›_ Codex Issues
Codex Issues / #21513

Session sync between CLI and App diverging

Resolved 💬 3 comments Opened May 7, 2026 by ebarti Closed May 7, 2026
💡 Likely answer: A maintainer (etraut-openai, contributor) responded on this thread — see the highlighted reply below.

What issue are you seeing?

I had a running session (/goal) for about 22h in the CLI that was correctly being represented/rendered in the Codex App.

At some point, I prompted Codex from the Codex App, which responded successfully and it streamed on both APP and CLI. After responding, the App took a differentvpath than the CLI and made conflicting changes, of which the CLI was completely unaware.

Apologies as I do not have the screenshots of when both CLI and Codex App were fully diverged. Therefore, the screenshot below for the CLI is not very representative, as I quit the session and resumed it, which forced the diverged conversation to be merged with the single thread. However I do have the screenshot of the diverged path in the Codexc App, which persisted until I quit and reopened the app.

CLI (after quit):
<img width="1004" height="996" alt="Image" src="https://github.com/user-attachments/assets/8e63e605-dba5-49f7-ba82-b598ecf3330c" />

Codex App (before quit):

<img width="734" height="558" alt="Image" src="https://github.com/user-attachments/assets/d6f91f24-5ab8-452b-aa05-6a86b7a8608d" />

Codex App (after quit)

<img width="756" height="565" alt="Image" src="https://github.com/user-attachments/assets/f19db537-aa59-48a7-ae01-77a9793e8ac5" />

As you can see, when both threads are merged into a single one the logical thinking makes no sense:

I’m fixing the eval subject path now and will update PR #330 instead of merging a partial fix.

And right after:

PR #330 is still mergeable and both CI jobs are green. I’m merging it now, then I’ll rerun from the merged state rather than the PR branch.

When both threads

What steps can reproduce the bug?

I was able to reproduce when:

  1. A goal is running
  2. I pause the active work from the Codex App
  3. Prompt in the Codex App
  4. The CLI continues its goal pursuit, diverging from the Codex App

What is the expected behavior?

_No response_

Additional information

_No response_

View original on GitHub ↗

3 Comments

etraut-openai contributor · 3 months ago

This issue isn't specific to the "goals" feature. In general, we don't currently support the use of a session from two different clients at the same time. It works fine if you start a session in one client (like the CLI) and then stop the session in that client and resume it in a second client (like the desktop app). The problem is that each client currently has its own instance of the app server, and those are different processes that don't know about what the other instances are doing.

ebarti · 3 months ago
This issue isn't specific to the "goals" feature. In general, we don't currently support the use of a session from two different clients at the same time. It works fine if you start a session in one client (like the CLI) and then stop the session in that client and resume it in a second client (like the desktop app). The problem is that each client currently has its own instance of the app server, and those are different processes that don't know about what the other instances are doing.

Thank you for responding. I guess the new codex release fixes the very reason why I had this happen

Experimental goals are now discoverable, stay paused across resume unless the user opts back in, and show clearer validation and multi-day duration output. (https://github.com/openai/codex/pull/20083, https://github.com/openai/codex/pull/20790, https://github.com/openai/codex/pull/20746, https://github.com/openai/codex/pull/20558)
ebarti · 3 months ago

I would suggest to document this explicitly - the user should not have to guess whether this is supported or not.

However, I think there's already enough feature requests regarding proper session management CLI/Codex App that I hope a better solution gets put in place sometime soon.