New ChatGPT desktop app stuck on local Ollama model — no way to switch back to cloud

I’d like to report an issue with the new ChatGPT desktop app for macOS. The app became locked into using a local Ollama model, and I could not find any way within the app to return to cloud mode.

■ Environment

  • Device: MacBook Pro 14-inch (M1 Pro, 16GB unified memory)
  • macOS version: [Apple menu → About This Mac]
  • ChatGPT app version: [check the About section in app settings]
  • When it occurred: early September 2026 (KST)

■ Steps to reproduce

  1. I installed Ollama to try out local LLMs and downloaded some models (qwen3.5:9b, gemma4).
  2. When I next opened the ChatGPT desktop app, Work mode was already using the local Ollama model without my having configured anything.
  3. I tried to switch back to cloud mode but could not find a way to do so inside the app.

■ What actually happens

  • The model picker in the composer shows only the local model (qwen3.5:9b). No GPT cloud models appear in the list.
  • In the “Where should this chat run?” popup, only “On my computer” is selectable. “In the cloud” is greyed out.
  • The top-left dropdown offers only ChatGPT Work and Codex. There is no entry for Chat.
  • The bottom-left slot, where account information normally appears, shows only “Ollama”.
  • Clicking Account in settings opens a web browser, and there is no flow that returns to the app afterward.
  • Fully quitting Ollama and restarting the ChatGPT app makes no difference.
  • The Option(⌥) + Space shortcut does nothing.

■ What I expected
Even after using a local model once, it should be possible to return to cloud models and Chat from within the app. At minimum, there should be a visible way to disconnect the Ollama integration or sign back into my account. As things stand, deleting and reinstalling the app seems to be the only option I have.

■ Note
ChatGPT Classic and the web version (chatgpt.com) both work normally with the same account. The problem occurs only in the new desktop app.

For someone encountering local AI for the first time, it is genuinely hard to tell why the app behaves this way, and easy to assume something is wrong with the account. It would help a lot if the point where the local integration is turned on and off were made clearer.

This topic was automatically closed after 24 hours. New replies are no longer allowed.