I want to deploy the gpt-5.3-codex for my team but my team only have limited budget. After discuss with Azure support team they told me I could create a APIM in azure, which is a gateway can manage our token quota. But when I use the endpoint and secret key provided by APMI I got ‘We’re having trouble finding the resource you requested. If the problem persists, please contact support.‘ error. Anyone have idea how to solve this?
Hey there, thanks for the details.
At the moment, Cursor’s Azure OpenAI BYOK expects a direct Azure OpenAI resource endpoint (like https://<resource-name>.openai.azure.com/).
Azure API Management (APIM) uses a different URL structure (e.g.,https://<apim-name>.azure-api.net/...) and different auth (APIM subscription keys), so it won’t work as a drop-in base URL in Cursor. This is why you’re seeing “We’re having trouble finding the resource you requested.”
What you can do instead
-
Use the direct Azure OpenAI endpoint in Cursor’s API Keys settings: API Keys docs.
-
If your goal is team quota/throttling, use Azure OpenAI deployment rate limits (TPM/RPM) rather than APIM in front of it.
-
If you specifically need admin-managed / centralized keys (where users don’t see the key), the only provider we currently support for that is AWS Bedrock.
Hope this helps!
Best,
Mohit
Using APIM with AOAI is ideal for my team as well.
It shouldn’t be that heavy of a lift. Cursor already takes in a base URL, deployment name, & api key. You would only need to add a checkbox to allow the user to enable APIM, a field for the ‘Header name’ (the default is api-key) for users that set one, & some conditional logic to change the request format. You could enforce that users keep the same routes as the base models for parity. Don’t see how it wouldn’t be quick to implement.
Using AOAI directly isn’t ideal for team use, since everyone would share the same key. I’d love to see this prioritized soon!!!
Hi there,
Thanks for the detailed feedback. The team governance concern is valid.
Unfortunately, Cursor’s Azure OpenAI integration currently only supports direct resource endpoints, and we don’t have a timeline for APIM support.
For now, if centralized key management is the primary goal, AWS Bedrock is the only provider we currently support that lets admins manage keys without individual users needing direct access. More details on setup: AWS Bedrock setup guide.
I’ve shared your feedback with the team, appreciate you spelling out the use case and implementation details.
We are also asking our team to use AOAI via APIM.
One reason for introducing APIM is that our quota is insufficient and we want to expand it, but the main purpose is traceability.
We cannot capture, through AOAI’s standard features alone, who used it, when, and how much they used. To do that, requests need to go through APIM. I believe this just means allowing an APIM URL, not only an AOAI URL, as the base URL.
Also, by using the Azure OpenAI v1 endpoint, it should be possible to use it transparently with the standard OpenAI-style interface rather than relying on the conventional Azure OpenAI API endpoint–specific methods. In other words, whether it is APIM or not should not matter; any OpenAI-compatible base URL should work.