[Windows Desktop 26.901.6511.0] Messages stop receiving replies without reconnect status; restarting Codex is required
Nobody has claimed this yet.
- Dominant language
- Rust
- Stars
- 125k
- Forks
- 19.4k
- PR merge metrics
- PR metrics pending
Description
What version of the Codex App are you using?
OpenAI.Codex Windows package 26.901.6511.0 (x64), verified locally on 2026-09-08.
What subscription do you have?
ChatGPT Pro (verified at submission time).
What platform is your computer?
Windows x64, version 24H2, OS build 26100.9168.
Model/settings at the time: GPT-6 Astra, highest reasoning setting, Full Access.
What issue are you seeing?
During normal use, Codex Desktop enters a state where sending a message no longer produces an assistant reply. Sometimes the message cannot be submitted successfully; sometimes it appears in the conversation but gets no response. The UI does not make it clear whether the message was delivered to the assistant.
I do not see an explicit reconnecting, disconnected, or offline status. My internet connection is working according to my observation. The only recovery that has worked for me is fully quitting and restarting Codex; after restarting, normal messaging resumes.
Two failure modes have been observed in separate conversations:
- Repeated queue-expiration errors: A follow-up message with two image attachments is attempted twice. Both attempts show the exact error:
The draft and attachments are still visible in the composer. There is no visible reconnect or recovery status explaining how to resume.App server request expired while queued - Silent stall / indefinite thinking: After earlier assistant progress and tool/file updates, short follow-up messages asking whether the assistant is stuck receive no visible reply. The UI remains at
正在思考(“Thinking”), with the stop button visible and no explicit network/reconnect error.
The queue-expiration error is real; the problem is not that every failure is completely silent. The other observed state provides no actionable explanation, and neither state recovers for me without restarting the app.
I cannot determine from the UI whether a stalled message never leaves the client, is stuck in the app-server queue, or fails later in processing. The underlying cause is not yet confirmed.
What steps can reproduce the bug?
The issue is intermittent; I do not yet have a deterministic minimal trigger. The observed sequence is:
- Open Codex Desktop on Windows and continue an existing task that was previously producing responses and tool/file updates.
- Send a follow-up message during normal work. One captured case includes image attachments.
- When progress stops, send another short follow-up asking whether the assistant is stuck.
- Observe either repeated
App server request expired while queuederrors or a persistent “Thinking” state without a reply. - There is no useful reconnect status or successful automatic recovery.
- Fully quit and restart Codex.
- Messaging works again after the restart.
The problem has occurred in separate conversations. It is not yet established whether they stalled at exactly the same time or whether attachments are required to trigger it.
What is the expected behavior?
- A submitted message should receive a response, or the app should show a clear delivery/failure state with an actionable retry.
- If the app-server or connection is stalled, Codex should detect it and display an accurate status.
- A timeout should lead to a recoverable state while preserving the message and attachments.
- Normal messaging should recover without forcing the user to restart the whole desktop app.
Additional information
Impact: Work is interrupted, follow-up instructions cannot be delivered reliably, and repeated submissions create uncertainty about which messages were received.
Workaround: Fully quit and restart Codex. This restores messaging in my observed incidents.
Related reports: #43525 and #37401 describe similar Windows app-server queue-expiration/restart symptoms. I have not verified that the underlying cause is the same. This report specifically documents the two UI states and the lack of useful reconnect/recovery feedback.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
No source file, test, or entry point is named. Start by reproducing the observed queue-expiration and persistent “Thinking” states on Windows, comparing behavior before and after restarting Codex. Done means messages receive replies or show an actionable delivery failure, with recovery that preserves drafts and attachments without restarting the app.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- rust
- Domain
- desktop, networking
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Needs clarification
- Newbie friendliness
- 35/100