Vector store is a term for the storage layer that holds and retrieves embedding vectors, often used interchangeably with vector database, though it sometimes implies something lighter weight. It is the component an application talks to when it needs to save vectors or find the ones most similar to a query.
The term is especially common in the world of LLM application frameworks like LangChain and LlamaIndex, where vector store is the standard abstraction for whatever system holds the embeddings — whether a full-featured vector database, a library like FAISS, or a simple in-memory structure. This abstraction lets developers swap the underlying implementation without changing their application code.
In practice the distinction between a vector store and a vector database is loose. Some use vector store to suggest a simpler component focused mainly on storing and retrieving vectors, reserving vector database for systems with richer features like advanced filtering, scaling, and operational tooling. But the terms heavily overlap, and both refer to the part of an AI system responsible for keeping vectors and serving similarity search over them.