Nvidia endpoint not working

[Shivanshmundraa]

Odysseus not working with NVIDIA endpoint — getting 404 error

Problem

I'm trying to connect Odysseus to an NVIDIA endpoint but getting a 404 error:

NVIDIA returned 404 — check the base URL and model name. 
(Function '23bd454d-b225-49a3-8118-582a62fc51b8': Not found for account 'cCju_r305G78HJvcuA9FBTHUxkhjQTHA9SJIenhoGjU')

Setup

• Odysseus: self-hosted, running locally
• NVIDIA endpoint: https://integrate.api.nvidia.com/v1
• Using NVIDIA NIM API with an API key

What I've checked

• Base URL looks correct
• API key is set up

Questions

1. Should I be using a specific model name or function ID?
2. Any documentation on configuring NVIDIA endpoints with Odysseus?
3. Why only gpt-oss working ?

Has anyone else hit this? How did you fix it?

Thanks!

[akakarantzas]

I think the API key/base URL are probably not the issue here. That Function ... not found for account error smells like Odysseus is sending NVIDIA a model/function id your account does not actually have access to.

I'd double check the exact model name you're passing in the Odysseus config. For NVIDIA NIM, it usually needs the real model slug from their catalog, not just any function UUID. The fact that gpt-oss works kinda supports that too. That one is probably mapped correctly, while the NVIDIA model name is getting sent wrong and NVIDIA just 404s it.

So I’d try:

• keep base URL as https://integrate.api.nvidia.com/v1
• use the exact NVIDIA model slug
• don’t use the function UUID unless Odysseus explicitly wants it
• test the same model/key with a direct NVIDIA curl request first

If direct NVIDIA works but Odysseus 404s, then it’s probably an Odysseus provider config/mapping issue.

[Shivanshmundraa]

Nvidia api if firing properly its issue with odysseus

[akakarantzas]

Yeah, if the NVIDIA API works when you call it directly, then it probably isn't your API key or base URL.

I'd take a look at what odysseus is actually sending to NVIDIA. The main thing I'd compare is the model name in the odysseus request versus the one that works directly with NVIDIA. That "Function not found for account" error makes me think it could be sending or resolving the wrong model/function ID somewhere along the way.

If you want, you can share the following (with any sensitive info removed) and I can try to help narrow it down:

• Your odysseus provider/model configuration
• The exact NVIDIA model name that works with curl
• Any debug logs or the request payload odysseus is sending

That should make it easier to see whether the issue is in the odysseus configuration or how it's mapping the model internally