FAISS

FAISS (Facebook AI Similarity Search) is Meta’s open-source C++/Python library for efficient similarity search and clustering of dense vectors. It is the reference implementation for ANN indexes and underpins or inspires many production vector stores.

Index Families

IndexClassCharacter
FlatIndexFlatL2 / IndexFlatIPExact, exhaustive; 100% recall baseline
LSHIndexLSHHash-bucket ANN; low-dimensionality only
HNSWIndexHNSWFlatGraph ANN; fast, high recall, memory-heavy
IVFIndexIVFFlatCluster/Voronoi ANN; needs training
IVF-PQIndexIVFPQIVF + Product Quantization for billion-scale

Supports L2 and inner-product metrics, GPU acceleration, and composite indexes.

Role in the Ecosystem

  • Library, not a service — no filtering, persistence, or dynamic updates on its own; it is a read-only index that must be periodically rebuilt.
  • The step past brute force. For corpora a plain scan still handles, an in-process exact scan needs no library at all; FAISS is what you reach for when the scan stops fitting but a hosted vector database is still more machinery than the problem deserves — “just load them all in memory in FAISS or something and call it done.”
  • Built on or wrapped by Milvus, OpenSearch (k-NN plugin), and used directly in many custom pipelines.
  • Teams needing live updates or metadata filtering often migrate to Weaviate, Qdrant, Vespa, or Pinecone.

Datasets

  • SIFT1M — the workload index-choice tutorials conventionally use

Articles

People

Company

  • Meta — creator and maintainer