Build a per-run spending controller with the Responses API

# Build a per-run spending controller with the Responses API An agent using the Responses API may call a model several times to finish one task. Organization and project spending limits cover overall usage, but they cannot tell you whether that task can afford its next request. Give each run its own budget. Before each model request, count the input tokens and set aside the most its response could cost. When the response arrives, record its cost and return any unused money to the budget. Stop before a request would exceed the remaining balance. Suppose support ticket #4821 asks when order ORD-42 will arrive. The order has shipped and is due Friday. Give the ticket a $0.02 budget. The first reply costs $0.01. Another reply could cost up to $0.0146. With only $0.01 left, the application stops. The controller handles synchronous, non-streaming Responses API requests on the default processing tier. Its budget covers model-token costs only; hosted tools and other charges are excluded. The prices, token counts, model names, and limits below are made up. They are not current OpenAI prices, real model limits, or a guarantee of your final bill.