Works fully offline: you point it at folders of .md files (or clip URLs) and it does hybrid search (BM25 and semantic) to return results scored with RRF. Everything stays on laptop and the db is a single SQLite file.
Since there is no query expansion, reranker or generation, it's extremely fast. It comes with a simple Python API, uses a 30M embedding model from HF, and a UI to visualize results more easily.
It's hackable, so anyone can play with it and customize it.