Typed Tools

• Key points are in the visual diagram above.

Stringly-typed tool calling is how SQL injection and silent None bugs land in agent stacks.

A FastAPI route: you declare types, the framework parses, you never json.loads by hand.

Scenario

get_order(order_id) returns Order. The model cannot pass {id: 1} if the field is order_id: str.

Solution

In PydanticAI, apply Typed Tools to this scenario: get_order(order_id) returns Order. 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 Typed Tools (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 Typed Tools happens in the code.

1from pydantic_ai import Agent, RunContext  # import dependencies2from pydantic import BaseModel  # import dependencies3
4class Order(BaseModel):  # define a data structure or component5    id: str6    total: float7
8agent = Agent("openai:gpt-4o-mini")9
10@agent.tool11async def get_order(ctx: RunContext[None], order_id: str) -> Order:12    """Look up an order by id."""13    return db.get(order_id)  # return the result

• Untyped **kwargs tools
• Doing business logic in the prompt

Quick recap — the most important points from this module.