DEFINITION
A database that stores text as series of numbers, called embeddings. It’s used to find passages with a similar meaning even when the words differ. Many assistants that answer questions from documents rely on one.
How it works
An embedding model turns a passage of text into a long list of numbers, a vector, so that passages with similar meanings end up with similar numbers. A vector database stores these vectors, each linked to the passage it came from, and quickly finds the ones closest to a new vector.
Search then works by meaning. The question “How much notice do I have to give?” and a lease clause about “ending the tenancy with one month’s warning” share few words, but their vectors are close, so the clause comes up.
Where it fits in AI assistants
Many assistants that answer from your documents work in two steps. The vector database first retrieves the passages closest to your question, then an LLM writes the answer from them. This is often called retrieval-augmented generation, or RAG. It lets a model use documents it was never trained on, and it’s why a good assistant can quote its sources.
Its limits
Closest in meaning isn’t always right. A vector search can return a passage on the right topic from the wrong document or the wrong year, and it handles exact values such as an invoice number less well than ordinary filters. That’s why many tools combine it with keyword search and filters.
Frequently asked questions
Is a vector database the same as a regular database?
No. A regular database finds records by exact values, such as a date or an invoice number. A vector database finds the entries closest in meaning to a query, which suits text but is less precise for exact figures.