faiss (facebook ai similarity search)

FAISS (Facebook AI Similarity Search) is a library for efficient similarity search and clustering of dense vectors. **Purpose**: Find nearest neighbors in high-dimensional spaces, orders of magnitude faster than brute force. Open source from Meta. **Key capabilities**: GPU acceleration, billion-scale search, multiple index types, clustering, dimensionality reduction. **Index types**: **Flat**: Exact search, baseline. **IVF**: Inverted file, clusters for faster search. **HNSW**: Graph-based, best accuracy/speed tradeoff. **PQ**: Product quantization for compression. **IVF+PQ**: Combined for scale. **Use pattern**: Build index on embeddings, query returns k nearest vectors by ID. **GPU support**: Dramatic speedup for large-scale search. Index can live on GPU. **Scale**: Handles billion-vector datasets with appropriate indexing and sharding. **Integration**: Python bindings primary, C++ core. Used under the hood by many vector databases. **Training**: Some indexes (IVF, PQ) need to be trained on representative data before adding vectors. **Comparison to vector DBs**: FAISS is library/building block. Vector DBs add persistence, filtering, APIs. **Use cases**: Core of similarity search systems, RAG pipelines, recommendation, and more.