Codex remote H100 connection shows false usage limit and unfamiliar chat history

What version of the Codex App are you using (From “About Codex” dialog)?

Codex App version 26.903.71938 (build 8576)

What subscription do you have?

ChatGPT Pro

What platform is your computer?

Darwin 25.6.0 arm64 arm

What issue are you seeing?

I am using Codex in the ChatGPT Desktop app with a ChatGPT Pro subscription. The issue occurs specifically in my “Multimodal Search” project connected to an H100 remote server.

Codex repeatedly displays:

You've hit your usage limit. Add credits to continue, or try again later.

The error occurs even in a newly created remote-project chat containing only:

hi

My Codex usage is not exhausted (approximately 64% used). I have not purchased credits and have not used my available banked reset.

Codex works normally in a non-remote/local chat under the same account. The false usage-limit error occurs specifically when I use the H100 remote project connection.

I also see remote-associated historical chat entries that I do not recognize as chats I created locally. Some titles reference paths such as:

/home/guests/zhen/dir1/LenVM/...

I did not open unfamiliar conversations further, to avoid accessing potentially unauthorized content.

What steps can reproduce the bug?

1. Sign in to the ChatGPT Desktop app with my ChatGPT Pro account.

2. Open the “Multimodal Search” project connected to the H100 remote server.

3. Create a new chat in that remote project.

4. Send a minimal prompt:

hi

5. Observe that Codex immediately displays:

You've hit your usage limit. Add credits to continue, or try again later.

6. Check Settings > Usage. The Codex allowance is not exhausted (approximately 64% used), and no credits or banked reset have been used.

7. Create a non-remote/local chat under the same account and send a prompt. It works normally.

8. In the remote project, inspect only the chat-title list. Remote-associated historical entries appear that I do not recognize as local chats.

What is the expected behavior?

The H100 remote project should use the same ChatGPT account, workspace identity, and Codex usage allowance shown in the Desktop app.

A new remote-project chat should work when the account has remaining usage.

Connecting to a remote server should not cause a false usage-limit error or display unfamiliar conversation-history entries.

Additional information

No response

Start by reproducing the issue in the ChatGPT Desktop app through the “Multimodal Search” H100 remote project, then compare its usage-limit and chat-title behavior with a local chat. Trace the remote connection’s account, workspace, allowance, and history handling; done means a remote chat works with remaining usage and shows only authorized conversation history.