The connection failed. Check connection.... VPN?

Hi everyone, I have a consistent connection issue for many weeks now. However I do not use a VPN.

Please check your internet connection or use a VPN and try again. If this persists, please email

I have contacted support, but it seems the issue persists, it is very inconsistent, where sometimes I get the modal about “connected issues” with VPN.

I really would like to use Cursor, but it seems the “code implementation” feature fails probalby 60% of the time.



What AI model are you using? And are you using Cursor’s requests or your own OpenAI Key?

Haven’t switched the Model, but had my key before subscribing, but the error happens with both with key and without it.

I am having the same issue today.

Same issue for me. No matter the model, connection failed in all the cases

I’m also having the issue starting today too.

For both GPT-3.5 and GPT-4.

I’m using Cursor request.

I also keep getting connection failed for GPT4 and 3.5

edit: oh it’s working now

Still failing for me :person_shrugging:

I am also encountering the same issue and am unable to use Cursor.

Yes the issue has been consistent for many weeks… the issue persists.

If you’re using a data center IP, you may be blocked from access. Use a home IP. Network errors are most often response errors from the OpenAI server.

No data center IP, also I use the OpenAI API with streaming with other apps and no problem.

Hey! As an immediate fix, could you try changing your model name? There’s some chance this is a bug that occurs if you were using an API-key-only-model and you switch from API key to no API key.

Is this persistent for you? Any reproduction steps if not?

Hi everyone, I have a consistent connection issue for many weeks now. However I do not use a VPN.

Does this only happen for you on Command-K requests over large ranges (> 300 lines of code)?

It happens more frequently on the inline chat with the code substitution, but it happens on side chat also, but the error is more generic.

Also there might be a correlation with context length and failure but it happens even with code of around 60-100 lines.

Not sure if it makes sense but the project I am working on is open source, not sure if the problem is deeper or even makes sense to try with it, since I know there are project specific indexing.

I have 4-10 files open, and it consistently fails.

FYI it does happen on other projects, but this might be the closest I can to giving you a reproducible example.

I worked today from a different place/connection and still got the error.

Other devs at the company which we are also validating cursor have seen the error but maybe less frequently?!?

Another clue, sometimes I can go a few runs without seeing the error, but once it happens, it becomes difficult to “not” get the error, and I have to stop using it.