RAG (Retrieval-Augmented Generation) introduces external memory during the inference phase to constrain the model's generation boundaries, by chunking documents, embedding them as vectors stored in a vector database, and retrieving relevant chunks to inject into the prompt at inference time.