Pydantic AI

Typed output · by Pydantic

Pydantic AI

Type safe agents. If you already trust Pydantic for validation, this feels like home.

29 lessons3h 20m reading6 sections

01Your first agent0/7 · 39m

Pydantic AI overview4m

Installation and setup5m

Agent basics and a first run5m

Requests and responses in a run6m

FunctionModel: write a stand-in model7m

Instructions: what the model is told6m

Real models: providers, keys and model names6m

Section quiz · 5 questionsNot taken

02Typed output0/4 · 27m

Structured output: a typed ticket7m

Validation retries: when the model gets it wrong7m

Output validators: rules Pydantic cannot check6m

Output functions and several output types7m

Section quiz · 5 questionsNot taken

03Tools and dependencies0/4 · 25m

Function tools: letting the model look things up7m

Dependencies: giving tools your data7m

Tool errors: ModelRetry and crashes6m

Usage limits: stopping a run that loops5m

Section quiz · 5 questionsNot taken

04Conversations and control0/6 · 37m

Message history: continuing a conversation6m

Trimming history with a history processor6m

Saving messages to JSON and back6m

Streaming: showing the answer as it is written5m

Tool approval: approve, deny or edit a refund9m

FallbackModel: when a provider is down5m

Section quiz · 4 questionsNot taken

05More than one agent0/3 · 25m

Multi-agent delegation: agents that call agents6m

MCP servers: tools from another program9m

Harness Coder: a ready-made coding agent10m

Section quiz · 5 questionsNot taken

06Testing, observability and shipping0/5 · 47m

Testing agents with pytest and TestModel8m

Evals: scoring the agent on many tickets12m

Observability: tracing runs with Logfire7m

Integrations: from stand-in to real backends8m

Support desk project: the finished agent12m

Section quiz · 5 questionsNot taken