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
{{ message }}
Repository navigation
Desktop App on Windows 26.623.101652: Unable to send message (Loading...), previous build failed with "no rollout found for thread id" #31004
The Desktop application creates a thread ID successfully.
The backend then responds:
no rollout found for thread id ...
This suggests thread creation succeeds but the rollout/task initialization fails.
The issue appears isolated to the Desktop application because both the web interface and CLI continue functioning normally.
Regression / version history
This appears to be a continuing Desktop App regression across recent builds.
The version before 26.623.81905 worked correctly.
After updating to 26.623.81905, the Desktop App started failing when creating new chats/tasks.
After updating again to 26.623.101652, the behavior changed: existing conversations can open, but sending messages now fails with Unable to send message / Loading....
So this is not a regression introduced only in 26.623.101652. The issue started in 26.623.81905 and persists, with changed symptoms, in 26.623.101652.
Adding another Windows Desktop data point that may help separate this from pure network/auth failures.
Environment:
Windows 11
Codex Desktop MSIX package currently installed: OpenAI.Codex_26.623.13972.0_x64__2p2nqsd0c76g0
WSL mode enabled for agent/runtime
WSL runtime itself is healthy: integrated terminal reports Linux/WSL and $HOME=/home/yury
Current WSL Codex CLI override was verified and points to an existing binary: codex-cli 0.142.5
Windows user CODEX_HOME is unset
bundled dependencies runtime validates successfully via setup.ps1
User-visible symptom:
After a Codex Desktop app restart, an existing chat can open but sending a simple message in that chat shows the modal:
Unable to send message
Loading...
New observation / workaround:
If I switch to a different chat after the app restart and send a simple test message there first, that message sends successfully.
After that, returning to the originally failing chat makes sending work there too.
This has reproduced more than once.
So in this case the failure does not look like a permanently broken CLI/runtime/WSL setup. It looks like some Desktop-side conversation/session/composer initialization state is not ready for the first target chat after restart, while sending in another chat initializes enough shared state for the original chat to recover.
This is different from the stale CODEX_CLI_PATH protocol-mismatch case in #29159: the CLI path is current and the WSL runtime is healthy. It also seems related to the broader Unable to send message / Loading... cluster in #29632, especially the reports where the first send attempt initializes enough state for later sends to work.
Windows Codex Desktop
App package: OpenAI.Codex_26.623.19656.0
CLI/runtime: 0.142.5
Project path: C:\dev\codex-empty
Symptom: creating a new project conversation spins forever.
Observation: backend/local rollout threads are created and recorded in .codex-global-state.json, but the project sidebar does not attach/open them.
Example hidden thread ids:
019f425c-554f-77e0-a395-cc4cfc3a0a97
019f425e-68ff-77e0-9ce9-e495e34a18ab
019f4261-60bc-7c31-959c-7732788d42d1
Other project C:\dev\codex-test works normally.
Activity
github-actions commentedon Jul 3, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
fibo-blip commentedon Jul 3, 2026
Adding another Windows Desktop data point that may help separate this from pure network/auth failures.
Environment:
OpenAI.Codex_26.623.13972.0_x64__2p2nqsd0c76g0$HOME=/home/yurycodex-cli 0.142.5CODEX_HOMEis unsetsetup.ps1User-visible symptom:
New observation / workaround:
So in this case the failure does not look like a permanently broken CLI/runtime/WSL setup. It looks like some Desktop-side conversation/session/composer initialization state is not ready for the first target chat after restart, while sending in another chat initializes enough shared state for the original chat to recover.
This is different from the stale
CODEX_CLI_PATHprotocol-mismatch case in #29159: the CLI path is current and the WSL runtime is healthy. It also seems related to the broaderUnable to send message / Loading...cluster in #29632, especially the reports where the first send attempt initializes enough state for later sends to work.ideajoker commentedon Jul 8, 2026
The same
Windows Codex Desktop
App package: OpenAI.Codex_26.623.19656.0
CLI/runtime: 0.142.5
Project path: C:\dev\codex-empty
Symptom: creating a new project conversation spins forever.
Observation: backend/local rollout threads are created and recorded in .codex-global-state.json, but the project sidebar does not attach/open them.
Example hidden thread ids:
019f425c-554f-77e0-a395-cc4cfc3a0a97
019f425e-68ff-77e0-9ce9-e495e34a18ab
019f4261-60bc-7c31-959c-7732788d42d1
Other project C:\dev\codex-test works normally.
ThEWiZ76 commentedon Jul 9, 2026
My issue was fixed with version 26.623.141536, closing issue