• cosine indices (exact + approximate) • drop-in for scikit-learn (metric='cosine') • Incredibly fast approximate IndexBinaryMultiHash with high recall (beats Faiss in some contexts)
https://github.com/pablocael/pynear