Repository navigation
Windows desktop: one conversation repeatedly fails remote compaction, even after restart and a one-word prompt #53779
Description
What issue are you seeing?
One of my existing conversations is stuck. Every recent attempt fails with:
Error running remote compact task: stream disconnected before completion:
An error occurred while processing your request.
The interface also shows Compaction interrupted and Reconnecting 2/2. Restarting the Windows app did not recover the conversation. Even sending just "Hi" fails before the assistant replies.
I lost my open research tabs when I restarted. I need a way to recover and continue the existing work without losing the conversation history.
Environment
- Product: Windows desktop app, local Codex conversation.
- Windows version reported by the runtime: Microsoft Windows NT 10.0.26300.0; 64-bit.
- Local app-server version reported in the desktop startup log:
0.162.0-alpha.17.2. This is the app-server version, not a verified desktop app version. - Desktop app version and subscription: not independently verified.
- Observed: October 13, 2026, America/Phoenix (UTC-07:00).
What steps can reproduce the bug?
This describes the observed failure in an existing conversation; I do not have a minimal fresh-conversation reproduction.
- Open the affected existing conversation.
- Submit a message.
- Observe the remote compaction error and reconnect attempts.
- Restart the desktop app and return to the same conversation.
- Submit "Hi". It fails with the same remote compaction error.
Confirmed diagnostic evidence
The six most recent inspected turns in that conversation have a terminal failed status and the same remote compaction error. Their completion times were:
| Time on October 13, 2026 (America/Phoenix) | Result |
|---|---|
| 1:39:34 PM | Remote compaction failed |
| 1:39:57 PM | Remote compaction failed |
| 1:40:19 PM | Remote compaction failed |
| 1:41:41 PM | Remote compaction failed; the message reported restarting and losing tabs |
| 2:13:55 PM | Remote compaction failed for a plain "Hi" |
| 2:16:06 PM | Remote compaction failed for "hi" with an attached screenshot |
Those failures took approximately 9 to 24 seconds. Six is the count in the inspected recent-turn sample, not a claimed total for all failures.
Two request IDs in the supplied screenshot exactly match the saved failed-turn records. The conversation history remains readable through the desktop thread-reading tool, but the affected conversation cannot generate a new response. Another conversation on the same installation is responding; this is not a controlled network or model comparison.
What is the expected behavior?
The existing conversation should complete compaction and answer, or provide an actionable recovery option that preserves history and work. Restarting or submitting a short message should not leave the conversation permanently stuck at the same compaction step. Open research tabs should survive or be recoverable after a restart.
Request
Please investigate the repeated compaction failure and provide a supported way to recover the existing conversation. If there is a known fix, please identify the affected component and fixed version.
The underlying cause has not been established. This report does not assume that the Windows client, connection, or service is solely responsible.
Request IDs, the affected conversation ID, and the screenshot are retained for a private support follow-up. This public report contains no raw conversation transcript.
Possibly related: #53653 and #52919. Similar symptoms do not establish a shared cause.
Activity
github-actions commentedon Oct 13, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action