Retrieval-Augmented Generation (RAG)
RAG is an AI technique that retrieves relevant text from approved documents and uses it as the basis for a language-model answer, rather than generating from model memory alone.
Definition
Retrieval-Augmented Generation (RAG) is an architecture for generative AI systems that separates knowledge from reasoning. At query time the system first retrieves relevant passages from a trusted knowledge source — a document store, vector database, or search index — and then passes those passages to a large language model as context for generating the answer. The model is instructed to answer only from the retrieved passages and to cite them. If nothing relevant is retrieved, a well-designed RAG system returns "I don't know" instead of inventing an answer.
Why it matters
RAG is the most common architectural choice for enterprise AI because it eliminates two fundamental problems of pure LLMs: knowledge freshness (the model's training data is frozen at a cutoff date) and hallucinations (the model fabricates plausible-sounding but wrong facts). By grounding answers in retrievable source material, RAG systems are auditable, citable, and easier to make compliant with regulations like the EU AI Act.
How Volentis.ai handles it
Every Volentis.ai agent uses a RAG pipeline over a customer's approved documents. The retrieval step is scoped to the role-based permissions of the asking employee — an agent only retrieves from documents the user is authorized to see. Every generated answer includes a citation pointing to the exact source passage, and the full retrieval-plus-generation trace is logged for audit.