Definition
Vector Database
A vector database stores embeddings and makes it fast to search for semantically similar documents, images or records.
Short definition
A vector database stores embeddings and supports similarity search. Instead of asking for exact keyword matches, applications can ask for items whose vectors are close in meaning.
How it works
Documents are converted into embeddings, saved with metadata and indexed for fast nearest-neighbor search. When a user asks a question, the question is embedded too, and the database returns the closest matching chunks.
Example
In a RAG assistant, a vector database can retrieve the most relevant policy paragraphs before the language model writes an answer.
Why it matters
Vector databases are important for semantic search, recommendations and knowledge assistants. Their quality depends on chunking, metadata, embedding models and retrieval evaluation. They do not solve data quality problems by themselves.
Vector search and relational search
A relational database is strongest when a query contains exact conditions such as an ID, date, status or price. A vector database finds semantically similar items. Production systems often combine both: structured filters restrict results to documents a user may access, and vector search ranks the most relevant passages.
What to watch
A similarity score is not proof that the result answers the question. Retrieval must be evaluated across chunking strategies, embedding models, result counts and relevance thresholds. Authorization is equally important: semantic search must never return a document that the requesting user is not permitted to read.