The domain troubleshooting guide can help with most custom domain configuration issues. You might be able to use that guide to solve it before a human is available to help you. Then you can come back here and share the answer for bonus points.

https://vercel.com/docs/projects/domains/troubleshooting

You can also use v0 to narrow down the possibilities.

Hi Mustafa,

We are triaging this request in ticket 00447461 currently and get back to you once we have more information.

Same problem with deepseek ai. Can you give more information and time resolving expectations?

Can you share the logs URL?

https://vercel.com/mflash123s-projects/ai-tm-fitness/logs

But what about 00447461 ticket with openai.com? I am sure they have same problem. So, i have same issue with openai too. If i understand from the previous reply, the ticket is not resolve ?

I've reported this issue to both Vercel and Inngest support teams. While Inngest has acknowledged the problem and indicated they would escalate it to their infrastructure team, Vercel support has unfortunately been unable to provide a resolution or clear explanation for the issue.

This problem appeared suddenly and without warning. The application was working perfectly fine up until a week ago, and there have been absolutely no code changes made to the project that could explain this sudden failure. All AI-related functions using Inngest's agent-kit with OpenAI integration are consistently failing with the same NAT64/IPv6 addressing error.

For reference, I've included a direct link to one of the error logs in the Vercel dashboard: https://vercel.com/oplog/nexus-workflow/logs?selectedLogId=67gsm-1744627347565-efce0e34f36c

While my initial suspicion is that this might be a Vercel infrastructure issue related to how outbound connections to OpenAI are handled, I recognize it could potentially be an Inngest-side problem as well. The fact that no code changes were made, and that everything works in local development but fails in production, strongly suggests this is an infrastructure/networking issue rather than a code problem.

I would greatly appreciate if both teams could coordinate efforts to investigate this persistent NAT64/IPv6 error that is currently blocking all AI functionality in our production environment.

I am working on ticket 00447461 and unfortunately don't have any further update. I have just ecalated the ticket internally.

Looks like vercel has a mass problem with inaccessibility to openai and deepseek.

We are also speaking with Inngest team internally.

@eraykeskinmac @mflash123 Hi there, my name is Shohei and I work in the compute team at Vercel.

In this case, since you're using Inngest's AgentKit, your Vercel Function is not making the requests to the external API endpoint directly, but it is through Inngest's AI Gateway proxy.

Vercel Function -> internet -> AI Gateway (managed by Inngest) -> internet -> External APIs (e.g., OpenAI)

The thing is, this particular NAT64 error is returned by the AI Gateway server, and your Vercel Function is just reporting what it got from the AI Gateway server.

This means that the request left the Vercel Function, and then the AI Gateway server tries to resolve the hostname and somehow gets a NAT64 answer, which then fails to proceed within its infra. Therefore, this is not related to our infra/Vercel Function

I believe the Inngest team is looking into this now. You may contact them directly on this matter.

Is it possible to use vercel and make request directly to external API without Inngest Gateway?

Thank you for the clarification, Shohei. That makes sense.

Yes, I've been in contact with Inngest support about this issue, and they are aware and investigating the potential problem within their AI Gateway. We've also discussed possible workarounds, potentially avoiding the AI Gateway for these calls.

Hopefully, we can find a resolution soon. I'll update this thread if a solution is found. Thank you for your help in pinpointing the origin of the error!

Thanks for the update and the context with the Inngest team. Since this is a public forum, your input is very helpful for others facing the same issue, so thank you!

We’ve also discussed possible workarounds, potentially avoiding the AI Gateway for these calls.

Would you mind sharing some of those details so that @mflash123 may be able to adopt the same workaround?

Hello!I hope you're doing well. Could you kindly let me know whether you've encountered similar challenges before and, if so, whether they were resolved? I’m currently unable to verify this on my end, as I had to pivot the project due to this issue. Any insights you could share would be greatly appreciated.

Thank you so much for your time!

@smaeda-ks Unfortunately, I had to pause the project temporarily as the error we encountered became a blocker, but I’m eager to revisit it when possible.