Levo help · For your kind of Mac
For AI and ML work: models, Hugging Face, Ollama
Models are the fastest way to fill a Mac. One download is often 5 to 40 GB, and they pile up while you try things. Levo shows them all, and leaves the choice to you.
What Levo looks for
- Hugging Face models and datasets.
- Ollama models.
- LM Studio models.
- PyTorch, Keras and NLTK downloads, and Whisper models.
- Python environments: virtualenvs, conda environments and the pip, uv and poetry caches.
- Jupyter runtime files and Python tool caches.
- Docker and other containers that hold training images. See below.
Models are listed, not ticked
A model costs time and bandwidth to download again, and some are fine-tunes that no longer exist online. So Levo shows models in the list with their size, and you tick the ones you want gone. Smartly doesn’t tick them at Careful or Balanced, whatever the cleanup level. A model can only be ticked by you.
Your agent helps by saying what each one is and when it last ran, which is the part Finder can’t tell you.
How a model is removed
Levo uses the tool’s own clean command where there is one. Anything else is deleted, or moved to the Trash if you turned on Move to Trash in Settings › Agent. A model is gone for good once you clean without it, so check the list first.
Containers and virtual machines
Docker, OrbStack, Colima and Podman keep one big disk image. Levo lists it with its size, and shows Docker’s build cache as a row you can tick. A virtual machine, such as UTM, Parallels or VMware, is listed too and only goes if you tick it, because it holds your own data.
A good routine
- Analyse, then open the AI models group.
- Sort by size and tick the models you tried once and forgot.
- Leave the ones you use. Clean up.
While Ollama is running, its models show as in use and are locked.