When is pgvector enough, and when is a separate service worth considering?
Answer out loud or in your head first, then compare.
Answer
If I already use PostgreSQL, the pgvector extension can provide vector search without running another system. That's convenient when the scale and requirements fit its capabilities. A separate search engine may be better for other requirements around performance, ranking or maintenance.
First I measure the real data: the number of vectors, how often they change and the response time. Fast approximate search can miss some results in exchange for lower computation cost. So I check it against an exact comparison. Permission filters can greatly reduce the number of hits; that has to be tested on the version and index configuration in use, not just on an open data set.
Why it matters: Sensible use of a familiar database instead of automatically adding a tool.
Possible follow-ups: What do you gain and what do you lose with approximate search?
How did it go? Be honest — “Not yet” puts the question on your review list.
Start with
- G-P01 · What is RAG? — RAG and retrieval
- G-P02 · What are embeddings? — RAG and retrieval
Further reading
- pgvector: official repository and documentation(opens in a new tab)Docs; accessed 22.10.2026. Exact vs ANN search, indexes, and how filters affect search.