RAG

Retrieval-augmented generation: fetch documents, then answer.

Incoming questions trigger a search over stored documents, and the relevant passages are handed to the model. It stays current and cites sources without retraining.