LM Studio and Grux OS solve different halves of local AI on a Mac. LM Studio is for finding, downloading and running models: it supports Apple's MLX on Apple silicon, runs on Mac, Windows and Linux, serves an OpenAI compatible API on your machine and can use MCP servers. It is free for home and work, but the app is closed source; only its lms command line tool is MIT. Grux OS is a free, open source (MIT) agent for macOS. It uses a local model to act on the mail, calendar, files and meetings already on your Mac from one Command Panel, with a shell whose every command can be undone. They work together: Grux defaults to Ollama, and it also treats an OpenAI compatible server on your Mac, such as LM Studio's, as a local model.
| Grux OS | LM Studio | |
|---|---|---|
| What it is | An agent that acts on your Mac | A model browser, runner and server |
| Source | Open source, MIT | Closed source app; the lms CLI is MIT |
| Price | $0 | Free for home and work |
| Platforms | macOS on Apple silicon | Mac, Windows and Linux |
| Models | Ollama by default, or a local OpenAI compatible server, or your own key | GGUF and MLX models it downloads and runs |
| Acts on your mail, calendar, files and shell | Yes | No, it is not built for that |
LM Studio's details are from its documentation and its free for work announcement, checked on 2 October 2026.
Run a model in LM Studio and start its local server, then point Grux's local model address at it in Settings. Grux's own tests treat a server on this Mac, LM Studio's included, as a local model, so a timeout reads as a busy model rather than a network problem. Or skip LM Studio and let Grux find Ollama by itself, which is the default.