Definition
RAG was described in a 2020 research paper by Lewis and colleagues. In a RAG system, a question is first used to search a knowledge source, usually with embeddings and a vector database, sometimes combined with keyword search. The best-matching passages are added to the model's context, and the model answers from them, ideally citing them.
RAG lets a general model answer from up-to-date, private information without retraining it, and makes answers easier to check. Its quality depends mostly on the retrieval: if the right passage is not found, the model will answer from memory or guess. Many enterprise AI projects are, at their core, RAG over company documents.