Agent chat built into SharePoint
A HubSite365 walkthrough shows a SharePoint Framework (SPFx) agent chat built with the OpenAI GPT SDK and Semantic Kernel, streaming responses via Azure Functions and wiring into Microsoft Graph for tenant data access. (hubsite365.com)
A new SharePoint demo shows how a chat agent can live inside a SharePoint page instead of a separate app, with the web part sending prompts to a cloud backend and streaming answers back into the page. (hubsite365.com) The front end uses SharePoint Framework, Microsoft’s client-side model for building SharePoint web parts and extensions with modern web tools. Microsoft says SharePoint Framework is its primary replacement technology for SharePoint add-ins in SharePoint Online and remains a core extensibility model across Microsoft 365. (learn.microsoft.com, learn.microsoft.com) In the walkthrough, the SharePoint web part collects a user’s message, sends it to an Azure Function, and receives a streamed response back to the browser. Azure Functions is Microsoft’s serverless service for running code without managing servers, and Microsoft documents it as a common way to build web APIs and event-driven backends. (hubsite365.com, learn.microsoft.com) The language model layer sits behind that backend, not in the browser, which keeps application secrets off the client side. The HubSite365 summary says the backend uses the OpenAI GPT software development kit with Semantic Kernel to handle chat completions and orchestration. (hubsite365.com, platform.openai.com) Semantic Kernel is Microsoft’s agent framework for wiring models to functions, plugins, and chat history, so the model can decide when to call code instead of only generating text. Microsoft describes it as middleware that translates a model’s function call into executable code and returns the result to the model. (learn.microsoft.com, learn.microsoft.com) The other key piece is Microsoft Graph, the application programming interface layer that exposes Microsoft 365 data such as SharePoint, OneDrive, Outlook, Teams, Planner, and users. Microsoft says apps must be registered, authenticated, and granted the right permissions before they can call Graph data on a user’s behalf or as an app. (learn.microsoft.com, learn.microsoft.com) That architecture turns a SharePoint page into a controlled front door for tenant data. Instead of asking a model to answer from general training data alone, developers can let it pull live information from Microsoft 365 services through Graph and then format the reply inside the SharePoint experience. (hubsite365.com, learn.microsoft.com, learn.microsoft.com) Microsoft has been signaling that SharePoint Framework is part of its artificial intelligence-era roadmap, with a March 24, 2026 developer update describing SharePoint Framework as a foundation for increasingly artificial intelligence-powered Microsoft 365 extensions. This demo fits that direction by keeping the interface in SharePoint while moving model calls and data access into managed services behind it. (devblogs.microsoft.com, hubsite365.com) The result is less a new SharePoint feature than a pattern for building one: web part in the page, serverless broker in Azure, model orchestration in Semantic Kernel, and tenant data through Microsoft Graph. For Microsoft 365 developers, that puts agent chat inside the same portal where employees already read documents, lists, and intranet content. (hubsite365.com, learn.microsoft.com, learn.microsoft.com)