Vector databases store embeddings — dense numerical representations of text — and support similarity search (e.g., cosine similarity or dot product). Popular options include FAISS (local), Chroma, Pinecone, and Weaviate.

FAISS (Facebook AI Similarity Search) is a library for efficient similarity search over dense vectors. It runs entirely in memory, making it ideal for small-to-medium corpora in development and testing scenarios. FAISS supports both exact search (Flat index) and approximate nearest neighbor search (IVF, HNSW) for larger datasets.

Choosing between vector stores depends on scale and deployment: FAISS is best for local dev and prototyping; Chroma adds persistence and a simple HTTP server; Pinecone and Weaviate are managed cloud services suited for production with millions of vectors.

Embedding models map text to fixed-size float vectors (e.g., 768 or 1536 dimensions). The quality of retrieval depends heavily on the embedding model — domain-specific or fine-tuned embeddings outperform generic ones on specialized corpora.
