How Microsoft 365 Copilot Works

Microsoft 365 Copilot uses Microsoft Graph, the GPT-4 large language model (LLM) from OpenAI, and optionally Bing and third-party services to generate responses. The process that occurs when a user enters a prompt into the Copilot chat UI includes the following:

1) User enters a natural language prompt, which can include references to files and potentially sensitive internal data.

2) The prompt is associated with specific domain data from Microsoft Graph so that the response returned is relevant to the context intended by the user through a process called grounding. Microsoft Graph is a set of services and APIs that provides access to organizational data in Microsoft 365 cloud services, such as documents, e-mails, and Teams meetings. The prompt can also optionally be grounded using Bings search results from the Internet and plugins, which can connect to third-party services, such as ServiceNow.

3) The grounded prompt is sent to the GPT-4 LLM supplied by Microsoft partner OpenAI and hosted in Azure. The LLM processes the prompt and generates a natural language response.