Windows Desktop: follow-up messages hang indefinitely after first turn; clean reinstall only fixes it temporarily
Nobody has claimed this yet.
- Dominant language
- Rust
- Stars
- 125k
- Forks
- 19.4k
- PR merge metrics
- PR metrics pending
Description
Summary
Codex Desktop on Windows frequently becomes unable to send follow-up messages.
The first message in a new conversation usually works. After the first turn completes, the next message stays in the composer with an infinite spinner and is never sent.
A completely clean reinstall temporarily fixes the problem, but the bug comes back again later.
This is a serious regression and, frankly, embarrassing for a production desktop client. Codex becomes practically unusable for normal multi-turn work.
Environment
- Windows 10 Pro 22H2
- Codex Desktop tested:
- 26.908.9136.0
- 26.903.9818.0
- Codex CLI: 0.154.0
- ChatGPT authentication
- No proxy
Reproduction
- Start Codex Desktop.
- Create a new conversation.
- Send a message.
- Wait for Codex to finish responding.
- Send a second message in the same conversation.
Actual behavior
The second message remains in the input box.
The Send button shows an infinite spinner.
No response is produced and the message is not submitted.
Creating a new conversation may temporarily work for one message, after which the same problem occurs again.
Expected behavior
Follow-up messages should create a new turn and be sent normally.
Diagnostics
codex doctor reports:
- authentication configured successfully
- WebSocket connected successfully (
HTTP 101 Switching Protocols) - provider endpoints reachable
- Desktop app-server initialized successfully
- state databases healthy
The first request successfully reaches the app-server:
method=turn/start
errorCode=null
However, when the UI hangs on the next message, no corresponding second turn/start is sent.
The Desktop log has also shown errors such as:
Received turn/started for unknown conversation
and:
Received turn/completed for unknown conversation
There are also IPC warnings such as:
Received broadcast but no handler is configured method=thread-stream-following-changed
Troubleshooting already attempted
- Restart Codex Desktop
- Create new conversations
- Upgrade Codex CLI
- Downgrade Desktop from 26.908.9136.0 to 26.903.9818.0
- Remove malformed rollout session
- Verify session/state database integrity
- Completely uninstall Codex Desktop
- Move/remove
.codexuser state - Remove local Codex application data
- Reinstall Codex Desktop from a clean state
- Re-authenticate
A clean reinstall initially made multi-turn conversations work again.
The problem later returned.
This strongly suggests that some Desktop renderer/conversation state becomes corrupted or stuck over time rather than this being a network or authentication failure.
Impact
This makes Codex Desktop effectively unusable for sustained coding work, because normal multi-turn conversations stop functioning and require repeated resets/reinstalls.
Please prioritize this Windows Desktop regression.
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
Start by tracing the Windows Desktop renderer's follow-up submission after the first turn, using the reported turn/start events and unknown-conversation logs to compare the first and second messages. Done means a second turn/start is sent for follow-up messages and the conversation completes without hanging or reporting unknown conversations.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- rust
- Domain
- desktop
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 52/100