The Faiss library is an industrial-grade toolkit for approximate nearest neighbor (ANN) search in vector databases. It provides a variety of indexing methods and related primitives for searching, clustering, compressing, and transforming vectors. Faiss is designed to be flexible, efficient, and usable both in simple scripts and as a building block for database management systems (DBMS). Unlike other libraries that focus on a single indexing method, Faiss contains a range of indexing methods that often involve a chain of components, such as preprocessing, compression, and non-exhaustive search. This flexibility is necessary to accommodate different usage constraints and efficiency requirements. Faiss is not a feature extractor, a service, or a database. It only indexes embeddings that have been extracted by other mechanisms. The library focuses on carefully implemented algorithms to ensure efficiency and accuracy. The core of Faiss is the *index*, which stores a set of database vectors and returns the vector closest to a query vector in terms of Euclidean distance. Faiss supports various search methods, including k-nearest neighbors, range search, and batch processing, and allows for trade-offs between speed, memory usage, and accuracy. The paper discusses the design principles of Faiss, including the trade-off between different constraints in vector search. It introduces two main tools for achieving this balance: vector compression and non-exhaustive search. Faiss is also engineered to be flexible and usable from other tools. The paper reviews several applications of Faiss, including trillion-scale indexing, text retrieval, data mining, and content moderation. It also discusses related work, including other indexing methods and software packages for vector search, such as FLANN, NMSlib, HNSWlib, and SCANN. The paper further explores the performance axes of a vector search library, including accuracy, resource metrics, and trade-offs between different constraints. It also discusses the exploration of search-time settings, the pruning of parameter spaces, and the exploration of the index space. Finally, the paper provides an overview of Faiss's compression levels, vector preprocessing techniques, and additive quantization options, highlighting the library's capabilities in handling large-scale vector search tasks.The Faiss library is an industrial-grade toolkit for approximate nearest neighbor (ANN) search in vector databases. It provides a variety of indexing methods and related primitives for searching, clustering, compressing, and transforming vectors. Faiss is designed to be flexible, efficient, and usable both in simple scripts and as a building block for database management systems (DBMS). Unlike other libraries that focus on a single indexing method, Faiss contains a range of indexing methods that often involve a chain of components, such as preprocessing, compression, and non-exhaustive search. This flexibility is necessary to accommodate different usage constraints and efficiency requirements. Faiss is not a feature extractor, a service, or a database. It only indexes embeddings that have been extracted by other mechanisms. The library focuses on carefully implemented algorithms to ensure efficiency and accuracy. The core of Faiss is the *index*, which stores a set of database vectors and returns the vector closest to a query vector in terms of Euclidean distance. Faiss supports various search methods, including k-nearest neighbors, range search, and batch processing, and allows for trade-offs between speed, memory usage, and accuracy. The paper discusses the design principles of Faiss, including the trade-off between different constraints in vector search. It introduces two main tools for achieving this balance: vector compression and non-exhaustive search. Faiss is also engineered to be flexible and usable from other tools. The paper reviews several applications of Faiss, including trillion-scale indexing, text retrieval, data mining, and content moderation. It also discusses related work, including other indexing methods and software packages for vector search, such as FLANN, NMSlib, HNSWlib, and SCANN. The paper further explores the performance axes of a vector search library, including accuracy, resource metrics, and trade-offs between different constraints. It also discusses the exploration of search-time settings, the pruning of parameter spaces, and the exploration of the index space. Finally, the paper provides an overview of Faiss's compression levels, vector preprocessing techniques, and additive quantization options, highlighting the library's capabilities in handling large-scale vector search tasks.