Vector Databases
What a Vector Database Stores and Does
A vector database is a database designed to store embeddings, the numeric vectors that represent the meaning of text, images or other data, and to find the stored vectors closest to a given query. Instead of matching exact values, it answers questions such as “which stored items are most similar to this one?”
Each stored record usually holds the vector plus metadata, such as the page URL, title, publication date or access permissions. Queries can combine similarity search with filters, for example “similar articles published in the last year in English”.
How Fast Similarity Search Works
Comparing a query against every stored vector is exact but slow at large scale. Most vector databases therefore use approximate nearest neighbor (ANN) search, which builds index structures that find very close matches quickly. The trade-off is that the search may occasionally miss the single best match. Common index approaches include graph-based and clustering-based methods.
Speed, memory use and recall, meaning how many of the true nearest matches the search returns, pull against each other. Tuning these settings is a normal part of running a vector store.
Dedicated Vector Stores or Database Extensions
There are two common ways to get vector search:
- Dedicated vector databases: systems built only for vector workloads, with scaling and filtering designed around embeddings.
- Vector extensions for existing databases: add-ons that let a general-purpose database store and search vectors. This is often simpler when your data already lives there and volumes are moderate.
Hosting considerations resemble those for other databases: memory, storage, backups and replication. Vector indexes often need a lot of RAM, so estimate the memory a dataset needs before you choose a plan. For how the vectors are created, see what embeddings are.
Searching a Help Center by Meaning
A software company with several thousand help articles wants visitors to find answers by describing their problem in their own words. Each article is split into passages, each passage is converted into an embedding, and the vectors are stored with the article URL and the product version.
When a visitor searches, the query is embedded and the database returns the closest passages, filtered to the product the visitor uses. The company chose a vector extension for its existing database because the content was already stored there and the volume was moderate. The same pattern underlies retrieval augmented generation, where the retrieved passages go to a language model instead of to a results page.
Frequently asked questions
Do I Need a Vector Database for a Small Website?
Usually not. For a few hundred or a few thousand documents, a simple index in your existing database, or even an in-memory library, can be enough. A dedicated vector database becomes more useful as the data grows, when you need filtering at scale, or when many users query at the same time.
Is a Vector Database the Same as a Regular Database?
No. A regular database is optimized for exact lookups, joins and transactions on structured fields. A vector database is optimized for finding the nearest vectors by similarity. Many modern databases offer both, so you can keep structured data and vector search in one place.