Vector Database

A vector database stores vectors and supports similarity searches. It can enable embedding-based text or image retrieval. These capabilities may be available in a specialist system or as an extension to an existing database.

Vector Database explained

The database supplies search infrastructure; it is not the language model. Retrieving document passages, generating an answer and checking its evidence are different tasks. Separating them helps diagnose problems: did retrieval find the wrong passage, or did generation misrepresent the right source?

Selection depends on collection size, updates, filters, access control and operations. A small internal asset collection differs from a large product catalogue. Start from those requirements; blanket cost comparisons between specialist and integrated systems do not provide a sound basis.

pgvector is a documented example of vector search within PostgreSQL. Depending on the method, a search can determine nearest vectors exactly or use approximation to favour speed. Technical neighbour accuracy differs from domain relevance: precisely similar vectors may still represent unsuitable documents.

Changed and deleted content needs a clear route into the live search collection. Permissions must take effect in actual queries. A successful demonstration does not establish that obsolete price lists, withdrawn assets or another project’s files will be excluded reliably.

Examples

Hypothetical application

An tests retrieval over approved product documents. Alongside speed and relevance, it checks that a withdrawn version disappears and each team can retrieve only its permitted documents.

Key Points

Separate retrieval infrastructure from answer generation.
Assess domain relevance separately from nearest-neighbour accuracy.
Test updates, deletion and permissions.

Practical application

Test selection against a realistic, approved collection, including and permission changes.

Useful measures

Relevant results

Assessment against documented search tasks.

Update delay

Time between a content change and its correct reflection in search.

Common mistakes

Using database speed as evidence of answer quality.
Leaving deleted or restricted content retrievable.

Sources and context

pgvector: Vector similarity search
Official documentation for vector search in PostgreSQL.

Frequently Asked Questions about Vector Database

No. Many applications need no vector search. The task determines the requirement.

Loading related terms…

All Terms

Vector Database

Vector Database explained