PRODUCTION SCENARIO
Vantage Health runs a tool-calling agent under Zero Data Retention with store set to false. Between steps the agent forwards only the function_call and function_call_output items, and quality on multi-step cases has dropped against the stored version.
What should the agent include in each follow-up request?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Every reasoning item, function call and function call output since the last user message
With store set to false the reasoning items carry an encrypted_content property, and the documented pattern is to pass back every reasoning item, function call and function call output since the last user message. A summary is a lossy rendering the model cannot resume from, previous_response_id reads server-side state that Zero Data Retention does not keep, and reasoning.encrypted_content on the include list asks the API to return encrypted reasoning rather than being the channel for sending it back, where one item would not cover the whole step sequence anyway.
Official Source: https://developers.openai.com/api/docs/guides/reasoning ↗