Vector database
A store built to answer one question quickly: which of these million embeddings sit closest to this one.
Comparing the query against every stored vector is exact and too slow past a certain size, so these stores use approximate indexes: they trade a small share of correctness for an answer in milliseconds. That trade is a setting you choose, not a hidden defect.
The word database oversells what most projects need. Postgres with pgvector, or the search engine already running, will carry tens of thousands of documents without adding a piece of infrastructure. A dedicated store earns its place at a scale most sites never reach.
Quality is decided upstream of the index, in how documents were cut into pieces. A chunk that separates a table from its heading, or a clause from the condition it depends on, produces confident wrong answers that no index can repair.