Usage Limit Interrupts Workflow

[REtoolsx]

What version of Codex is running?

visual stuido code extension 0.5.11

Which model were you using?

gpt-5-codex

What platform is your computer?

Microsoft Windows NT 10.0.26100.0 x64

What steps can reproduce the bug?

This issue occurs when GPT reaches its usage limits while continuing its workflow. More specifically, if the limit is exceeded during a patch or code modification, GPT may leave faulty code fragments in different parts of the project.

What is the expected behavior?

Usage limits should not interrupt GPT while it continues its workflow. The limit check should only occur when the user submits a new request. This way, edits to the project will not be abruptly stopped, and potential errors can be avoided.

What do you see instead?

When GPT runs in automatic mode, hitting the usage limit interrupts the process mid-function, causing the operation to end up incomplete, partial, or erroneous.

Additional information

No response

[wbdb]

I agree with you, and there have been several suggestions here for some time now, such as an API fallback, etc.

1. Automatic use API key once chatGPT account limits exceeded #2992 (upvote here)
2. API fallback #3551
3. When logged in with ChatGPT Plus/Pro while also supplying an API key, automatically fall back to using API credits once ChatGPT credits run out #2478

[jacko555]

Same issue or similar.

I have it connect to openai account on pro.

But i get this issue stream disconnected before completion: Your input exceeds the context window of this model. Please adjust your input and try again..

Then doesn't let me continue working within chat.

[baran312]

stream disconnected before completion: Your input exceeds the context window of this model. Please adjust your input and try again

[jacko555]

Its likely because the addon in sidebar forgets to compact conversion like what cursor does. I haven't used terminal too much yet but i assume it auto compacts like cc?

[wbdb]

@jacko555 I mean, with gpt-5, it did compact automatically. With gpt-5-codex, I haven't experienced that in the CLI even with < 30% context.

Edit: checked the code no default auto compact actually. But can set via config.toml.

[jacko555]

I think its just in the chat add on them in ide