A vector database finds the closest matching embeddings out of millions in milliseconds.
DefinitionWhat it means
A vector database, such as Pinecone, Weaviate, or pgvector, is a data store purpose-built to hold large collections of embedding vectors and to answer similarity queries efficiently, typically returning the nearest neighbors to a query vector rather than exact matches. It combines vector indexing algorithms with the metadata filtering and durability features expected of a normal database.
Why it mattersWhy you should care
Vector databases are the retrieval backbone of most RAG systems, semantic search products, and recommendation engines, since a naive brute-force comparison across millions of embeddings would be far too slow for real-time use. Choosing between a dedicated vector database and a vector extension bolted onto an existing database like Postgres is a common architecture decision that trades operational simplicity against specialized scale and speed.
At a glanceSee it
How a vector database chooses between an exact brute-force scan and an approximate index — trading recall for speed.
The greedy graph walk inside an HNSW index — hop to ever-closer neighbors, then descend to read off the top matches.
Where you see itIn the wild
- RAG architecture diagrams showing a vector store step
- Vendor comparisons between Pinecone, Weaviate, and pgvector
- Infrastructure cost discussions for indexing millions of embeddings