows no window, takes no focus, and keeps its own record of its windows.
Silent. With ~/.grux/SILENT present it makes no sound.
Command line. grux reads answer with the app closed, from files on disk; writes go over a private Unix socket and Grux opens no network port.
Phone companion (Labs, opt in). Pairs over your local network as a remote microphone and control surface; if you never pair a phone, the Mac never opens a listening socket.

Where to check this

Claim	Source
8 GB for 3B models, 16 GB for 7B, 32 GB for 13B on macOS	Jan README, System Requirements
LM Studio serves models on OpenAI-like endpoints, locally and on the network	LM Studio docs
Ollama answers a REST API on localhost	Ollama README
Grux's headless and silent modes, switched by files under ~/.grux	CHANGELOG, 3.0.0
Grux reads work with the app closed; writes go over a Unix socket	Grux README, The command line
The phone companion pairs over the local network, encrypted, and is opt in	Grux README, The phone companion

Questions

Can a Mac mini run local AI?: Yes. A Mac mini with Apple silicon runs local models with Ollama or LM Studio. Its memory decides the size: about 3B parameter models on 8 GB, 7B on 16 GB, 13B on 32 GB.
Can a Mac mini be a home AI server?: Yes. LM Studio can serve models to other machines on your network, so a Mac mini can answer for laptops and other apps in the house.
Does Grux OS need a screen on the Mac mini?: No. With ~/.grux/HEADLESS present it runs with no window and takes no focus, and ~/.grux/SILENT turns its sound off.

Download Grux 3.0.0 Read the source

Free, MIT licensed. macOS 14 or later, Apple silicon. 23.7 MB, signed and notarized by Apple. No account, no server, no subscription.