PurgatoryFS

 O   O
  \ /
   V

Local LLM code quarantine scanner.

This is a Proof-of-Concept and is intended for fast, cursory scans with an LLM. When it is expanded to a FUSE fs, it will present interesting problems to work through and learn.

Usage

cd /path/to/watch python3 /path/to/purgatory-watch.py # watch current folder python3 /path/to/purgatory-watch.py /custom # watch specific folder python3 /path/to/purgatory-watch.py --model llama3.2:3b # use a different model python3 /path/to/purgatory-watch.py --log scan.log # log all LLM responses

Clean codebases move to ../safe/ Suspicious ones keep PURGATORY_SCAN_REPORT.md in place.

Setup

pip install -r requirements.txt

Install ollama (Linux):

curl -fsSL https://ollama.com/install.sh | sh

Pull a model - pick one based on your RAM:

ollama pull deepseek-r1:8b # default, good balance (~5GB) ollama pull llama3.2:3b # lighter option (~2GB) ollama pull qwen2.5-coder:7b # code-focused alternative

Verify ollama is running:

ollama list

Run python3 purgatory-watch.py --help for all options.