What is RAG?

A large language model only knows what it learned in training. It doesn't know your products, your prices, your policies or a customer's order history, and what it does know stops at a date. RAG fixes that by looking things up first. The AI searches your own sources for the passages that matter, and answers from those, often with a link to where the answer came from.

An example. A trade buyer asks your portal: "What's the lead time and fire rating on this door closer?" With RAG, the assistant finds the product's datasheet and current stock, answers in a sentence, and links to the datasheet. Without it, the model can only guess.

Why it matters

• Current without retraining. Update a document or a price, and the next answer uses it.
• Answers you can check. Because the answer comes from a source, it can show you which one.
• Each person sees only what they should. The search can respect permissions, so a customer's assistant only draws on what that customer is allowed to see.

What it doesn't fix

RAG is only as good as what it searches. Out-of-date, contradictory or missing documents give poor answers, and the search can miss what matters if the content isn't organised for it. Most of the work in a good RAG system is in the content and the search, not the model.

RAG is often confused with fine-tuning. Fine-tuning changes how a model writes and behaves. RAG changes what it knows at the moment it answers. For answering from business information that changes, RAG is usually the right starting point.

When plain RAG is enough and when you need a knowledge graph: RAG, GraphRAG and when you need each, from Enceladus AI, the PureNet Group's AI strategy company.