[macOS][Quick Chat] Hotkey popup hides after Send; response appears only in main window

What version of the Codex / unified ChatGPT app are you using?

26.818.41509 (build 6962)

What subscription do you have?

Not provided

What platform is your computer?

macOS 26.5.2 (build 25F84), Apple Silicon (arm64)

What issue are you seeing?

In the current unified ChatGPT desktop app, the global hotkey popup opens and accepts the first message, but the popup hides/disappears before the assistant response is displayed. The response completes successfully and is visible only after opening the main ChatGPT window.

Because the popup is gone, the conversation cannot be continued in the popup. This began immediately after an app upgrade and has now been reproduced on two separate Apple Silicon Mac laptops.

Restarting the app and resetting the popup-related local state did not change the behavior.

Steps to reproduce

1. Open the current unified ChatGPT desktop app on macOS.
2. Invoke the configured global hotkey window shortcut.
3. Enter any message in the popup and press Send.
4. Observe that the popup initially accepts the request.
5. Before the assistant reply is rendered, observe that the popup hides/disappears.
6. Open the main ChatGPT window.
7. Observe that the assistant reply completed and appears in the main window instead of the popup.

Expected behavior

The hotkey popup should remain visible while the response is generated and display the completed reply. It should preserve the same conversation so follow-up messages can be sent from the popup.

Actual behavior

The popup becomes hidden before the response is displayed. The response is generated normally but appears only in the main window, leaving the popup workflow unusable for multi-turn conversation.

Diagnostic evidence

The affected app log shows that:

• the hotkey conversation is created and turn/start succeeds;
• the hotkey thread window is visible and focused at submission;
• the response completes successfully;
• by completion, the hotkey thread window is reported as visible=false and focused=false.

The generated assistant response is also present in the corresponding local rollout, confirming that generation did not fail. No conversation text, account information, or raw logs are attached here.

Regression confirmation

The same failure occurred on a second laptop only after upgrading the app. This makes local preference corruption or a machine-specific configuration unlikely.

Related issues checked

• #33753 concerns inability to send follow-up messages in the main Chat interface.
• #37572 concerns sending while a response is still generating.
• #38182 concerns Quick Chat becoming a white screen after large input.

Those reports are related to Chat/Quick Chat behavior but do not describe this exact popup lifecycle regression.