RAG — Retrieval-augmented generation
Retrieval-augmented generation, or RAG, is the technique of finding relevant information in your own sources and handing it to a language model so it writes the answer with that. The model does not memorise the content: it looks it up at answer time.
What is retrieval-augmented generation?
Retrieval-augmented generation, or RAG, is the technique of finding relevant information in your own sources and handing it to a language model so it writes the answer with that. The model does not memorise the content: it looks it up at answer time.
Also: RAG
RAG is not training a model
This is the most common confusion. Training modifies the model with your data, costs a lot and freezes the knowledge at training time. RAG leaves the model untouched and passes it the relevant documents with every question. That is why, with RAG, fixing an article today changes the answer today.
Why it matters
What changes in a SaaS
It is what lets an AI system talk about your product without making it up. It is also what makes your documentation the ceiling: the model cannot answer better than what is written, and if the wrong passage is retrieved the answer will be flawless and wrong.
Retrieval-augmented generation in detail
How it works, in four steps
The documentation is split into passages; each passage becomes a vector representing its meaning; when a question arrives, the vectors closest to it are found; and those passages go to the model along with the question.
Why it searches by meaning, not words
Because customers do not use your words. If your article says "retention policy" and the customer writes "how long do you keep my data", exact matching finds nothing. Vector search does, because it compares meanings.
Where it fails
In retrieval, nearly always, not in the writing. If the wrong passage comes back, the model writes a convincing answer based on it. That is why most of the work improving one of these systems is improving what gets retrieved, not swapping the model.
What RAG does not solve
Data that lives in no document: which plan this customer is on, how much quota is left, whether the last payment went through. That is not retrieved from documentation, it is looked up in your system. It is the difference between a documentation chatbot and an agent.
How Intake handles it
Questions about retrieval-augmented generation
What does RAG stand for?
Retrieval-Augmented Generation: finding relevant information in your own sources and passing it to the model so it writes with it.
Is RAG the same as fine-tuning a model on my data?
No. Fine-tuning changes the model and freezes what it knows; RAG leaves it untouched and looks documents up on every question. With RAG, updating an article changes the answer immediately.
Why does my RAG system answer incorrectly?
Almost always because it retrieves the wrong passage, not because the model writes badly. Looking at what was retrieved for the failed questions usually explains the whole problem.
Related terms
A term on its own is only half understood. These come up in the same conversation.
Embedding
An embedding is a text represented as a list of numbers that encodes its meaning. Two texts that say the same thing with different words produce similar lists, and that numeric closeness is what makes it possible to search by meaning rather than by exact match.
LLM — Language model
A language model is a system trained on enormous amounts of text that, given a fragment, predicts how it continues. Writing, summarising, translating and holding a conversation all come out of that one simple capability: they are all ways of continuing a text plausibly.
Hallucination
A hallucination is an answer from a language model that sounds right and is not: an invented figure, a feature that does not exist, a plausible number nobody checked. It is not a bug, it is a direct consequence of how a model works — it produces the most probable continuation, not the verified one.
AI agent
An AI agent is a system that, beyond generating text, reads data and takes actions to finish a task. In support that means it checks the actual state of the customer account before answering, and when the answer involves doing something, it does it instead of explaining how.
Knowledge base
A knowledge base is the organised set of articles documenting how a product works and how to resolve what gets asked often. It serves two audiences at once: customers looking up the answer themselves, and the team using it as the source for their replies.
Chatbot
A chatbot is a program that holds a written conversation with a user. In support, current ones index the company documentation and answer from it; earlier ones followed decision trees written by hand.
