FAISS

FAISS, Facebook AI Similarity Search, is Meta's library for efficient similarity search and clustering of dense vectors, and it underpins much of the modern vector database and retrieval ecosystem. Written in C++ with complete Python and NumPy wrappers, it implements a wide range of index types: exact brute-force baselines, inverted file and product quantization indexes for compressed billion-scale search, binary vector indexes, and graph-based methods such as HNSW and NSG. Supported metrics include L2, inner product, and cosine similarity, and key algorithms have GPU implementations for NVIDIA CUDA, AMD ROCm, and the NVIDIA cuVS backend. FAISS is designed to scale to collections that do not fit in RAM, with tooling for evaluating the speed and accuracy tradeoffs of each index type. Released under the MIT license, it is used by researchers and engineers doing nearest neighbor retrieval for embeddings, recommendation systems, and semantic search.