Vector database
A vector database stores content as high-dimensional number sequences (embeddings) and finds entries that are similar in meaning rather than merely identical in wording. It is the technical basis of semantic search and many RAG systems.
In practice
Well-known examples are the Postgres extension pgvector and open-source systems such as Qdrant, Weaviate and Chroma, or managed services such as Pinecone. Classic full-text search compares wording, vector search compares meaning. A search for “cancel invoice” will also surface documents that talk about credit notes instead.
In practice many systems combine both: vector search for meaning, a word-based method such as BM25 for exact matches like error codes or article numbers. The result lists are then merged, often via reciprocal rank fusion.
Small knowledge bases do not necessarily need a dedicated system. An extension of the existing database is often enough – or no vector index at all, if the content fits into the request to the language model in full.
In practice many systems combine both: vector search for meaning, a word-based method such as BM25 for exact matches like error codes or article numbers. The result lists are then merged, often via reciprocal rank fusion.
Small knowledge bases do not necessarily need a dedicated system. An extension of the existing database is often enough – or no vector index at all, if the content fits into the request to the language model in full.