swayam-rajput/lura ? reverse-engineered prompt

Build me a fully offline document Q and A tool in Python.

I want to drop in a folder of files like txt, md, pdf, and docx, have them get split up and indexed locally, and then ask natural language questions against that content. It should use a local embedding model and FAISS for search, and a local GGUF model for the final answer, with everything staying on my machine and no external APIs.

Please make the flow simple from the terminal, with options to ingest a file or folder, search by meaning, ask a question with sources, and reset the index if I change the embedding model or want a clean start. If the answer is not in the uploaded documents, it should say so instead of making things up.

I’d also like it to be careful about keeping the embedding model consistent so the index does not get corrupted. If anything is unclear, look up the current docs online if you need to.