How can I retrieve the used OpenAI responses ID?

[rodriciru]

Hi.

Now I can get the Gemini Interaction ID via an AG-UI CUSTOM event named geminiinteractionId, but how can I get this value but for a OpenAI responses, so I can resume the conversation from an ID instead of a full messages conversation??

[DahanItamar]

With the current OpenAI Responses adapter, the ID is exposed on the RUN_FINISHED event, under metadata.tanstack.responseId:

let responseId: string | undefined

for await (const event of stream) {
  if (event.type === 'RUN_FINISHED') {
    responseId =
      event.metadata?.tanstack?.responseId ?? event.responseId
  }
}

The metadata path works on the server stream. The top-level fallback also handles events where the client has restored TanStack metadata. This is the provider's resp_... ID; runId and threadId identify TanStack's run/thread.

For the next turn, pass it through modelOptions:

const nextStream = chat({
  adapter,
  messages: [{ role: 'user', content: followUp }],
  modelOptions: {
    previous_response_id: responseId,
    store: true,
  },
})

Capture a defined ID before starting that continuation, and keep response storage enabled on the initial request too. Send your system instructions again on subsequent requests; previous_response_id does not carry over the previous top-level instructions. See OpenAI's conversation-state guide.

I verified this with @tanstack/ai@0.69.0 and @tanstack/ai-openai@0.30.0 using mocked Responses SSE events: the finish event contained the response ID, and the next request forwarded it as previous_response_id while sending only the follow-up message. This was a local fixture check, with no live API call.

Relevant source: Responses finish event and continuation option.

[rodriciru]

Tested with @tanstack/ai@0.68.0 and @tanstack/ai-openai@0.29.1. No responseID:

{
  "type": "RUN_FINISHED",
  "runId": "7f0c40fc-61de-4ea3-9a47-7d71baba6a21",
  "threadId": "responses-test-0d0c9c64-9b1a-4d08-8128-9168c12b2c1f",
  "timestamp": 1791906732006,
  "usage": {
    "promptTokens": 22,
    "completionTokens": 96,
    "totalTokens": 118,
    "completionTokensDetails": {
      "reasoningTokens": 87
    }
  },
  "metadata": {
    "tanstack": {
      "model": "gpt-6-luna",
      "finishReason": "stop"
    }
  }
}

Tested with @tanstack/ai@0.69.0 and @tanstack/ai-openai@0.30.0. I get a responseID:

{
  "type": "RUN_FINISHED",
  "metadata": {
    "tanstack": {
      "responseItems": [
        {
          "id": "msg_0bf10f3d323473f9006acf364c99dc87d1a3b7e20970fa30f0",
          "phase": "final_answer"
        }
      ],
      "source": {
        "provider": "openai",
        "api": "openai-responses",
        "model": "gpt-6-luna"
      },
      "model": "gpt-6-luna",
      "responseId": "resp_0bf10f3d323473f9006acf364a871087d18c3526104a916565",
      "finishReason": "stop"
    }
  },
  "runId": "2ed67396-8c45-468d-9d4a-bb9c04725c13",
  "threadId": "responses-test-33773070-7077-423f-b0f8-0799fd38dd9e",
  "timestamp": 1791964748941,
  "usage": {
    "promptTokens": 22,
    "completionTokens": 131,
    "totalTokens": 153,
    "completionTokensDetails": {
      "reasoningTokens": 122
    }
  }
}

[DahanItamar]

Thanks for testing both. That matches the changelogs: responseId on RUN_FINISHED (metadata.tanstack.responseId) was added in #1698, released in @tanstack/ai 0.69.0 and @tanstack/ai-openai 0.30.0. So those are the minimum versions for this approach, and on 0.68.0 / 0.29.1 the field doesn't exist yet.