Running Local AI on a Small Mac: Where to Keep Ollama, Hugging Face and LM Studio Models
Local AI models are 4–40 GB each and fill a 256 GB Mac fast. Here's where Ollama, Hugging Face and LM Studio store models, and how to move them to an external drive.
Clearbay TeamGuides from the people building Clearbay
Running language models locally on Apple silicon is genuinely practical now. The catch is storage. A small 7–8B model is around 4–5 GB, mid-size models are 15–40 GB, and it's easy to collect a dozen while you experiment. On a 256 GB or 512 GB Mac, that's a big share of your disk.
Here's where the popular tools keep their models, and how to move them somewhere roomier.
Where models live
Tool | Default location |
|---|---|
Ollama |
|
Hugging Face (transformers, hub) |
|
LM Studio |
|
Because these folders start with a dot, Finder hides them. Press ⌘⇧. in Finder to show hidden files, or check sizes in Terminal with du -sh ~/.ollama/models.
Step 1: Remove models you don't use
Ollama: list with
ollama list, remove withollama rm <model>.Hugging Face: run
huggingface-cli delete-cachefor an interactive list of cached models you can remove.LM Studio: open the My Models tab and delete models from there.
Step 2: Move the rest to an external SSD
Models load straight from disk into memory, so a fast USB-C or Thunderbolt SSD works well. Load times are a little longer than from the internal disk, and generation speed is the same once the model is loaded.
Ollama
Ollama reads the OLLAMA_MODELS environment variable. Quit Ollama, move the folder, then point Ollama at the new location:
mv ~/.ollama/models /Volumes/MyDrive/ollama-models
launchctl setenv OLLAMA_MODELS /Volumes/MyDrive/ollama-modelsRestart the Ollama app. To make it permanent, add export OLLAMA_MODELS=/Volumes/MyDrive/ollama-models to your shell profile and set it again after a reboot.
Hugging Face
Set HF_HOME (or just HF_HUB_CACHE) to a folder on the drive in your shell profile:
export HF_HOME=/Volumes/MyDrive/huggingfaceThen move the existing cache there, or let it download again.
LM Studio
In LM Studio, open My Models and change the models directory to a folder on your drive. Move the existing model folders there first.
What happens when the drive is unplugged?
The tools simply won't find those models until you reconnect. Nothing is lost. If you use one model every day, keep that one on your internal disk and offload the rest. That's usually the best balance.
A simple routine
Keep one or two daily-driver models on your Mac.
Offload the models you're “saving for later” to an external SSD.
Once a month, remove the experiments you'll never run again.
The short version
Ollama lives in ~/.ollama/models, Hugging Face in ~/.cache/huggingface, LM Studio in ~/.lmstudio/models. Remove what you don't use, and move the rest to a fast external drive with OLLAMA_MODELS, HF_HOME or LM Studio's settings.
Clearbay finds Ollama, Hugging Face and LM Studio models on your Mac and offloads the ones you aren't using to your drive, with a verified copy and one-click restore. Coming soon to the Mac App Store.
