Faiss

FAISS, short for Facebook AI Similarity Search, is a library for efficient similarity search and clustering of dense vectors. It provides algorithms to find nearest neighbors in high-dimensional spaces and to cluster large collections of vectors. Designed for scalable performance, FAISS aims to enable fast searches on modern hardware for very large datasets.

Developed by Facebook AI Research (FAIR), FAISS is open-source and primarily implemented in C++ with Python bindings (pyfaiss). It runs on CPU and GPU, supports indexing and searching across billions of vectors, and is widely used to retrieve similar embeddings produced by neural models in production and research settings.

FAISS offers a spectrum of index types. Exact search can be performed with a flat (brute-force) index. Approximate search uses inverted-file indices (IVF) with product quantization (IVF-PQ) and optimized product quantization (OPQ). It also provides HNSW-based indexes for efficient graph-based search. Indices can be trained on sample data prior to indexing.

The library emphasizes GPU acceleration and supports multi-GPU configurations, batch queries, and integration with Python, NumPy, and PyTorch workflows. It includes clustering tools such as k-means and utilities for preprocessing, normalization, and dimensionality reduction. Outputs include the indices of nearest neighbors and their distances.

FAISS is used in research and production for vector search in recommender systems, image and text similarity, and retrieval pipelines. It is maintained as an open-source project and hosted on GitHub under a permissive license.