Structured Results

• Key points are in the visual diagram above.

Downstream code needs OrderDecision, not markdown that maybe contains JSON.

A typed API response. If the body is wrong, you 400. Here the runtime asks the model to fix it.

Scenario

Reply(action='refund', amount=12.5, reason='duplicate charge'). Your FastAPI route returns that object.

Solution

In PydanticAI, apply Structured Results to this scenario: Reply(action='refund', amount=12. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Structured Results (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Structured Results happens in the code.

1from typing import Literal  # import dependencies2from pydantic import BaseModel  # import dependencies3from pydantic_ai import Agent  # import dependencies4
5class Reply(BaseModel):  # define a data structure or component6    action: Literal["refund", "deny", "ask"]7    amount: float | None8    reason: str9
10agent = Agent("openai:gpt-4o-mini", result_type=Reply)11result = await agent.run("Charged twice for order 99")12print(result.data.action, result.data.amount)  # show output for debugging

• result_type=str for a structured workflow
• Huge nested models the model can't fill

Quick recap — the most important points from this module.