Skip to content

Pydantic AI

Pydantic AI is a Python framework for structured model output and workflows with app-defined tools.

Visit Pydantic AI

How does the app review one proposed action?

Pydantic AI can defer a tool request so your app can ask a person. Show the exact action and its arguments, check that the reviewer is allowed to decide, then pass the decision back for that call. An approval of one draft should not authorize a revised draft.

Can a reviewer come back after a restart?

Save the information needed to continue the waiting run. Pydantic AI documents snapshots of finished steps separately from durable execution that can restart an interrupted run. Record the specific approval in your app and decide which effects would be safe to repeat.

What costs sit outside the Python library?

Budget for model requests, hosting, storage and any durable execution service you choose. Keep the Python project, approved action format and decisions so you can reproduce what happened.

What happens if an editor changes a proposed event title?

On paper, have an assistant draft a name for a fictional pottery class and show that exact draft to an editor.

  1. Record the decision
    Write which version the editor approved and who was allowed to approve it.
  2. Change the title
    Create a second draft and show that it needs its own decision, even if the first approval is still saved.