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
Codex desktop app only offers Local when continuing a visible Cloud task #29694
In the Codex desktop app on macOS, I can see a Codex Cloud task under a project/environment, but when I try to continue it from the desktop app composer, the only available destination is Local. There is no Cloud option.
The composer text says:
Create a new local task that references this cloud task
and the destination dropdown shows only:
Continue in
- Local
This makes it look like the desktop app can view/reference the cloud task, but cannot continue it in Cloud.
Expected behavior
When viewing a Codex Cloud task in the desktop app, I expected to be able to continue/follow up in the same Cloud environment, or at least choose Cloud in the destination selector.
Actual behavior
Only Local is offered. Continuing from the desktop app creates a local task that references the cloud task instead of continuing the cloud task in Cloud.
Evidence / checks
The same Cloud environment is visible and works in the web app.
The desktop app sidebar shows the project/environment and the Cloud task.
codex cloud list on the same machine can see the task successfully.
CLI output shows the task as [READY], environment label rose, with no diff.
codex cloud --help is available locally.
Redacted CLI/task detail:
codex-cli 0.128.0
codex cloud list
https://chatgpt.com/codex/tasks/task_e_[redacted]
[READY] Hello
rose • ~20m ago
no diff
Environment
Platform: macOS
Surface: Codex desktop app
CLI version: codex-cli 0.128.0
Cloud environment label: rose
Branch shown in web app: main
Related/possibly adjacent
I found a similar/adjacent issue about desktop app Cloud task availability, but this case is different because Cloud auth works and the task is visible in both the web app and CLI; only desktop follow-up routing is limited to Local.
Additional data point: I can reproduce a similar issue on a second Windows machine as well.
On that Windows machine:
Cloud environment is configured
GitHub repository is connected
The Desktop app no longer exposes the quick commands/actions for running tasks from the configured environment
I also no longer see a way to send work to Codex Cloud from the Desktop app
So this does not look macOS-specific. It appears to affect the Desktop app UI/workflow across at least macOS and Windows after a recent update.
On my macOS machine:
Codex Desktop: 26.623.101652
Bundled CLI: codex-cli 0.142.5
codex cloud list works
Local state shows Codex Cloud access enabled and an existing Cloud environment
But Desktop only exposes the local path instead of the expected Cloud workflow
Impact:
This blocks using Codex Desktop for hosted Cloud tasks even when Cloud access and environments are configured. The workaround is to use Codex web or CLI, but the Desktop UI no longer exposes the expected Cloud launch path.
I can reproduce a closely related issue on Windows using Codex Desktop 26.715.8383.0.
Unlike the original report, this occurs when creating a new task from an existing Cloud project, rather than when continuing an existing Cloud task.
Expected behavior
Creating a task from the configured Cloud project should start a task in its associated Cloud environment.
Actual behavior
Codex Desktop creates a local, projectless task. The resulting task metadata records:
environment_id = "local"
workspace_kind = "projectless"
A local working directory under C:\Users\<USERNAME>\Documents\Codex\…
At the same time, the application state indicates:
Cloud access is enabled
Work run location is set to Cloud
A valid Cloud project is selected
The configured Cloud environment is writable
There is no Cloud submission or rejection error—the client appears to create the task as Local from the outset. The Cloud selector and /cloud-environment command are also unavailable.
This appears consistent with the Desktop Cloud-to-Local routing problem described in this issue, but through the new-task flow.
Reproduced on macOS with desktop version 26.901.51231 (bundled app-server reports 0.153.4).
The existing Cloud task loads successfully in the desktop app and has a cloud icon in the sidebar. However, the “Run follow-up in” dropdown contains only Local; Cloud is absent. The same task can be continued on the web, and a live codex cloud status check succeeds and reports READY. This is not evidence of an expired or inaccessible task.
Reproduction
Open an existing Codex Cloud task in the desktop app.
Use a desktop context where the follow-up composer does not resolve a local Git repository.
Open the follow-up location dropdown.
Observe that only Local is offered, although the original cloud task is accessible on the web.
Implementation evidence from the installed build
Read-only inspection of the installed frontend identifies the follow-up location component in composer-utility-bar-731e84ddf1fb.js:
It obtains a desktop execution context and queries Git metadata for that context's directory.
It renders the Cloud choice only when the Git metadata contains a repository root.
After metadata loading finishes without a repository root, an effect switches the composer mode to Local.
This appears to couple continuing an existing Cloud task to local Git discovery, rather than to the existing task's cloud environment. The observed menu matches this branch. The precise runtime directory was not captured, so the remaining question is why Git discovery does not resolve a root for this task.
There is also a desktop warning for the affected task: “No owner repo found for remote task”. Inspection shows this warning can occur when resolving the task's environment label to environment repository metadata fails. It does not by itself prove that the environment was deleted or that the local clone is missing.
The persisted desktop Cloud-access state is enabled (cached state, not a live entitlement check). No credentials, cache, configuration, or application files were modified during diagnosis.
Expected behavior
Allow an existing Cloud task to continue in its cloud environment without requiring a local Git checkout. If cloud continuation is unavailable for another reason, show that reason explicitly instead of silently hiding Cloud and switching to Local.
This confirms the behavior on the version above; it does not establish which release introduced it. Task identifiers and unrelated screenshot content are omitted for privacy.
I am seeing a closely related regression in the current Codex Cloud workflow, with an additional continuity problem that makes iterative PR work difficult.
Observed behavior:
I start a new Codex Cloud task. The initial Cloud task does not show a model picker.
After the Cloud task completes, I try to send a follow-up.
At that point a model picker appears, but the continuation flow is Local-oriented rather than continuing the same Cloud task/workspace.
More importantly, the original Cloud task is no longer reopenable/resumable from the UI in a way that allows another Cloud turn.
As a result, I cannot use the previous workflow of:
Cloud task -> implement -> PR -> CI feedback -> same Cloud task -> update same PR
while preserving the Cloud task's context/workspace.
To continue using Cloud, I effectively have to start a new Cloud task. That loses the previous task/session context and makes iterative work on an existing PR substantially more cumbersome; in practice it pushes the workflow toward one bounded Cloud task per iteration instead of repeated updates from the same Cloud session.
The model-picker behavior also suggests that the follow-up composer may be routing into a new Local continuation rather than resuming the original Cloud execution context: the picker is absent for the initial Cloud task but appears only when attempting the follow-up that is Local-oriented.
Expected behavior:
A completed Cloud task should remain reopenable/resumable.
Follow-up should allow another turn in the same Cloud task/workspace.
The same Cloud task should be usable for iterative development on the same branch/PR, including responding to CI failures and updating the existing PR without reconstructing task context from scratch.
If Local handoff is offered, it should be optional rather than the only continuation path.
This is especially disruptive for longer-lived engineering work where a PR is refined over several CI/review cycles. The issue is not merely losing chat history: the practical regression is loss of Cloud task/workspace continuity between iterations.
Update: Cloud continuation has become available again for me, but the underlying iterative-PR problem remains.
I tested continuing work from the same Cloud session. The session itself can now accept another Cloud turn, but it cannot update the existing branch/PR created by the earlier turn. To continue the work, I am effectively required to create a new PR.
That means the practical workflow is still roughly:
Cloud task/session -> PR A -> follow-up in same Cloud session -> cannot update PR A -> create PR B
This is not materially better than starting a brand-new Cloud session for the next iteration, because the normal engineering loop I need is:
Cloud task -> implement -> PR -> CI/review feedback -> same task/session -> push to same branch -> update same PR
Expected behavior:
A resumed Cloud task/session should retain or recover the original branch/worktree association.
A follow-up should be able to commit/push to the same branch and update the existing PR when that is the user's intent.
Creating a new PR should be optional, not required for every follow-up iteration.
So Cloud-session continuity appears partially restored, but branch/PR continuity is still broken. For iterative development, this remains a significant regression.
Activity
demiurg92 commentedon Jul 3, 2026
Additional data point: I can reproduce a similar issue on a second Windows machine as well.
On that Windows machine:
So this does not look macOS-specific. It appears to affect the Desktop app UI/workflow across at least macOS and Windows after a recent update.
On my macOS machine:
codex cloud listworksImpact:
This blocks using Codex Desktop for hosted Cloud tasks even when Cloud access and environments are configured. The workaround is to use Codex web or CLI, but the Desktop UI no longer exposes the expected Cloud launch path.
Ayanmullick commentedon Jul 22, 2026
I can reproduce a closely related issue on Windows using Codex Desktop
26.715.8383.0.Unlike the original report, this occurs when creating a new task from an existing Cloud project, rather than when continuing an existing Cloud task.
Expected behavior
Creating a task from the configured Cloud project should start a task in its associated Cloud environment.
Actual behavior
Codex Desktop creates a local, projectless task. The resulting task metadata records:
environment_id = "local"workspace_kind = "projectless"C:\Users\<USERNAME>\Documents\Codex\…At the same time, the application state indicates:
There is no Cloud submission or rejection error—the client appears to create the task as Local from the outset. The Cloud selector and
/cloud-environmentcommand are also unavailable.This appears consistent with the Desktop Cloud-to-Local routing problem described in this issue, but through the new-task flow.
wenyuanyu commentedon Sep 8, 2026
Reproduced on macOS with desktop version 26.901.51231 (bundled app-server reports 0.153.4).
The existing Cloud task loads successfully in the desktop app and has a cloud icon in the sidebar. However, the “Run follow-up in” dropdown contains only Local; Cloud is absent. The same task can be continued on the web, and a live
codex cloud statuscheck succeeds and reports READY. This is not evidence of an expired or inaccessible task.Reproduction
Implementation evidence from the installed build
Read-only inspection of the installed frontend identifies the follow-up location component in
composer-utility-bar-731e84ddf1fb.js:This appears to couple continuing an existing Cloud task to local Git discovery, rather than to the existing task's cloud environment. The observed menu matches this branch. The precise runtime directory was not captured, so the remaining question is why Git discovery does not resolve a root for this task.
There is also a desktop warning for the affected task: “No owner repo found for remote task”. Inspection shows this warning can occur when resolving the task's environment label to environment repository metadata fails. It does not by itself prove that the environment was deleted or that the local clone is missing.
The persisted desktop Cloud-access state is enabled (cached state, not a live entitlement check). No credentials, cache, configuration, or application files were modified during diagnosis.
Expected behavior
Allow an existing Cloud task to continue in its cloud environment without requiring a local Git checkout. If cloud continuation is unavailable for another reason, show that reason explicitly instead of silently hiding Cloud and switching to Local.
This confirms the behavior on the version above; it does not establish which release introduced it. Task identifiers and unrelated screenshot content are omitted for privacy.
Raskolnikov-kai49 commentedon Oct 5, 2026
I am seeing a closely related regression in the current Codex Cloud workflow, with an additional continuity problem that makes iterative PR work difficult.
Observed behavior:
I start a new Codex Cloud task. The initial Cloud task does not show a model picker.
After the Cloud task completes, I try to send a follow-up.
At that point a model picker appears, but the continuation flow is Local-oriented rather than continuing the same Cloud task/workspace.
More importantly, the original Cloud task is no longer reopenable/resumable from the UI in a way that allows another Cloud turn.
As a result, I cannot use the previous workflow of:
Cloud task -> implement -> PR -> CI feedback -> same Cloud task -> update same PRwhile preserving the Cloud task's context/workspace.
To continue using Cloud, I effectively have to start a new Cloud task. That loses the previous task/session context and makes iterative work on an existing PR substantially more cumbersome; in practice it pushes the workflow toward one bounded Cloud task per iteration instead of repeated updates from the same Cloud session.
The model-picker behavior also suggests that the follow-up composer may be routing into a new Local continuation rather than resuming the original Cloud execution context: the picker is absent for the initial Cloud task but appears only when attempting the follow-up that is Local-oriented.
Expected behavior:
This is especially disruptive for longer-lived engineering work where a PR is refined over several CI/review cycles. The issue is not merely losing chat history: the practical regression is loss of Cloud task/workspace continuity between iterations.
Raskolnikov-kai49 commentedon Oct 5, 2026
Update: Cloud continuation has become available again for me, but the underlying iterative-PR problem remains.
I tested continuing work from the same Cloud session. The session itself can now accept another Cloud turn, but it cannot update the existing branch/PR created by the earlier turn. To continue the work, I am effectively required to create a new PR.
That means the practical workflow is still roughly:
Cloud task/session -> PR A -> follow-up in same Cloud session -> cannot update PR A -> create PR BThis is not materially better than starting a brand-new Cloud session for the next iteration, because the normal engineering loop I need is:
Cloud task -> implement -> PR -> CI/review feedback -> same task/session -> push to same branch -> update same PRExpected behavior:
So Cloud-session continuity appears partially restored, but branch/PR continuity is still broken. For iterative development, this remains a significant regression.