noun · AI infrastructure
Vector database
also called vector store
A library shelved by meaning instead of by title. You ask a question and it hands back the items that mean something similar.
A database built to store embeddings and quickly return the items whose vectors are closest to a query vector.
It stores embeddings alongside the original text and answers a query by finding the stored vectors nearest to the query's vector. To stay fast on large collections it usually uses approximate nearest-neighbor search, which can miss a close match.
Real-life example
You save thousands of support articles as embeddings. A customer types a messy question and your app turns it into an embedding too. The database returns the five articles whose numbers sit closest, even though none shares the customer's exact words.
Where the analogy breaks
-
Nearest is not the same as best. The closest items are only the most similar in meaning, so the right answer can rank lower or be missing.
-
Results are often approximate. To stay fast on big collections, many systems trade a little accuracy for speed.
-
It is tied to one embedding model. If you switch models, every stored item has to be embedded again.