ScaNN

ScaNN, short for Scalable Nearest Neighbors, is Google Research's library for vector similarity search at scale. It combines search-space pruning with learned quantization, and its signature contribution is anisotropic vector quantization, which weights quantization error by its effect on inner-product ranking; a 2023 addition called SOAR further improves indexing for approximate search. The library is tuned for maximum inner product search as well as Euclidean distance, scales to billion-vector datasets, and posts competitive results on standard nearest-neighbor benchmarks. A Python API wraps the C++ core, with optional TensorFlow integration and Docker images for TensorFlow Serving. It requires Linux with Python 3.9 to 3.13 and x86 processors with AVX and FMA support, or ARM with NEON. Released under the Apache 2.0 license within the google-research repository, ScaNN is used for recommendation systems, semantic search, and retrieval layers behind RAG applications.