store=false still persists conversation messages to the database

[jaideepr97]

This bug was identified by Claude.

Bug

When a user calls POST /v1/responses with store: false and a conversation ID, the response object is correctly not persisted, but conversation messages and items are still written to the database.

This violates the semantics of store=false, which should prevent all server-side persistence for the request.

Root cause

In _create_streaming_response() (openai_responses.py, lines 1118-1130):

# Incremental persistence (correctly gated)
if store:
    await self._persist_streaming_state(...)

# Conversation sync (BUG: does not check `store`)
if (
    stream_chunk.type in {"response.completed", "response.incomplete"}
    and final_response
    and failed_response is None
):
    if conversation:
        # These two writes execute even when store=False
        await self._sync_response_to_conversation(conversation, input, output_items)
        await self.responses_store.store_conversation_messages(conversation, messages_to_store)

The store flag correctly gates _persist_streaming_state (line 1110), but the conversation sync block (line 1125) only checks if conversation: and ignores store entirely.

Reproduction

Step 1: Create a conversation

$ curl -s http://localhost:8321/v1/conversations \
  -H "Content-Type: application/json" -d '{}'

{
    "id": "conv_cfe5bb383e510ebcd729360c012bd07c3c8e5643c38789da",
    "object": "conversation",
    "created_at": 1774477837,
    "metadata": null,
    "items": null
}

Step 2: Create a response with store=false on this conversation

$ curl -s http://localhost:8321/v1/responses \
  -H "Content-Type: application/json" \
  -d '{"model":"openai/gpt-4o-mini","input":"What is 2+2?","store":false,"conversation":"conv_cfe5bb383e510ebcd729360c012bd07c3c8e5643c38789da"}'

{
    "id": "resp_0217f2ac-425e-4d8a-8779-aceb7c6700b3",
    "object": "response",
    "status": "completed",
    "output": [
        {
            "content": [{"text": "2 + 2 equals 4.", "type": "output_text"}],
            "role": "assistant",
            "type": "message",
            "id": "msg_14412d01-45fe-445a-9603-e254b293c372",
            "status": "completed"
        }
    ],
    "store": false
}

Step 3: Verify the response object was NOT stored (correct)

$ curl -s http://localhost:8321/v1/responses/resp_0217f2ac-425e-4d8a-8779-aceb7c6700b3

{
    "detail": "Response 'resp_0217f2ac-425e-4d8a-8779-aceb7c6700b3' not found. Use 'client.responses.list()' to list available Responses."
}

The response object was correctly not persisted — store=false was honored here.

Step 4: Check conversation items (BUG)

$ curl -s http://localhost:8321/v1/conversations/conv_cfe5bb383e510ebcd729360c012bd07c3c8e5643c38789da/items

{
    "object": "list",
    "data": [
        {
            "content": [{"text": "2 + 2 equals 4.", "type": "output_text"}],
            "role": "assistant",
            "type": "message",
            "id": "msg_14412d01-45fe-445a-9603-e254b293c372",
            "status": "completed"
        },
        {
            "content": [{"text": "What is 2+2?", "type": "input_text"}],
            "role": "user",
            "type": "message",
            "id": "msg_5863422ba1218f2c9fa62c6ed47114c9c8522072037854a6",
            "status": null
        }
    ],
    "first_id": "msg_14412d01-45fe-445a-9603-e254b293c372",
    "last_id": "msg_5863422ba1218f2c9fa62c6ed47114c9c8522072037854a6",
    "has_more": false
}

Both the user input and assistant output were persisted to the conversation despite store=false.

Impact

• Privacy: A user explicitly opting out of storage (store=false) still has their conversation content written to the database. This is especially problematic for sensitive or PII-containing queries.
• Semantic violation: The store parameter's contract is broken — partial persistence is worse than no persistence because the user believes their data was not stored.
• Storage leak: Ephemeral conversations accumulate in the conversation_messages table with no cleanup path since the user never intended them to be stored.

Fix

Add and store to the condition on line 1120:

if (
    stream_chunk.type in {"response.completed", "response.incomplete"}
    and final_response
    and failed_response is None
    and store  # <-- gate conversation sync on store flag
):