The Codex Desktop app on Windows breaks with an error that isn't a logout or an expired token: the account screen still shows your email, but usage, plugins, and task history disappear, all carrying the same status code in the network tab. It's error 432, "Workspace routing is unavailable", and it only shows up in the graphical app. The terminal, on the same machine, with the same account, keeps working.
What does the 432 "Workspace routing is unavailable" error mean?
It means the Codex backend can't resolve workspace routing for your account inside the Desktop app, even with a valid login. The message in the app logs is direct:
errorMessage="Workspace routing is unavailable"
status=432
routePattern=/settings/userThe same error repeats across several workspace-dependent routes: /wham/usage, /wham/tasks/list, /accounts/verified_access, /subscriptions/credits/discount-offer. None of them are real authentication: the app-server confirms login and account reads without error (account/read, getAuthStatus). Only the account/workspace layer that Desktop uses to fill in the profile, usage, and plugins screens is the part that breaks.
Why does only Desktop break while the terminal keeps working?
Because they're two different clients reading the same backend. The Codex CLI talks directly to the task API and doesn't go through the account/workspace routing layer that only the Desktop app uses to build profile, usage, and plugins screens. Across at least two issues opened on the official repository (#46382 and #46786, Sep 18 and 20, 2026), multiple users confirmed the same pattern: Desktop shows the 432, but codex in the terminal, on the same machine and account, keeps authenticating and running tasks normally.
| Where | With the bug active |
|---|---|
Login and account reads (account/read, getAuthStatus) | Works on both clients |
Running tasks with codex in the terminal | Works normally |
| Profile, usage, and plugins screen on Desktop | Broken, 432 error |
| Task history sync on Desktop | Broken, 432 error |
How to fix it, in the order that worked for other users
- Check the
CODEX_CLI_PATHvariable. If it's set and pointing to the wrong binary, it makes Desktop load an incompatible runtime and triggers the 432. On Windows, check and remove it via PowerShell:
Restart Desktop after removing it.[Environment]::GetEnvironmentVariable("CODEX_CLI_PATH", "User") [Environment]::SetEnvironmentVariable("CODEX_CLI_PATH", $null, "User") - No variable set? Update the app-server. The regression showed up in the 0.155.x range. Users confirmed fixing it both by rolling back to the previous version and by moving to a newer build:
Check the installed version in "About Codex" in the app, or withnpm install -g @openai/codex@latestcodex --versionin the terminal. - Meanwhile, work from the terminal.
codexkeeps authenticating and running tasks normally even with Desktop broken, because it doesn't depend on the same account/workspace routing layer.
Has OpenAI already fixed this?
Partially. The original issue (#46382) was closed as resolved on September 20, 2026, but later comments, including from someone who reproduced the problem in a controlled way by swapping only the app-server version on the same Desktop and account (0.155.1 fails, 0.154.0 works), asked for it to be reopened. If you're still on an affected build, the three steps above remain the fastest path, without waiting for another release.
If your workflow depends on a heavy Desktop app, it's worth considering a more direct path: the Verboo Code runs entirely from the terminal, with no separate graphical app layer that can break while the engine keeps running. To try it: npm install -g @verboo/code and then verboo /login. Learn more at verboo.ai.
