CLU is an LLM based concept I’ve been working on that monitors package feeds like pypi.org to hunt for potentially malicious packages while I do more important things.
It uses a combination of
built in heuristics
levenshtein distance checks for popular package typosquats
LLM analysis thanks to Ollama and the Qwen model (configurable)
This recording shows it logging to console. Plan is to poke this out to a log analyzer or possibly a standalone webservice