When will the ChatGPT app be able to connect directly to Codex running on a Linux machine?

[uuuuu-diwu]

When will the ChatGPT app be able to connect directly to Codex running on a Linux machine?

My daily workflow is that my Mac connects to our Linux devbox. My Mac is not suitable for running services or doing a lot of messy/heavy work, so all my code and workspaces are on the Linux machine.

But right now, the ChatGPT app still cannot directly connect to Codex on Linux. It used to work in version 137, but for some reason it has stopped working properly over the past few days. I previously tried asking Codex to read the Codex source code, and it discovered that remote control could be started directly on Linux and then viewed directly from the phone. But now it no longer works at all.

This makes it very hard for me to work anytime, anywhere. I carry my Mac around a lot, and obviously I often close the lid. But my Linux machine runs 24/7.

[EricSpencer00]

I usually have a persistent Codex App running which I think is necessary to keep a remote session enabled for the phone.

Is it possible to compile the codex app on the Linux machine and then sign in? Another option is a dedicated Mac (I use mac mini) as a global codex router. I have my Codex there and whenever I need to do work on another machine, I setup SSH known hosts to the device and then let my Codex router speak to those machines via SSH. If Linux doesn't support the app, then that is what I would do.

Finally you could try SSH directly from something like TailScale into the machine and just run multiple Codex sessions without a fancy app UI.

[hoodrichpirobo]

As far as the current documented setup goes, there does not seem to be a public ETA for “ChatGPT mobile/app connects directly to a Linux Codex host” as a first-class remote host.

The supported model today is:

ChatGPT mobile / another Codex app
        -> connected Codex App host on macOS or Windows
        -> optional SSH project on Linux

So for your workflow, the intended setup is not to expose a Linux codex app-server directly to ChatGPT. It is to run the Codex App on a Mac or Windows machine, then connect that app to your Linux devbox over SSH.

The practical setup is:

1. On the Mac, add the Linux devbox to ~/.ssh/config:

   Host devbox
     HostName your-linux-host
     User your-user
     IdentityFile ~/.ssh/id_ed25519

2. Confirm this works from the Mac:

   ssh devbox

3. On the Linux devbox, install/authenticate Codex and make sure `codex` is on PATH for the SSH login shell.

4. In the Codex App on the Mac:
   Settings -> Connections -> add/enable the SSH host -> choose the remote project folder.

5. For phone access:
   keep the Codex App host awake, online, signed in, and connected through the app’s mobile/remote setup.

The important limitation is that the phone is still connecting to the Codex App host, not directly to Linux. If your Mac sleeps, closes the lid, loses network, or the Codex App is closed, remote access can stop even if the Linux devbox is still running.

If you need 24/7 access to a Linux workspace, the most reliable current workaround is an always-on Mac/Windows machine acting as the Codex App host/router, with the actual project running on Linux over SSH. A Mac mini or small Windows box works better than a laptop for that. The other lower-level option is SSH/Tailscale into the Linux box and run Codex CLI sessions there, but that will not give you the same ChatGPT mobile/Codex App remote-control experience.

I would avoid depending on a directly exposed Linux codex app-server for this. The app-server WebSocket transport exists for integrations, but it is documented as experimental/unsupported, and direct remote exposure is specifically not the recommended path.