Local AI tools download models, and models are big. Each tool keeps them in a hidden folder, so the space can vanish without you seeing where. Here are the default locations from each tool’s own documentation, and one read-only command to check how big each folder is.
The three places to check
| Tool | Default location on a Mac | Documented in |
|---|---|---|
| Ollama | ~/.ollama/models |
Ollama FAQ |
| LM Studio | ~/.lmstudio/models/ |
LM Studio docs, “Import models” |
| Hugging Face libraries | ~/.cache/huggingface/hub |
huggingface_hub docs, “Understand caching” |
The Ollama FAQ also says you can use a different directory by setting the OLLAMA_MODELS environment variable. The Hugging Face docs say the default cache directory can be changed with the HF_HOME or HF_HUB_CACHE environment variable. LM Studio’s docs mention changing the models directory under My Models in the app. If someone set up your tools, or you changed these yourself, the folder may not be where the table says.
The Ollama FAQ describes ~/.ollama as holding models and configuration, and another Ollama page puts its logs in ~/.ollama/logs.
See how big each one is
du is read-only, and it reports the space a folder uses on disk:
du -sh ~/.ollama/models ~/.lmstudio/models ~/.cache/huggingface/hubIf a path does not exist on your Mac, du says so. That is fine; it means you do not use that tool. I checked du against a Hugging Face cache on a test folder and it reported the real size. None of the three default folders exists on the Mac I used for this article, so the paths come from the documentation and I did not see them filled.
The tools also have their own lists, which give per-model sizes. See the guides for Hugging Face and Ollama.
What is in each folder
- Ollama. Models you pulled or created. Ollama’s CLI reference documents listing and removing them, so remove them through Ollama, not through Finder.
- LM Studio. Model files arranged as publisher, then model, then file, for example
infra-ai/ocelot-v1/ocelot-v1-instruct-q4_0.ggufin LM Studio’s own example. - Hugging Face cache. One folder per repository, such as
models--bert-base-cased. Inside, theblobsfolder holds the files, andsnapshotsholds symlinks to them. Datasets and other repository types share the same cache, so it can hold more than models.
Safe, risky, never
| Item | Verdict | What breaks | Cost to recreate |
|---|---|---|---|
| A model you no longer use, removed with its own tool | Safe | Nothing, unless a project names that model | A re-download, sized like the model |
| The same model stored by two tools | Safe to keep one | The tool you removed it from | Re-download, or re-import, into that tool |
| Hugging Face cache entries | Safe with its own commands | Scripts that load that repository download it again | Re-download size shown by hf cache ls |
| Model files deleted by hand in Finder | Risky | The tool may list a model it can no longer load | Pull again, with a possible broken state in between |
A whole ~/.ollama, ~/.lmstudio or ~/.cache/huggingface folder |
Never, unless you mean to start over | All models and, for Ollama, its configuration too | Every model re-downloaded |
What a re-download costs
Nothing is lost for good if you only downloaded it, but time and bandwidth are real. Before removing a large model, check its size in your tool’s listing so you know what you would pull again. Ollama’s model pages and Hugging Face’s repository pages show sizes. If your connection is slow or metered, keep what you use often.
Keeping models on another drive
Both Ollama and the Hugging Face libraries document an environment variable for this: OLLAMA_MODELS for Ollama, and HF_HOME or HF_HUB_CACHE for Hugging Face. LM Studio’s documentation says you can change its models directory under My Models. I did not test any of these. If you move a folder, set the variable first, make sure the tool is quitted, and check afterwards that models still load, before you remove the old folder.
A calm order
- Run the
ducommand above to find the biggest folder. - Open the tool’s own listing for that folder.
- Remove models you have not used in months, through the tool.
- If two tools hold the same model, see the duplicates guide.
- For uninstalling a tool and what stays behind, see AI tool leftovers.
How DiskJoy fits in
DiskJoy is our product. It is pre-alpha: it scans and shows reclaimable space, so a large model folder would appear in the results. Cleanup is not built yet, so it does not remove any models today. The download page shows the current status of the scan.
Frequently asked questions
Where does Ollama store models on a Mac?
Ollama documentation gives ~/.ollama/models as the macOS location. You can point it somewhere else by setting the OLLAMA_MODELS environment variable.
Where does LM Studio store models?
LM Studio documentation gives ~/.lmstudio/models/ as the default models directory, organised as publisher, model, then the model file.
Where is the Hugging Face cache on a Mac?
The default is ~/.cache/huggingface/hub, according to the huggingface_hub documentation. The HF_HOME or HF_HUB_CACHE environment variables change it.
How big are local AI models?
It varies a lot, from hundreds of megabytes to tens of gigabytes. Check the folder with du, or the listing in the tool itself, instead of guessing.