Microsoft 365 Copilot Retrieval API

Key Takeaways

[Music] We recently released in preview the copilot retrieval API which allows you to consume the enterprise content of your Microsoft 365 tenant using a rest API available through the Microsoft graph. Let me dig into how it works and let me show you a brief demo about how you can leverage this new capability. So let's see what the retrieval API is and it is a natural language quering model over your Microsoft 365 content via a secure compliant and scalable platform. You can use the retrieval API in your own applications to consume your enterprise content using the same approach that end users can use in Microsoft 365 copilot. Currently the retrieval API targets the SharePoint online content as well as the copilot connectors and you can use filter expressions to define for example with KQL or with filters the specific scope of your queries. In order to consume the retrieval API from a graph perspective and from a permission perspective you will need to have permissions for files readall size. Roll and external item read all. This technology is nowadays in public preview. So, let's have fun with it. Let's assume that we have an HR department with a bunch of documents about HR related topics like for example how to hire people in our company and stuff like that. Now, let's say that we want to use the retrieval API provided by Microsoft Graph through the graph explorer. And to do that you can simply go to aka.ms/ge to go to the graph explorer. From there once you sign in in your account you can rely on the new retrieval API which is available in the beta uh family of APIs for Microsoft graph and it is the copilot/ retrieval. You need to make a post request and you need to provide a JSON payload in the request which will define the actual content of your request. So here for example I'm going to provide this prompt. The query string defines the prompt you send to the copilot engine and it will be how can we hire new employees in our company. I want to target SharePoint data sources and as I told you you can use and target shareepoint online or copilot connectors. I'm going to select that I want to retrieve from the documents that are matching my criteria the title and the author and I want to get up to 20 items in the output. So let me try to run this request. By doing that you can see we can get back a list of it which will define the text matching our criteria as well as the metadata of the items and the URL of the documents matching the criteria. And as you can see in the URL, we get content also from other sides than the one I was showing you before, which is this one. If you want, however, you can rely on the filter expression property in the request. And as you can see right here, I still say that I want to get data from SharePoint as the data source. But now in the filter expression, the path that I'm interested in is the one of the SharePoint online set collection that I was showing you before. And all of the other criteria are the same. So if I will run this query now, we can see that we will get back responses only from the specific site collection. So this is really powerful because you can target the specific scope of your prompt. Moreover, if you want, you can also target the copilot connectors and as such if you want to do so, you can select as the data source the external items in the copilot connectors. Still we can use a prompt. In this scenario, I'm targeting my tenant where I know I have a copilot connector with content about how to write prompt for copilot. And so my prompt is how can I write a prompt for copilot? And still I'm interested in retrieving up to 20 items. So if I will run this query again now we can see that the output comes out from the copilot connector we can see the URL of the items matching my criteria as well as the text excerpt extracted from the document and which is matching the criteria provided in the prompt. So this is really really powerful and easy to use and allows you to have access to the copilot capabilities through npi based on Microsoft graph. Here you can find additional technical information about the retrieval API in the Microsoft graph documentation as well as the public announcement of the public preview of this technology. Let's have fun with it. Thank you.