RAG (retrieval-augmented generation)

A language model asked an open question will answer from its training data — confidently, and sometimes wrongly. RAG changes the flow: first retrieve relevant documents from a search index, then instruct the model to answer only from those documents, citing them.

The search index is what makes RAG work, which is why AACSearch ships it as a first-class feature: assistants are bound to collections, answers stream with citations to the exact source documents, and a knowledge graph adds entity-level structure behind the retrieval.