Use a custom function in an AI agent

Why it matters

Function calling is the base tool contract: the model proposes a call, and your code decides whether and how it runs.

What to practise

Define three FunctionTool objects with JSON-schema parameters, strict=True and additionalProperties: False, and attach them to a PromptAgentDefinition with the Microsoft Foundry SDK for Python. The agent’s response returns function_call items; your client runs the matching Python function and sends function_call_output items back on the same conversation. Watch the agent call two tools in one turn, which arguments it fills from earlier messages, and the report file one tool writes. The exercise notes that some of its technologies are in preview.

How it fits

The strict schemas and client-side execution are 2.3.1’s typed contract and execution owner. The exercise returns outputs on the conversation, not through previous_response_id, because the next message would then fail; that is 2.2.1’s continuity choice. Ask what 2.3.9 would check before running a call the model proposed.