What is a vector database?

A vector database is a database built to find content by meaning. It stores text or images as embeddings, which are lists of numbers that represent meaning, and quickly finds the items closest in meaning to a search. Most AI systems that answer questions from company documents use one to find the right passages. Some are standalone products, and some are add-ons to ordinary databases, such as pgvector for Postgres.

Why it matters for a business

In a retrieval (RAG) system the vector database decides which passages the model sees, and when a RAG system answers wrongly, the usual cause is the wrong passage coming back from this step. It is also a store of company data in its own right. The fingerprints are derived from the original documents and carry their sensitivity, so the database needs the same access rules and deletion policy as the documents themselves. It is easy to forget, because on most diagrams it is a small box behind the model.

Go deeper

How AI uses tools and memory

On Atlas: Vector databases, compared

Related terms

• Embeddings
• Semantic search
• RAG (retrieval-augmented generation)
• Grounding

All terms, A to Z