AI models are filling your disk: Ollama, LM Studio, Hugging Face and ComfyUI

A single local model can be 4 to 40 GB. Try Ollama, LM Studio and ComfyUI and you may have the same model three times, in three different folders, with three different names.

Where the models live

Ollama: C:\Users\<you>\.ollama\models (blobs named by hash).
LM Studio: C:\Users\<you>\.lmstudio\models, or the folder set in its My Models tab.
Hugging Face: C:\Users\<you>\.cache\huggingface\hub.
ComfyUI: the models folder inside your ComfyUI install.

Remove models with each tool’s own command first

Each tool tracks its models. Removing them through the tool keeps its list correct. For Ollama, list and remove models like this:

ollama list
ollama rm <model-name>

Duplicates are the hidden cost

The same GGUF or safetensors file downloaded by two tools takes the space twice. Deleting one copy breaks that tool. A hard link solves it: both paths point to one copy on disk, so both tools keep working and the space is counted once.

Watch the running servers too

Model servers keep gigabytes of memory while loaded, and some listen on your network. Check which ones are running and close those you are not using.

→ where Tidemark helps

Where Tidemark helps

Apps shows each AI tool’s real footprint, including its models.
Duplicates finds identical model files by content and can share one copy through an NTFS hard link, so every tool still works (Pro).
AI tools shows which model servers are running, their memory and ports, and warns when one is reachable from your network.
Limits can cap AI agent session logs that grow forever (Pro).

coming soonto the Microsoft Store

See every feature