Source: Conversation state (OpenAI guide)

OpenAI's guide to multi-turn state in the Responses API: previous_response_id chaining, the Conversations API, 30-day storage, manual replay and compaction pointers.

Contents
  • What: official API documentation page
  • Publisher: OpenAI
  • Date: living document; read 2026-11-14
  • Original: Conversation state

Summary

Describes the options for keeping a conversation going across Responses API calls and how long OpenAI keeps the state.

Key claims

  • previous_response_id chains a response to an earlier one; all earlier input tokens in the chain are still billed as input.
  • In WebSocket mode, if a previous response ID cannot be resolved from cache, resend the full context with previous_response_id: null.
  • The Conversations API creates a durable conversation object (messages, tool calls, tool outputs) passed via the conversation parameter; it can be reused across sessions, devices or jobs.
  • Response objects are stored for 30 days by default; conversation objects and their items are not subject to the 30-day TTL.
  • Manual mode: replay the full output array for reasoning models so reasoning items and assistant phase values stay intact; models with persisted reasoning can use reasoning.context: "all_turns".
  • The context window counts input, output and (for some models) reasoning tokens; excess generated tokens may be truncated.
  • Compaction is available server-side (context_management with compact_threshold) or via a standalone /responses/compact endpoint.

Feeds

OpenAI Responses API

Suggest changes →