PydanticAI

Most agent frameworks treat type safety as an afterthought. PydanticAI brings Pydantic's validation philosophy to agents — typed dependencies, structured results, and compile-time-like safety that catches errors before runtime. Built for developers who want robust, testable agent code.

PydanticAI is like TypeScript for agents — you define exactly what goes in and what comes out, and the framework catches mismatches before they become production bugs.

Scenario

Support agent with typed deps (database, config) returns a structured SupportResponse (answer, confidence, suggested_actions). Tools access the database via RunContext. Tests use FunctionModel — no API calls needed.

Solution

In Agent Frameworks, apply PydanticAI to this scenario: Support agent with typed deps (database, config) returns a structured SupportResponse (answer, confidence, suggested_actions). 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 PydanticAI (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 PydanticAI happens in the code.

1from pydantic_ai import Agent, RunContext  # import dependencies2from pydantic import BaseModel  # import dependencies3from dataclasses import dataclass  # import dependencies4
5@dataclass6class SupportDeps:  # define a data structure or component7    db: DatabaseConn8    customer_id: str9
10class SupportResponse(BaseModel):  # define a data structure or component11    answer: str12    confidence: float13    suggested_actions: list[str]14
15agent = Agent(16    "openai:gpt-4o",17    deps_type=SupportDeps,18    output_type=SupportResponse,19    system_prompt="You are a helpful support agent.",20)21
22@agent.tool23async def lookup_order(ctx: RunContext[SupportDeps], order_id: str) -> dict:24    return await ctx.deps.db.get_order(ctx.deps.customer_id, order_id)  # return the result25
26result = await agent.run(27    "Where is my order #12345?",28    deps=SupportDeps(db=db, customer_id="cust_1"),29)30print(result.output.answer)  # typed SupportResponse

• Untyped deps — loses the main benefit
• Not using structured output — parsing strings manually
• Testing against real APIs instead of FunctionModel
• Using PydanticAI for complex multi-agent orchestration
• Ignoring async — framework is async-first

Quick recap — the most important points from this module.