Responses API: stored responses, web search, compaction for Codex, and pre-flight suggestions
POST /v1/responses now keeps a response when you send store: true, so the next turn can send previous_response_id instead of the whole conversation, and GET, DELETE and input_items read it. OpenAI models can search the web with the hosted web_search tool and max_tool_calls. Codex can compact long sessions through /v1/responses/compact, and /v1/suggest takes Responses bodies.
Stored responses are encrypted for your team, kept for 30 days from the latest turn of a chain, visible only to the project and environment that made them, and never sent to the provider, which always gets store: false. Nothing is kept without store: true. Where Directives require zero data retention, storing is refused and turning it on deletes what is stored.
web_search needs max_tool_calls: OpenAI can run one more search than it when the model browses, so the hold reserves max_tool_calls + 1 searches, and you are charged for the ones that ran. Compaction serves OpenAI models on OpenAI itself and is billed from the usage it returns.
More in Stored responses.