Vector databases store embeddings for similarity search:
Embeddings: ML models convert text, images, or audio into fixed-size vectors. Similar items have similar vectors.
Use cases: Semantic search, recommendation systems, image search, RAG for LLMs.
Popular options:
- Pinecone: Managed, scales easily
- Milvus: Open source, feature-rich
- pgvector: PostgreSQL extension
Trade-offs: Approximate nearest neighbor (ANN) algorithms trade accuracy for speed. Index build time can be large.