Simple definition
A vector database indexes numerical embeddings to retrieve content by semantic similarity.
Purpose and use
In a RAG, Retrieval-Augmented Generation, application, a vector database can retrieve documentary passages to enrich the model's context.
How does it work?
The pipeline splits documents, calculates embeddings and retains metadata such as source, version and applicable access rights.
Example and verification
For a documentary assistant, verify retrieval of useful passages and filtering that respects the user's permissions.
Points to watch
The database guarantees neither correct answers nor automatic model training. Deleted documents and changed permissions must also be reflected in the index.
In short
Its RAG role connects retrieval, source traceability and access control beyond similarity ranking alone.