Based on the user's query, the Agent Planner retrieves a list of tasks or steps in the form of API requests required to gather all relevant information from Dynamics 365. These tasks are executed sequentially, with the results serving as context for the LLM to respond to the user's query.

The process begins when a user sends a request, including system instructions for a Large Language Model (LLM). The LLM utilizes these instructions to generate a list of API requests that the Dynamics 365 connector must execute.

The process then sequentially executes each generated API request against the Dynamics 365 database. The result of each execution is used as context for the subsequent LLM query. This cycle continues, using the outcome of each API request as context data for the next query to the LLM, in addition to the user's initial instructions. The loop persists until all queries derived from the user's input have been processed.

The LLM then uses the combined context to craft a response to the user's original query. The process ensures that the LLM leverages data from Dynamics 365 to deliver a comprehensive and informed response to the user's query.