LocalMac
Last updated Oct 2026

Tools for running AI locally on a Mac

You don’t need to code to get started. Here are the four tools most people use, from easiest to most hands-on.

LM Studio

Best for: beginners who want a normal app.

A desktop app with a built-in model browser and chat window. On Macs it can run both GGUF models and Apple’s MLX format through its MLX engine, and it can serve a local API for other apps.

lmstudio.ai

Ollama

Best for: simple one-line model downloads and connecting other apps.

A Mac app plus command-line tool that downloads and runs models with one command and provides a local API many apps support. Ollama announced MLX-powered Apple silicon support in preview in March 2026.

ollama.com

llama.cpp

Best for: tinkerers who want control.

The open-source engine behind the GGUF format, with a command-line chat and llama-server for a local web UI and API. New model support often lands here first.

github.com/ggml-org/llama.cpp

MLX and mlx-lm

Best for: getting the most out of Apple silicon, and fine-tuning.

MLX is Apple’s open-source ML framework; mlx-lm is its Python package for running and fine-tuning language models. Many ready-made MLX models are on Hugging Face under mlx-community.

github.com/ml-explore/mlx · mlx-lm

Which should I use?

If you want…Use
A friendly app, no terminalLM Studio
To plug a local model into other apps or coding toolsOllama or LM Studio’s local server
Maximum control over settings and the newest models firstllama.cpp
Best Mac performance and fine-tuning from Pythonmlx-lm

Your first local model in 10 minutes

  1. Install LM Studio from lmstudio.ai.
  2. Open the model search and pick a model that fits your memory (see the model table). On a 16GB Mac, start with a model under about 10GB.
  3. If offered, choose the MLX version and a 4-bit download.
  4. Load it and start chatting. Everything stays on your Mac.

Prefer the terminal? With mlx-lm installed (pip install mlx-lm), you can run a model straight from Hugging Face:

mlx_lm.generate --model <hugging-face-model-id> --prompt "Explain unified memory in one paragraph"

Commands and features change between versions; check each tool’s docs.