← AI Terminology

FAISS - Facebook AI Similarity Search

FAISS is a library from Meta for efficient similarity search and clustering of dense vectors, including exact and ANN indexes on CPU and GPU.

It is the classic toolkit behind countless retrieval systems.
Why It Matters in AI
Before managed vector DBs were everywhere, FAISS was how teams shipped dense retrieval. It remains the benchmark library for index research and high-performance local search.
Key Points
Aspect Description
Use Research, custom RAG, embedding evals
Origin Meta/Facebook AI Research
Indexes Flat, IVF, HNSW, PQ, OPQ composites
Related ANN, ScaNN, vector databases
Hardware CPU and CUDA GPU indexes
Ecosystem Wrapped by LangChain, many vector DBs historically
Simple Analogy
A Swiss Army knife workshop for building custom “find similar vectors” machines — choose blades (indexes) for your scale and memory.
Common Usage Examples
  • faiss.IndexFlatIP for exact inner product
  • IndexIVFFlat + train on sample
  • GPU GpuIndex for bulk search
  • Export/import index files for deploy
Summary
In short: FAISS is Meta’s high-performance library for vector similarity search — the classic foundation for dense retrieval systems.