Clean up Ollama and Hugging Face models on a Mac

30 August 2026

Run local AI for a few months and your disk quietly loses 50–150 GB. One pulled model can be 40 GB, every experiment leaves another behind, and none of the tools ever suggest deleting anything.

Where the models live

• Ollama, ~/.ollama/models
• Hugging Face, ~/.cache/huggingface (models and datasets)
• LM Studio, ~/.lmstudio/models
• PyTorch, Whisper, Keras, ~/.cache/torch, ~/.cache/whisper, ~/.keras/models

Delete the tool-native way

Don't rm -rf Ollama's folder, the right way is per model, so Ollama's own store stays consistent:

ollama list then ollama rm llama3:8b

Hugging Face ships huggingface-cli delete-cache for the same reason. Everything is re-downloadable, so the worst case of any deletion is waiting for a pull again.

Or manage them from one screen

ddeploy's AI Junk screen shows every model store with its real size, flags libraries untouched for 60 days, and its model manager lists each Ollama and LM Studio model individually, deleting via ollama rm under the hood, one model at a time, never the whole library by accident. Claude, ChatGPT, Copilot and MCP server caches get the same treatment.