What is a vector database?

Short answer

A vector database stores embeddings — numeric representations of text, images or audio — and answers nearest-neighbour queries over them, so you can retrieve items by meaning rather than by exact match. It does this with approximate indexes such as HNSW or IVF, trading a small amount of recall for large speed gains. It is a component of retrieval, not a synonym for it: most production systems combine vector search with keyword search and a reranking step.

How it works

An embedding model maps content into a fixed-dimension space where distance approximates semantic similarity. The index makes search sub-linear: HNSW builds a navigable small-world graph, IVF partitions the space into clusters and searches a few. Both expose a recall-versus-latency dial, and both need filtering support so results can be restricted by tenant, permission or date without destroying recall.

Example

"How do I cancel?" should retrieve a passage titled "Ending your subscription" — no shared keywords, close in embedding space. The same mechanism fails on "error 0x80070005", where the exact token is the whole query. That is the case for hybrid retrieval, and it is why dedicated vector stores and vector extensions to existing databases both remain viable choices.

Why it matters

The choice is usually operational rather than algorithmic: does the workload need a separate service, or is a vector index inside the existing database enough? Data volume, filtering complexity, update rate and the cost of a second system to operate decide it — not benchmark queries per second.

Common misunderstandings

That you need a dedicated vector database. At moderate scale, a vector index in an existing relational database is often sufficient.

That embeddings are interchangeable. Vectors from different models are incomparable; changing embedding models means reindexing.

That higher dimensions mean better retrieval. Dimension trades storage and latency against marginal quality.

What is a vector database?

Short answer

How it works

Example

Why it matters

Common misunderstandings

Limitations of this answer

Sources

Related answers