Inquiry About Copilot Studio Agent Accessing SharePoint Data and Use of Vector Databases

(1) Share

Report

Posted on by CU08122003-1

Hello,

I am using Copilot Studio and would like to understand how the agent accesses my data in SharePoint. The documentation mentions that "GraphSearch" is utilized for this purpose.

Could you kindly clarify the following:

Data Access Method:
- Is a vector database created where the data from SharePoint is stored, or does the agent directly access the SharePoint content without intermediate storage in a vector database?
Functionality of GraphSearch:
- What exactly is the "GraphSearch" function?
- How does it work in detail?
- Does it use semantic search, or is it based on keyword search?

Thank you in advance for your assistance, and I look forward to your response.

Best regards,

Categories:

General topics

I have the same question (0)

All responses (2)

Answers (0)

Suggested answer

Artur Stepniak 1,539 Moderator on at

Like (0)

Report
Copy link

Link copied!

Hello,

1. It's not possible for GPT model to directly search the content. Vector representation of the data must be created first for GPT to perform a search. I don't know how it's designed from architecture perspective - I assume that it's stored in a vector database somewhere.

2. GraphSearch function is a feature which's used by GPT model to communicate with the Graph API to retrieve various information based on the user prompt. I don't know how it works in detail and MS won't share it I guess, because it would show others how to design it. I assume that it performs an API call via Graph API to retrieve the data first and then it uses RAG technique to generate a response - that's an obvious approach in that case.

In case of any other questions, let me know. If the answer helped you, mark it, so that others can benefit from it.

Best regards,

Artur Stepniak

Was this reply helpful? Yes No
Craig_Humphrey 190 on at

Like (0)

Report
Copy link

Link copied!

Hey CU,

are far as I'm aware:

1. The SharePoint Connector in Copilot Studio is primarily a wrapper for Microsoft Graph API, which is scoped to just SharePoint's semantic index. So, there's no new semantic index, just SharePoint existing one.

Thislooks to be an internal (to Microsoft) only facing API, as I can't find any solid documentation around it.

Back in the days of FAST (SharePoint 2010), SharePoint had public facing search APIs that supported semantic search (as well as traditional keyword/metadata), and while it sounds like Microsoft has kept the semantic index, they no longer expose the APIs to query it.

2. As per the answer to Q1, I believe it's leveraging Microsoft's Graph API, but it appears to be an internal only endpoint, so there's no real detail.

The closest I can find is you can do your own queries against Copilot Studio's Retrieval API (Microsoft 365 Copilot Retrieval API Overview (Preview) | Microsoft Learn)

Hopefully that's helpful, even if there's some speculation involved.

If you got a better answer elseware, perhaps you'd like to update this post.

Thanks

Craig

Was this reply helpful? Yes No