I have recently installed codex app. Authenticated with my account which has subscription to chatgpt. When I try to start a conversation, I see this error - Error starting conversation.
How should I solve this problem ?
cosburn
February 4, 2026, 11:20pm
2
Same issue. and Codex CLI and I have tried everything, even uninstalled and reinstalling. Was working like a champ just yesterday.
Carl
Hey @Sumit_Gupta and @cosburn, this looks very similar to the incident we had on Feb 3, 2026, where some users were seeing issues loading or starting conversations.
The āError starting conversationā message matches what was happening then, and the issue was resolved shortly after. You can see the details on the status page here .
Things should be working normally now, but if youāre still experiencing the issue, try logging in via an incognito window to confirm the fix is correctly reflecting on your account, then start a fresh conversation.
āThe issue is still occurring, and existing conversations are showing a āFailed to resume taskā error.ā
I have the same issue. Newest version of Codex on MacOS
Same issue today on the latest version of Codes on MacOS
Having this issue on Mac app too. Getting this popup every time I enter a new query
MedVox
March 12, 2026, 1:49pm
8
Same problem. āFailed to resume taskā after update to newest version.
EricGT
March 12, 2026, 2:01pm
9
For those reporting the problem here.
You are more than welcome to report such problems here.
For persistent issues such as noted in this topic also consider the official place to report such at the GitHub OpenAI Codex issue tab.
Search for your issue first and if you find it give the first post a thumbs up to show that you also have the problem.
If your issue is not a duplicate then then report it as a new issue.
I do not have a Mac so can not test this.
This issues seems to be the one noted here.
opened 09:04PM - 11 Mar 26 UTC
bug
app
regression
### What version of the Codex App are you using (From āAbout Codexā dialog)?
26ā¦ .305.950 (863)
### What subscription do you have?
ChatGPT Enterprise
### What platform is your computer?
Darwin 24.6.0 arm64 arm
### What issue are you seeing?
Codex Desktop App Version 26.305.950 (863) on macOS 15.7.4 fails to start or load any session with the error:
`failed to resume task`
This affects:
ā¢ All existing sessions
ā¢ Any brand-new session created after reinstall
The error appears immediately when attempting to start or load a session.
ChatGPT (web) and Codex online function normally under the same account.
No meaningful local log entries are generated at the time of failure.
### What steps can reproduce the bug?
1. Install Codex Desktop App 26.305.950 (863) on macOS 15.7.4.
2. Sign in using any OAuth provider (Microsoft, Google, Apple all tested).
3. Launch the app.
4. Attempt to:
- Open an existing session, or
- Create a new session.
5. The app immediately returns:
`failed to resume task`
This occurs even after:
- Full uninstall
- Manual removal of:
```
~/Library/Application Support/Codex
~/Library/Caches/Codex
~/Library/Preferences/com.openai.codex.plist
~/Library/Saved Application State/com.openai.codex.savedState
```
- Reboot
- Fresh reinstall
- Full OAuth reauthentication
After reinstall, previous sessions repopulate from the server immediately.
The failure appears to occur during backend session bootstrap, before local execution begins.
Session ID not available because new sessions fail before creation completes.
### What is the expected behavior?
- Existing sessions should load normally.
- New sessions should create a fresh task/thread.
- If resume metadata is invalid, the app should fall back to creating a new task instead of failing entirely.
### Additional information
- There was a recently reported Codex service incident marked as resolved.
- The issue began during or after that incident.
- ChatGPT (web) and Codex online are fully functional.
- The issue appears isolated to the Codex Desktop App namespace.
- Old sessions sync from server even after a full local wipe, suggesting server-side session metadata persists.
- The failure happens immediately (no prolonged spinner).
which has been noticed by OpenAI staff
