Glossary

What is RAG?

Retrieval Augmented Generation, RAG for short, lets AI answer from your own documents instead of only from what the model learned during training.

5 min readEnterprise AI

In short

RAG, or retrieval-augmented generation, is a method that improves the answers of large language models: relevant information is first retrieved from a database or a set of documents, and the answer is then generated based on that information. For organisations this means AI answers grounded in their own, current knowledge, with sources and within permissions. The quality of a RAG answer depends on what is retrieved: well-structured documents with metadata, versions and access rights make the difference.

On this page

How does RAG work?
Why RAG matters for organisations
RAG, fine-tuning and Agentic RAG
What makes RAG work well?
Frequently asked questions about RAG

How does RAG work?

Step 1
Question
A user asks a question in natural language.
Step 2
Retrieve
The system searches the connected sources for the most relevant documents and passages, within the permissions of that user.
Step 3
Augment
Those passages are added to the question as context for the language model.
Step 4
Generate
The language model writes the answer based on that context and shows its sources.

Why RAG matters for organisations

Current information

Answers use today’s documents, not the model’s training data.

Your own knowledge

Contracts, procedures and project files become a source for AI.

Sources you can check

Every answer refers to the documents it is based on.

Fewer hallucinations

The model answers from retrieved facts instead of guessing.

Permissions respected

Only documents the user may open are used.

No retraining

New documents are available immediately, without training a model.

RAG, fine-tuning and Agentic RAG

Three ways to bring your knowledge to AI
Approach	What it does	When it fits
Fine-tuning	Trains a model further on your data	A fixed style or vocabulary, not knowledge that changes
RAG	Retrieves relevant documents for each question	Answers from current, company-specific knowledge
Agentic RAG	Reasons across information, retrieves the right content and helps complete complex tasks	Questions and tasks that span several sources

What makes RAG work well?

A RAG answer is only as good as what is retrieved. Three things make the difference:

Structured documents: metadata, versions and a clear final version, so the right document is found.
Relevance: semantic search and a relevance model that pick the passages that matter, also called Context Engineering.
Permissions: the system enforces access rights for every question.

That is why SoftAdvice starts with the document foundation in M-Files or SharePoint, and uses Mindbreeze InSpire to connect the sources and provide the context.

What is Context Engineering? · Agentic AI and Context Engineering

Frequently asked questions about RAG

What is RAG?

What is the difference between RAG and fine-tuning?

Fine-tuning trains a model further on your data, which suits a fixed style or vocabulary. RAG retrieves relevant documents for every question, so answers always use current knowledge and can show their sources.

Does RAG prevent hallucinations?

It reduces the risk considerably, because the model answers from retrieved facts and shows its sources. The quality still depends on the documents that are retrieved.

Does RAG respect permissions?

It should. A good RAG system only retrieves documents the user may open. Mindbreeze inherits and enforces existing enterprise permissions for every query.

What is Agentic RAG?

With Agentic RAG, AI goes beyond simple answers: it reasons across information, retrieves the right content and supports users in completing complex tasks, while results stay connected to reliable company knowledge.