Guide
Glossary
Retrieval-augmented generation (RAG)
Retrieval-augmented generation (RAG) is a way of making an AI model answer from a chosen set of documents rather than from what it learned in training alone. When someone asks a question, the system first searches the documents, such as precedents, policies or procedures, and retrieves the most relevant passages. It then gives those passages to a large language model along with the question, and the model writes an answer based on them, usually citing its sources. RAG does not retrain the model, so documents can be added, updated or removed at any time. It reduces hallucinations but does not remove them, so answers still need checking against the cited sources.
Also called RAG, retrieval augmented generation, grounded generation
Last updated:
Example
In an accounting firm
For example, a 40-person accounting firm sets up a RAG assistant over its internal procedures, engagement letter templates and tax-time checklists. A graduate asks how the firm handles a client who has missed a BAS lodgement, and the assistant answers from the firm's own procedure, linking to the document it used so the graduate can read the source.
Guides that explain it in context
Secure by design. Set up correctly. Fully managed.
Talk to us before you commit to anything
Start with a free 45-minute discovery call. We look at your systems and priorities, then recommend a first step with a fixed scope, or tell you if we are not the right fit.