Pydantic Agents

Learn Pydantic models, validation, and pydantic-ai agents through hands-on katas.

The @dojocho/pydantic-agents dojo teaches Pydantic and pydantic-ai the same way the rest of Dojocho does: one failing test at a time, with a Socratic sensei refusing to write the answer for you.

Install

dojo setup
dojo add pydantic-agents

Then in your agent:

/kata

Katas

The current set covers the fundamentals:

Kata	Focus
001 — basic-model	Defining your first `BaseModel`, fields, types
002 — validation	Validators, constraints, error handling
003 — simple-agent	Building a `pydantic-ai` agent backed by your models

More katas (tools, structured outputs, RAG patterns) are on the roadmap.

Test runner

The dojo uses pytest driven by uv:

uv run pytest katas/001-basic-model/test_solution.py -v

Each kata has its own test_solution.py. The runner adapter is exit-code — green when pytest exits 0, red otherwise. Your sensei runs these for you after each change.

Teaching contract

Lifted from the dojo's DOJO.md:

Never give solutions. Your role is Socratic guide.

Ask questions that steer toward the answer

Point to Pydantic docs or class/method names

Narrow scope: "Focus on the first failing test"

Never write or show solution code

Test-only utilities (fixtures, assertions) belong to the harness — the sensei won't attribute them to you.

Requirements

Python 3.11+
uv for the test runner.
The dojo's prepare.sh installs Python deps the first time you run it.

Source

The dojo source lives in the Dojocho monorepo.