---
title: "Secure RAG Starter Accelerator — Datagist"
description: "Retrieval-augmented generation over SharePoint, OneDrive, and Confluence that keeps document permissions, cites its sources, and logs every request."
url: https://datagist.ai/accelerators/secure-rag
---

[Accelerators](https://datagist.ai/accelerators) Secure RAG Starter

[GenAI & Agents](https://datagist.ai/services/genai-agents) 

# Secure RAG Starter

An assistant over your documents that only answers from files the user can already open.

GIST step [Iterate](https://datagist.ai/approach#iterate) [Ship](https://datagist.ai/approach#ship)

 Download the PDF 

## Get the Secure RAG Starter overview

A short PDF to share with your team. It covers what is in the package, how it works, and a worked example.

On this page The problem What's included How it works Worked example Where it fits

The problem 

## An assistant can reveal what a search box never would.

Point a model at your document stores and it can quote the salary file or the board pack to anyone who asks. Instructions hidden inside a document can also steer what it says.

What's included 

## Answers from your content, inside your access rules.

01 

### Permission-aware ingestion

Every chunk carries the permissions of the document it came from. Ingestion also screens text for hidden instructions and leaves those passages out.

02 

### Filtered search and cited answers

Azure AI Search applies the permission filter inside the query. Azure OpenAI writes the answer, treats retrieved text as untrusted data, and cites the documents it used.

03 

### Logging and limits

Every request is logged, with daily limits on requests and cost. An offline mode with keyword search and extractive answers runs tests and demos without a model.

04 

### Connectors

SharePoint Online and OneDrive through Microsoft Graph with Sites.Selected, and Confluence Cloud through a read-only service account.

How it works 

## How permissions work.

1. ### Principals at ingestion  
Every document gets its allowed and denied groups or users, read from the source system.
2. ### Default is nobody  
A file that matches no rule never gets indexed.
3. ### Chunks inherit  
Splitting a document never widens who can see it.
4. ### Filter before ranking  
Content a user cannot read is never scored or returned.
5. ### Check again  
The assistant re-checks every result and raises an alert if one fails.
6. ### Deny wins  
A denied group overrides an allowed one, whatever other groups the person is in.

Worked example 

## Two employees ask the same question.

Bob works in engineering and Alice works in HR. Both ask for the salary range for band L4.

What happens 
- Bob gets “I could not find this in the documents you have access to.”
- Alice gets the answer, with a citation
- A supplier FAQ holds a hidden instruction to the AI
- Ingestion reports it and leaves that paragraph out

Result 

The rest of the supplier FAQ stays searchable. Every request is logged, and a failed permission re-check raises an alert.

Where it fits 

## What it runs on, and where it stops.

### Works with

- Azure AI Search and Azure OpenAI
- SharePoint Online, OneDrive, and Confluence Cloud
- Microsoft Entra ID groups

### Assumptions and limits

- Permissions are only as current as the last sync, so we agree a sync schedule with you
- Injection screening catches common patterns and is one of three layers of defense
- Production adds vector search to Azure AI Search

Related accelerators 

[GenAI Evaluation Tests Tests the assistant before every release. View details of the GenAI Evaluation Tests](https://datagist.ai/accelerators/genai-evaluation) [AI Agent Controls Governs any agent built on top of it. View details of the AI Agent Controls](https://datagist.ai/accelerators/agent-controls) [AI Governance Starter Lists the assistant in the AI inventory with its risk tier. View details of the AI Governance Starter](https://datagist.ai/accelerators/ai-governance)

## Worried about what your assistant might say?

Tell us where your documents live and who can open them. We will walk you through how those permissions carry over.

Download the PDF [Talk to us ](https://datagist.ai/contact)
