Plan context windows and prompt budgets.
Check whether your prompt, history and context fit a model's window.
Budget system, example and user tokens against a model.
Estimate the added cost of few-shot examples.