In a nutshell
A language model only knows what was in its training text, so it can't cite your private documents or this week's news, and it may make things up. Retrieval-Augmented Generation fixes that: it turns the question into an embedding — a list of numbers capturing its meaning — searches a vector store of your documents for the chunks closest in meaning, pastes those chunks into the prompt, and lets the model answer grounded in that real, retrieved text. To update what it knows, you just update the documents — no retraining.