ModelRefs / FAISS — AI Glossary
FAISS — AI Glossary
Meta AI's open-source library for efficient similarity search and clustering of dense vectors, the foundational ANN toolkit.
Overview
FAISS (Facebook AI Similarity Search) provides highly optimized GPU/CPU implementations of IVF, HNSW, and PQ index families. Supports billion-scale search with <10 ms latency. Used as the search backend by LangChain, LlamaIndex, and many production vector stores. Primarily a C++ library with Python bindings; not a database (no persistence layer).
Reference details
| Topic | rag |
|---|---|
| Last reviewed | 2026-06-24 |
Related terms
Example: The index has to be trained before you can add to it
The clustered index families are not drop-in containers. They must be trained on a representative sample first — that is the step that learns the partition centroids — and only then can vectors be added. Add to an untrained index and it fails; train on an unrepresentative sample and the partition is wrong in a way that silently costs recall, with no error to point at. The other thing to internalise is that this is a library, not a database: an index lives in memory and disappears on restart unless you serialise it yourself, and there is no built-in persistence, replication, filtering or access control.
Commonly confused with
The library provides the index; a vector database wraps one and adds the things production needs — durability, metadata filtering, updates and deletes, backups, multi-tenancy. Choosing between them is choosing whether to build that layer. Many vector databases use this library underneath, so “which index” and “which database” are separate questions.
When to use it
Reach for it when:
- Embedded, read-heavy search inside a single process where you control the lifecycle
- Research and offline batch work — clustering, deduplication, nearest-neighbour analysis
- Benchmarking index families before committing to a hosted vector store
Reach for something else when:
- Where durability, concurrent writes or metadata filtering are requirements
- Frequently changing corpora, since deletes and updates are awkward in clustered indexes
- Without measuring recall against exact search — an approximate index fails silently
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to FAISS — AI Glossary.
Frequently asked questions
What is FAISS?
Meta AI's open-source library for efficient similarity search and clustering of dense vectors, the foundational ANN toolkit.
What concepts are related to FAISS?
Closely related concepts include ann, ivf, pgvector.