Connecting SharePoint to AI: your documents become searchable in plain language
SharePoint and OneDrive can be connected to AI. IntraGPT does it in two ways at once: files are synchronised into the knowledge base so you can search them as if they were one document, and an agent can look inside the library live to open a specific file. Answers come with the source attached, so you always know which document something came from.
Two routes the AI takes to your documents
Synchronise into the knowledge base
Files from the chosen site and library are retrieved and indexed. After that the AI answers substantive questions from several documents at once, with a reference to the file the answer came from.
Walk the folders live
The agent can traverse the folder structure to see what is there. Useful when someone does not know the file name but does know roughly where it lives.
Search on file name
Short, distinctive search terms work best in SharePoint: a project number, a client name, a year. The agent searches with intent rather than requesting whole folders.
Open and read a file
Once the right document is found, the agent reads the contents straight from SharePoint. That is the route for a file that has just changed and may not yet be in the knowledge base.
Inspect metadata
The agent can request a file's details, for example when it was last modified. That lets it say which version it read.
Several sites side by side
Each site or library gets its own connection with its own name. The model therefore keeps the sources apart and does not pull a contract from the wrong department library.
Where daily work speeds up
Healthcare
- Situation
- A nurse wants to know which protocol applies to a given situation, but the protocol library holds hundreds of documents.
- What the agent does
- The agent searches the indexed protocols, answers in plain language and names the document and the version.
- Result
- The answer arrives within seconds, with a source the colleague can check.
Legal
- Situation
- A lawyer is looking for the latest version of a framework agreement with a supplier, but three folders have similar names.
- What the agent does
- The agent searches on the supplier name, compares the modification dates and opens the most recent one.
- Result
- One document instead of half an hour of clicking, with certainty that it is the current version.
Sales
- Situation
- An account manager is drafting a quote and knows something similar was written last year.
- What the agent does
- The agent searches the quotes library by industry and service and summarises how the quotes found are structured.
- Result
- A first draft based on what was actually sent before, not on an invented example.
Onboarding
- Situation
- New colleagues ask the same questions about arrangements and ways of working during their first weeks.
- What the agent does
- The agent answers those questions from the indexed handbooks and points to the full document.
- Result
- New joiners get going faster and the questions do not all land with the same person.
From consent to first answer
-
1
Consent in Microsoft 365
An administrator grants access from your own Microsoft 365 environment for the site and library that may be included. That consent sits with you and can be withdrawn there.
-
2
Choose the scope
Per connection you pick which site and which library are included. Confidential libraries are simply left out; what is not in scope does not exist for the AI.
-
3
First synchronisation
The files are retrieved and indexed. After that the connector keeps up with the library, and the test connection tells you whether access still works.
-
4
Attach the agents
You decide which agents and which roles may use this source. A customer service agent can reach the manuals without reaching the personnel files.
Why this is not a second Copilot
Many organisations compare this with Microsoft Copilot. The difference is where the model runs and how sharply you can bound the scope.
Where the data lives: Nederland of de EU
The model runs on your own server
The contents of your documents are processed on your own server with open-weight models, not in an American cloud.
Scope per library
You explicitly choose which sites and libraries are included. There is no button that switches on the entire tenant at once.
Roles decide who sees what
Per agent you record which sources belong to it and which roles may use it, so a source is never available more widely than intended.
Always with a source
The answer states which file the information came from. That makes what the AI says checkable and stops a summary from taking on a life of its own.
Questions about the SharePoint connector
Can I connect SharePoint to ChatGPT?
ChatGPT has no built-in, managed connection to your SharePoint, and uploading documents to a public chatbot is not an option for most organisations. IntraGPT connects SharePoint inside your own environment, with a scope you choose yourself and a model that runs on your own server.
Does IntraGPT work with SharePoint and OneDrive?
Yes. SharePoint and OneDrive have a native implementation in IntraGPT with synchronisation into the knowledge base and a live file tool. There is also a migration path from the older OneDrive sources.
What is the difference with Microsoft Copilot?
Copilot works inside Microsoft 365 and runs in Microsoft's cloud. IntraGPT runs on your own server with open models, connects your other systems such as ERP and CRM alongside SharePoint, and lets agents carry out work rather than only answering.
Can the AI change or delete files?
No. The connector reads: walking folders, searching, inspecting metadata and reading contents. There is no action that edits, moves or deletes a file.
What does a SharePoint connector cost?
The connector is part of the project we run together. What drives the work is the number of sites and libraries, the size of the document estate and how the permissions need to be divided. We work that out in the first call.
How long does setup take?
Consent and setting the scope take a short session with your Microsoft 365 administrator. After that the size of the library determines how long the first indexing runs; nobody has to sit and wait for it.
Will everyone see every document?
No. Per agent you decide which sources are attached and which roles may use it. An agent that only knows the handbooks cannot say anything about the contracts folder, even when asked.
What about a document that was just changed?
That is what the live file tool is for. If a change is not in the knowledge base yet, the agent opens the file directly in SharePoint and reads the current contents. The answer names the file it used.