LLM Token Cost Management

Pydantic AI re-sends the whole run history every step, so token spend goes quadratic by design

A dev.to walkthrough of the run graph finds one append-only history per run, copied whole into each model call. The step count sets the exponent; your tool outputs set the price.