What is Vector store?
Database optimized for similarity search over embeddings — the storage layer under most RAG setups. Examples: Chroma, LanceDB, Qdrant, Pinecone. Agents query it to find context relevant to the current task.
What it is
A vector store is a database of embeddings: numerical representations of text chunks that sit close together when their meaning is similar. It is the search index behind retrieval-augmented generation. OpenAI describes its vector store as a container that indexes your files for its Retrieval API and file search tool.
How adding data works
When a file is added, the system splits it into chunks, creates an embedding for each and indexes them. In OpenAI's implementation the default chunk is 800 tokens with a 400-token overlap between neighbours, adjustable within stated limits. A query is embedded the same way, and the store returns the nearest chunks with a similarity score and their source file, so a question can match by meaning even when the wording differs.
Things to watch
Retrieval quality depends on chunking and on the embedding model, so test with real questions. Retrieved text goes into the model's context window and is untrusted input. Deletion can lag: OpenAI notes that removed files may briefly still appear in results. Limits and prices differ by vendor; OpenAI documents a 512 MB cap per file and storage that is free up to 1 GB, but check the current page.
Example
You add your help-centre articles to a vector store. A customer asks "how do I get my money back?" and the store returns the refund-policy chunk, which shares almost no words with the question but means the same thing.
Sources
See also
← Back to the full AI agent glossary.