• 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