Repository navigation
Windows Desktop: first prompt succeeds, second message cannot be sent; reinstall only fixes it temporarily #45797
Description
Summary
I am seeing a recurring Codex Desktop issue on Windows where a conversation works for exactly one turn, but after the first response completes I can no longer send another message in the same conversation.
This has now happened to me multiple times. Reinstalling Codex Desktop fixed it temporarily once, but the same problem returned again on September 15, 2026.
Environment
- OS: Windows 10 (64-bit)
- App: Codex Desktop for Windows
- Date reproduced again: September 15, 2026
- The issue affects existing conversations after the first successful turn
Steps to reproduce
- Open Codex Desktop on Windows.
- Open a project/conversation (or start a new one).
- Send the first prompt.
- Wait for Codex to finish responding normally.
- Type a second message in the same conversation and try to send it.
Actual behavior
The first prompt and response work normally, but the second message cannot be sent / does not start a new turn. The conversation effectively becomes unusable after one successful exchange.
In my case, reinstalling the Windows desktop app previously restored normal behavior, but only temporarily. The bug has now returned, which makes it look more like a recurring Desktop client / renderer state issue than corrupted project or conversation data.
Expected behavior
After the first response completes, I should be able to continue sending follow-up messages in the same conversation normally.
What I have observed
- First turn: works normally.
- Second/follow-up message: cannot be sent.
- Reinstall: previously fixed the problem temporarily.
- The problem later returned without intentionally modifying my Codex conversation data.
Related reports
This appears very similar to several recent Windows Desktop reports, including:
- [Windows App 26.820.7780.0] Composer stays disabled after the first completed turn in a new task #40872
- Windows Codex desktop: Send button spins forever and prompts never submit #40968
- Windows desktop app: second/follow-up message hangs indefinitely in Work and Codex after Sep 5 update #43366
- Windows Desktop 26.903.61454: follow-up messages cannot be sent after the first completed turn #44102
- [Windows Desktop] Second message stalls after first turn completes #44295
- Windows desktop: existing-chat sends indefinitely blocked by loading-local-config / pending codex-home; main-window reload recovers, normal restart can recur #44342
- Codex desktop (Windows): follow-up messages in existing threads never send #45069
- Still reproducible on Windows Desktop 26.908.40834 — first turn completes, but follow-up message does not create a new turn/start #45315
I am opening another report to confirm that this is still reproducible on a separate Windows installation as of September 15, 2026.
I can provide additional logs or run specific diagnostics if that would help identify whether this is the same loading-local-config / renderer state problem described in some of the related issues.
Activity
github-actions commentedon Sep 15, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
BilibiliSZdaomei commentedon Sep 15, 2026
Additional reproduction and workaround confirmation on Windows package
OpenAI.Codex 26.908.9136.0(September 15, 2026).The same user-visible failure is still reproducible on this newer build:
I then tested the renderer-restart workaround without reinstalling Codex, clearing caches, resetting
.codex, or deleting any conversation data.Using PowerShell, I listed the running
ChatGPT.exerenderer processes (--type=renderer). The renderer corresponding to the main Codex window was the clearly dominant renderer by working set / CPU in this session. I terminated only that renderer process withStop-Process -Id <PID> -Forcewhile the previous model turn was already complete and no task was actively running.Observed result:
.codexreset, login reset, or conversation migration was needed.This strongly matches the renderer/local-initialization stuck-state behavior described in #44342 and #45069, and confirms that the issue is still present on
26.908.9136.0.Important limitation: identifying the main renderer by highest memory/CPU is only what worked in this specific session and should not be treated as a generally safe identification method across all builds/windows. A supported in-app renderer reload / local-state retry action would still be preferable.
shanshan1231 commentedon Sep 16, 2026
I finally fixed this issue on my machine.
My symptom was exactly the same: the first message in a new Codex conversation worked, but the second message would stay stuck indefinitely.
In my case, the root cause turned out to be stale authentication. My Google account password had been changed, but Codex Desktop still appeared to be logged in and did not ask me to authenticate again.
After changing/resetting the password and then fully signing out of Codex and signing back in with ChatGPT/Google, multi-turn conversations immediately started working again.
So if you have this issue, I recommend trying a full logout and re-login, especially if your Google/OpenAI account credentials were changed recently. The app may appear authenticated even though its cached authentication state is no longer valid.
o-khard commentedon Sep 17, 2026
I tried relogging in - issue persisted. So relogging isn't a universal solution.
Nokanokanoka commentedon Sep 18, 2026
Still reproducible on a newer Windows Desktop build.
Environment:
Reproduction:
Also affects existing chats.
Troubleshooting already tried:
codex doctor --all: 21 ok, 1 idle, 2 notes, 2 warn, 0 failRelevant desktop log behavior:
turn-completenormally and the desktop notification is emitted.Received broadcast but no handler is configured method=thread-stream-following-changedReceived broadcast but no handler is configured method=thread-read-state-changedThis looks like a Windows Desktop renderer/thread-state synchronization issue: backend turn completion succeeds, but the composer does not become able to start the next turn.
I can provide additional sanitized logs if useful.
GoMaaars commentedon Sep 20, 2026
Same root cause as #44342: the renderer's first batch of
vscode/*local requests (codex-homeincluded) gets its responses dropped during startup and nothing retries them. Re-issuing those queries after startup recovers in milliseconds without reloading the window. Details and a small MIT workaround script (manual + optional auto-run on app start) here: https://github.com/GoMaaars/codex-desktop-send-button-fix — happy to add more traces if useful.shanshan1231 commentedon Sep 21, 2026
太专业了!这个问题在我电脑上出现了好久,我之前的解决办法就是重新登录codex和网页端的账号,就能恢复正常(后续也会复现问题,但是按照这种重新登录的办法也能解决),但是我一直没有溯源问题的根本,我大概清楚问题是出现在哪里,今天终于一探究竟了。不愧是MIT
Sebastian-Habenicht commentedon Sep 21, 2026
Additional reproduction on Windows desktop:
Windows 11 Pro x64, OS build 26200.9457
Codex Desktop: 26.915.31945
This matches the first-turn-only behavior reported here and adds the brief spinner and unaffected regular-chat/voice observations.
EDIT:
val235 commentedon Sep 24, 2026
Reproducing this across my machines as of September 24, 2026. On one affected machine: Windows 11 Pro x64, OS build 26200, Codex Desktop 26.917.9434.0.
A new conversation may accept its first prompt, but follow-ups do not send; existing conversations are also blocked. Remote can still continue the conversation while the Windows desktop composer cannot. Restarting a
ChatGPT.exerenderer, which previously helped, no longer restores sending for me.On the affected machine,
codex doctor --allreports healthy local databases, successful app-server initialization, and a working model connection. I have not established where the desktop send fails. Please investigate a durable fix and provide a supported recovery action that preserves drafts and conversations.