Use LM Studio

LM Studio reads models from a folder laid out as <publisher>/<repo>/<files>. hangar lmstudio install downloads a model into your local Hugging Face cache and links it into that folder:

hangar lmstudio install bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M
hangar lmstudio install mlx-community/SmolLM-135M-Instruct-4bit

• For a GGUF Repo, it installs one quant group and any mmproj-*.gguf projector; without a quant it picks Q4_K_M, or the closest one available.
• MLX and other Repos are installed whole.
• Files are symbolic links into the cache on macOS and Linux, so the disk is used once even when Python libraries use the same model. Use --link-mode hardlink or --link-mode copy if LM Studio's folder is on another disk, or on Windows (where the default is hardlink).

Restart LM Studio, or open My Models, to see them.

Where the models folder is

In order:

1. --lmstudio-dir, if you give it.
2. The downloadsFolder setting in settings.json in LM Studio's home folder. The home folder is the path in ~/.lmstudio-home-pointer if that file exists, otherwise ~/.cache/lm-studio if it exists, otherwise ~/.lmstudio. This is how LM Studio itself finds its home.
3. models in that home folder.

Keep track

hangar lmstudio list            # what is installed, and whether a newer Revision exists
hangar lmstudio remove bartowski/Llama-3.2-3B-Instruct-GGUF
hangar lmstudio remove bartowski/Llama-3.2-3B-Instruct-GGUF --prune   # also free the cache

The CLI records installs in ~/.config/hangar/lmstudio.json. --prune deletes the model's cached files unless another install uses them; Python code that used the same cached files downloads them again next time.

To add models you already have in LM Studio to the Arsenal, see Importing an existing cache.