Also known as: vector store
In plain English
A vector database is like a library organised by meaning. You ask a question, it turns the question into numbers, and it finds the documents with the closest meaning, even if they use different words.
In practice
Vector databases are a core part of most RAG systems. Options range from dedicated products like Pinecone, Weaviate and Qdrant to extensions of existing databases, such as PostgreSQL with pgvector, which is often enough to start.
Under the hood
Vector databases index high-dimensional vectors with approximate nearest-neighbour algorithms such as HNSW, trading a little accuracy for large speed gains. Production systems combine vector search with metadata filters and often keyword search, known as hybrid search.
Example
"Our policy documents are split into chunks and stored in a vector database."