You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
autocompact fails for me too: 019db1fd-7d09-7370-bea2-7e518769131e and shows ■ Error running remote compact task: stream disconnected before completion: error sending request for url (https://api.openai.com/v1/responses/compact)
no fix? how to fix?
this is making the 5.5 update completely unuasble, especially given that 5.5 only is 258k tokens right now even with settings to 400k. autocompaction gets stuck -> copy thread id for a new session to resume, one turn it its autocompaction again. completely unusable right now. @etraut-openai pls fix thnx!
This still happens during real coding work. The failure mode is not just slow compaction; the thread gets stuck in the autocompaction transition and needs manual steering / additional prompts to become usable again.
Observed today:
Thread was mid-task on a multi-file code/debugging fix.
Autocompaction interrupted the task.
The Codex app thread became stuck enough that Alan had to continue through the CLI instead of the Codex app.
After prompting again from CLI, Codex resumed with a compacted summary, but the active control flow had been disrupted.
The summary preserved facts like changed files, tests run, root cause, and next step, but the thread no longer continued cleanly from the pending action.
User had to ask meta questions to re-establish whether the session was unstuck.
So there seem to be two related failure surfaces:
The remote compact request can fail/stall with stream disconnected before completion.
Even when the session becomes promptable again, the transition can leave the agent operationally “stuck” until the user manually reorients it, and in this case the practical workaround was switching out of the Codex app to the CLI.
Expected behavior:
If remote compaction fails, return control with a clear recoverable error and do not leave the thread in a half-compacted state.
If compaction succeeds after retry/recovery, preserve the active task continuation explicitly: latest user intent, immediate next action, pending blocker/tool step, and whether injected workspace/session instructions are informational or should supersede the active task.
A useful diagnostic would be exposing whether the failure happened in:
To be clear, after testing, this is not autocompaction failing, this is compaction failing in general. I get this error when invoking compaction manually (now that it's available) as well!
Also having same issue, compaction fails in codex app, but not in cli, not sure why @etraut-openai keeps removing the app label from these issues, where clearly it is a codex app issue.
Thanks for reporting this problem. Until recently, Codex used a separate endpoint and server-side logic for compaction. We recently switched to a more robust approach where the compaction logic is moved locally using the same endpoint as normal turns. This eliminates "remote compaction" errors and increases the stability and reliability of compaction operations. If you see further problems with compaction, please use /feedback and open a new bug report.
Activity
github-actions commentedon Apr 12, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
lasertiger commentedon Apr 22, 2026
autocompact fails for me too: 019db1fd-7d09-7370-bea2-7e518769131e and shows ■ Error running remote compact task: stream disconnected before completion: error sending request for url (https://api.openai.com/v1/responses/compact)
no fix? how to fix?
infiloop2 commentedon Apr 22, 2026
+1, almost consistently failing for me starting today
alanxchen85 commentedon Apr 24, 2026
this is making the 5.5 update completely unuasble, especially given that 5.5 only is 258k tokens right now even with settings to 400k. autocompaction gets stuck -> copy thread id for a new session to resume, one turn it its autocompaction again. completely unusable right now. @etraut-openai pls fix thnx!
alanxchen85 commentedon Apr 24, 2026
Additional live occurrence on 2026-04-24:
This still happens during real coding work. The failure mode is not just slow compaction; the thread gets stuck in the autocompaction transition and needs manual steering / additional prompts to become usable again.
Observed today:
So there seem to be two related failure surfaces:
stream disconnected before completion.Expected behavior:
A useful diagnostic would be exposing whether the failure happened in:
/backend-api/codex/responses/compactalanxchen85 commentedon Apr 24, 2026
To be clear, after testing, this is not autocompaction failing, this is compaction failing in general. I get this error when invoking compaction manually (now that it's available) as well!
[-]Autocompaction fails[/-][+]Compaction/Autocompaction fails[/+]hchafiatheal commentedon Apr 24, 2026
Also having same issue, compaction fails in codex app, but not in cli, not sure why @etraut-openai keeps removing the app label from these issues, where clearly it is a codex app issue.
lasertiger commentedon Apr 24, 2026
psaul20 commentedon May 3, 2026
Getting the same issue - 019dea5d-63bc-7a52-acf0-48000ae08f95
etraut-openai commentedon Jun 29, 2026
Thanks for reporting this problem. Until recently, Codex used a separate endpoint and server-side logic for compaction. We recently switched to a more robust approach where the compaction logic is moved locally using the same endpoint as normal turns. This eliminates "remote compaction" errors and increases the stability and reliability of compaction operations. If you see further problems with compaction, please use
/feedbackand open a new bug report.