A vector database stores embedding vectors (numerical representations of text, images, audio) and supports approximate nearest-neighbour search across them. It is the retrieval substrate behind most RAG (retrieval-augmented generation) implementations.
Common choices in 2026: Pinecone, Weaviate, Qdrant, pgvector (PostgreSQL extension), MongoDB Atlas Vector Search, and AWS OpenSearch. Each makes different trade-offs across operational model (managed vs self-hosted), query latency, recall quality, and hybrid (vector + keyword) search support.
Indian enterprises increasingly default to pgvector when the data is already in Postgres -- one fewer system to operate -- and reach for Pinecone or Weaviate when query volume crosses a few million vectors with strict latency SLAs.