Retrieval-augmented generation
Finding the relevant documents first and putting them into the prompt, so the model answers from your material rather than from memory.
how it works · the vocabulary
RAGretrieval
The standard shape of every question-answering product: embed the corpus, search it at query time, paste what comes back into the context, and cite it. It is what embedding models and vector databases exist to serve, and it is the cheapest way to make a model right about something it was never trained on.