Glossary
What is RAG (Retrieval-Augmented Generation)?
Last updated:
RAG (Retrieval-Augmented Generation) is a method in which a language model first retrieves matching documents before answering and grounds its answer on those passages, rather than on its trained memory alone.
The process has two stages: a retriever finds the relevant passages in a document corpus, and the model formulates the answer from them. This keeps answers current (the corpus can be maintained without retraining) and verifiable (the source passage can be named).
RAG is the difference between a model that sounds plausible and one that is right. For company knowledge like product catalogues, policies and contracts, it is the standard architecture, because the model answers only from the documents released to it.
Why it matters for the mid-market
RAG is what makes AI usable in the mid-market: the model answers questions from the company’s own documents instead of the internet, and the answer remains verifiable against what the company released.