Codex Windows app exits immediately when USERPROFILE contains Korean characters
What version of the Codex App are you using (From “About Codex” dialog)?
Windows app package: OpenAI.Codex_26.608.1337.0_x64__2p2nqsd0c76g0 CLI version: codex-cli 0.139.0
What subscription do you have?
ChatGPT Pro
What platform is your computer?
Microsoft Windows NT 10.0.26200.0 x64
What issue are you seeing?
On Windows, Codex desktop app version 26.608.1337.0 launches several Codex.exe processes, then exits within 2-3
seconds. Crashpad dumps show "__char_to_wide: Illegal byte sequence" while staging cua_node under a user profile path
that contains Korean/non-ASCII characters.
What steps can reproduce the bug?
- Use a Windows account whose user profile directory contains Korean/non-ASCII characters, for example:
C:\Users\<non-ASCII Korean username>\...
- Install or update to Codex Windows desktop app version:
OpenAI.Codex_26.608.1337.0_x64__2p2nqsd0c76g0
- Launch Codex from the Windows Start Menu, or run the app entrypoint:
C:\Program Files\WindowsApps\OpenAI.Codex_26.608.1337.0_x64__2p2nqsd0c76g0\app\Codex.exe
- Observe that several Codex.exe processes start briefly, then the desktop app exits within about 2-3 seconds.
- Crashpad dump strings repeatedly include:
__char_to_wide: Illegal byte sequence
C:\Users\?????????\AppData\Local\OpenAI\Codex\runtimes\cua_node\.staging-...
C:\Users\?????????\.codex\config.toml
USERPROFILE=C:\Users\??????
Expected:
The Codex desktop app should open normally even when the Windows user profile path contains non-ASCII characters.
Actual:
The app exits during startup. AppModel Runtime logs show the Desktop AppX container being created and then destroyed
shortly after. A new Crashpad dump is generated on each launch.
Troubleshooting attempted:
- Reset the Chromium/Electron profile by moving LocalCache\Roaming\Codex\web\Codex aside.
- Manually copied packaged resources\cua_node to AppData\Local\OpenAI\Codex\runtimes\cua_node.
- Launched with --disable-gpu and --disable-gpu-compositing.
- The app still exits within a few seconds.
No code snippet is required to reproduce this because it is a Windows desktop app startup/path encoding issue. Session
id, token limit usage, and context window usage do not apply.
What is the expected behavior?
The Codex Windows desktop app should launch successfully and show the main UI regardless of whether the Windows user
profile path contains Korean or other non-ASCII characters.
The app should correctly handle paths such as:
C:\Users\<non-ASCII Korean username>\AppData\Local\OpenAI\Codex\...
It should not fail during cua_node runtime staging or crash with "__char_to_wide: Illegal byte sequence".
Additional information
This appears to be a regression in the newer Windows desktop app build. The issue started after updating to
OpenAI.Codex_26.608.1337.0_x64__2p2nqsd0c76g0.
Local diagnostics suggest this is not a corrupted cache issue. Resetting the Chromium/Electron profile and manually
restoring the packaged cua_node runtime did not fix the crash.
The strongest signal from Crashpad dump strings is that the Korean/non-ASCII Windows username path is being corrupted
into question marks, followed by:
__char_to_wide: Illegal byte sequence
The crash appears related to staging or initializing:
C:\Users\<non-ASCII username>\AppData\Local\OpenAI\Codex\runtimes\cua_node\.staging-...
A likely workaround is to run Codex under a Windows account with an ASCII-only user profile path, but the desktop app
should ideally support non-ASCII Windows usernames.
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗