You pulled a model in Ollama. It worked. You pulled three more to compare them. Then a few weeks later you notice your Mac is suspiciously low on storage — and Ollama is not the first thing you think to blame.
It should be.
The fast answer
Ollama stores model files in a hidden system folder. The files are named after cryptographic digests, not human-readable model names. There is no obvious way to look at the folder and know what is safe to delete.
Each model you pull can be anywhere from 4 GB to over 40 GB. Pull a handful and you have quietly committed 60–80 GB of your SSD to models you may or may not still use.
Why manual cleanup is harder than it looks
When you open the folder, you do not see llama3.safetensors. You see filenames like sha256-3c4.... There is no label, no date, no indication of which model each file belongs to.
Ollama uses a manifest system — a separate set of small JSON files that point to the actual blobs. If you delete a blob without understanding its manifest, you can corrupt a model without realising it until you try to run it.
There are also orphaned blobs — files left behind after models were removed — that are completely invisible to ollama list but still occupy real disk space. The only way to find them is to cross-reference the manifest files against the blob directory manually.
That is not a weekend project. That is a chore you will do once, get wrong, and never do again.
Want to skip the hidden-folder hunt?
LLM Cleaner scans your Mac for local AI models, caches, indexes, and project memory — then shows what you can review, reveal, export, or safely move to Trash.
What makes this riskier than other file cleanup
Most file cleanup is straightforward: find the big file, decide you do not need it, delete it. Ollama model storage is different because:
- Files are shared between models. One blob can be referenced by multiple manifests. Deleting it breaks all models that use it.
- Ollama may be running in the background. Deleting an active model's files while Ollama is running can cause silent failures.
- There is no built-in size overview.
ollama listshows names and rough sizes, but does not show you the total footprint, orphaned blobs, or duplicate storage.
The right approach
Before removing anything from your Ollama storage, you need to see the full picture: which models are present, how large each one actually is on disk, when it was last used, and whether any blobs are orphaned.
Doing that manually is possible. It is also tedious, error-prone, and the kind of thing you will put off until your disk is completely full.
Local AI storage is messy. Cleanup should not be.
Find Ollama, LM Studio, Hugging Face, Cursor, Windsurf, Claude Code, Whisper, and loose model files in one place.