Skip to main content

Vector database

A vector database is a database built to store and search embeddings, the numerical vectors that represent the meaning of text. Its special skill is finding, very quickly, the items whose meaning is closest to a query, even across millions of entries.

Where a traditional database answers "find the rows where this field equals that value", a vector database answers a different question: "find the items most similar in meaning to this one". That shift, from exact matching to similarity, is what makes semantic search possible.

How similarity search works

Every chunk of your knowledge is stored as a vector, a point in a high‑dimensional space. When a query arrives, it is turned into a vector too, and the database returns the nearest stored vectors, the ones whose meaning is closest. These nearest matches are the most relevant pieces of knowledge for that query.

To stay fast at scale, vector databases use specialised indexes that find near‑neighbours without comparing against every single entry.

Vector database vs traditional database

Traditional databaseVector database
StoresStructured rows and fieldsVectors (embeddings)
Finds byExact values and rangesSimilarity of meaning
Best forPrecise, structured queriesOpen‑ended, semantic search
Example query"orders where total > 100""passages about late deliveries"

The two are complementary, not rivals: structured questions still belong in a traditional database, which is why retrieval and SQL tools often work side by side.

In Rational AI

The vector database is the engine behind retrieval. The embeddings generated from your Knowledge are stored there, and at answer time it supplies the most relevant chunks as context for the model. You don't manage it directly; it works quietly underneath the Knowledge and RAG features.


Additional resources