ChatGPT desktop upgrade leaves Remote Control daemon on an older Codex runtime, hiding new models in Remote projects
Summary
After upgrading the macOS Codex desktop app to the unified ChatGPT desktop app, the foreground app updated successfully but the durable Remote Control app-server daemon remained on an older standalone Codex runtime.
As a result, local Tasks/projects can select GPT-5.6, while projects opened on the same machine through Remote cannot: the remote host's model catalog does not include any GPT-5.6 variants.
Expected behavior
Updating and relaunching the ChatGPT desktop app should update or migrate the managed Remote Control daemon as well. If that cannot happen automatically, the product should clearly identify the host-runtime mismatch and offer a supported repair path.
The remote model picker should expose models available to the same signed-in account when the host is otherwise healthy.
Actual behavior
The updated app and the persistent daemon use different runtimes:
- ChatGPT desktop app:
26.707.30751(build5018) - Embedded foreground CLI/app-server:
0.144.0-alpha.4 - Managed Remote Control daemon:
0.142.2
Sanitized output from the current embedded CLI:
{
"status": "running",
"backend": "pid",
"managedCodexPath": "$HOME/.codex/packages/standalone/current/codex",
"managedCodexVersion": "0.142.2",
"cliVersion": "0.144.0-alpha.4",
"appServerVersion": "0.142.2"
}
The standalone current target had only the 0.142.2 release installed. The Remote Control daemon persisted from before the desktop-app update; restarting the visible app did not replace it.
The app's Remote model picker performs host-specific model/list requests. Both requests succeeded, but returned different catalogs:
| Host/runtime | Returned models |
| --- | --- |
| Updated local app-server (0.144) | GPT-5.5, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, GPT-5.4, GPT-5.4 Mini, and others |
| Remote Control daemon (0.142.2) | GPT-5.5, GPT-5.4, GPT-5.4 Mini, GPT-5.3 Codex Spark, Auto Review |
This is not a failed model-list request or an account-wide availability issue: the same signed-in account can use GPT-5.6 locally, while the older Remote daemon successfully returns its own older model catalog.
Observed scenario
- Enable durable Remote Control on a macOS host running an older Codex app-server daemon.
- Update the desktop app to the July 9 unified ChatGPT desktop release and relaunch it.
- From another signed-in ChatGPT desktop app, open a project hosted on that machine.
- Compare the local model picker with the Remote-project model picker.
The local picker contains GPT-5.6 variants; the Remote-project picker does not.
Requested improvements
- Migrate/update the standalone managed daemon when the desktop app updates.
- Detect and surface foreground-vs-daemon version skew in Remote settings.
- Provide a supported in-product repair action that updates/restarts the managed daemon without requiring users to inspect internal state.
- Ensure a Remote model picker cannot silently remain on a stale per-host catalog after a desktop-app migration.
No usernames, hostnames, account/workspace IDs, IPs, tokens, or raw logs are included.
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗