Glossary

RAG (Retrieval-Augmented Generation)

RAG (Retrieval-Augmented Generation) is a technique in which an AI language model retrieves relevant content from a knowledge source before answering and grounds its response in those passages. Instead of relying on training data alone, the model uses current, company-owned documents. For SMEs, this means answers are based on their own knowledge and can be verified through source references.

Last updated: 2026-08-03

RAG combines two steps. First, the system searches a knowledge base – typically via semantic search, which stores documents as so-called embeddings and finds passages that match in meaning even when the query uses different words. Then the language model composes its answer from exactly those passages and cites them as sources. The model itself does not need to be retrained: new knowledge becomes available as soon as a document is added to the knowledge base.

In everyday business, RAG solves a practical problem: language models know nothing about internal processes, current price lists, or contracts. With RAG, the AI answers questions like "How does our onboarding work?" directly from your own documents – instead of guessing from general training knowledge. Because every answer lists its sources, employees can see what a statement is based on and check it against the original document. That lowers the risk of plausible-sounding but incorrect output, known as AI hallucinations.

Chifty implements RAG with Chifty Knowledge: documents, notes, and connected sources such as Google Drive, OneDrive, SharePoint, and HubSpot form a shared knowledge base that Chifty Chat and Chifty Office build on. One principle applies throughout: the AI must not bypass permissions. The permission filter is applied inside the search itself on every request, before a result exists. Answers draw exclusively on sources the asking person is allowed to see – and display the sources used.

The quality of RAG depends on the knowledge base: outdated or contradictory documents lead to correspondingly weak answers. Anyone introducing RAG should therefore decide which sources to connect, who may maintain content, and how approvals are handled – turning scattered company knowledge into a verifiable basis for answers.

Frequently asked questions

What is RAG in simple terms?

RAG stands for Retrieval-Augmented Generation: before answering, the AI searches a knowledge base – such as your company documents – for relevant passages and builds its answer on them, with source references. The AI answers from your knowledge rather than from its training data alone.

Why does RAG matter for businesses?

Language models know nothing about internal documents, processes, or current data. RAG makes that knowledge usable without training the model on company data: the content stays in the knowledge base, and every answer cites its sources – making statements verifiable.

Does RAG prevent AI hallucinations?

RAG reduces them significantly but does not eliminate them entirely. Because the answer is tied to specific passages that are shown as sources, every statement can be checked against the original document – so incorrect claims surface faster.

RAG (Retrieval-Augmented Generation)

Frequently asked questions

Related terms