PromptDustCaches
llama.cpp
llama.cpp caches downloaded model weights (GGUF) here; it holds model data, not conversation content, and is a large_growth signal.
Exposure
- Stores found
- 1
- Holds
- —
- Confidence
- likely
Where it writes
1 store in the current signature database. Paths shown for a typical macOS install.
files
- macOS
~/Library/Caches/llama.cppdir - Linux
~/.cache/llama.cppdir - Windows
~/AppData/Local/llama.cppdir
llama.cpp caches downloaded model weights (GGUF) here; it holds model data, not conversation content, and is a large_growth signal.
- Remove unused GGUF files to reclaim disk space; recent builds may instead use ~/.cache/huggingface/hub.
Source: huggingface.co
What to do about it
PromptDust flags these stores and leaves them in place; cleanup is your call. If a store holds a live secret, rotate the secret first, since deleting the file won’t un-leak a key that already sat in plain text.
For ongoing work, keep sensitive material out of sessions where you can, and prune old transcripts and caches now and then.
This describes where llama.cpp stores data. Metadata only, never your content. Confidence: likely. Spot a mistake? Tell us.
See if this one’s on your machine.
One pass finds every store, for this tool and the other 51.