AI glossary
Vector database
A database designed to store embeddings and quickly find the ones most similar to a query.
Normal databases find exact matches; vector databases find nearest neighbours — the stored items whose meaning is closest to your question. They are a common piece of RAG systems: documents are split into chunks, each chunk is embedded and stored, and at question time the most relevant chunks are retrieved and handed to the model.
Many general-purpose databases now offer vector search as a feature, so a separate product is not always needed.
Example: A company stores 5,000 help articles as vectors. When a customer writes “my verification code isn’t arriving”, the database returns the articles about verification texts and emails, even though they do not share the same words.
In practice
- With only a few documents you do not need one: putting them in the context may be enough.
- Store each chunk’s source (document and section) alongside it so you can cite it.
- Combining keyword and meaning-based search (hybrid search) usually gives better results.