Retrieval-augmented generation

A · 10-second intuition RAG looks up relevant passages at question time and pastes them into the prompt, so the model answers from provided text instead of memory alone.