Hi! I want to use VAPI through my Azure OpenAI account and its model there. I have linked it successfully in the Provider Credentials. The message there is "Note: When credentials are set, Azure OpenAI models are automatically used." - Now, if I select "OpenAI" on the Provider list, it uses the regular VAPI-provided OpenAI model instead of mine in Azure OpenAI (I can see that through the activity protocol). Also if I select "custom-llm" as provider and give the endpoint to the model deployment into the Custom LLM URL field doesnt work. What am I doing wrong? In the support section its said to choose OpenAI and "gpt-4o-cluster" as the model (the model deployment also is a gpt-4o), but it still uses the VAPI-provided OpenAI model, no activity at my deployment. The call id is 55c9dbc7-2344-4a1b-b868-628a0b4d8797 (menu selections: Provider "OpenAI", model "gpt-4o-cluster"). I need it for EU data protection reasons btw., maybe VAPI could provide a finished solution for that straight in the model selection, with models runned on EU servers so it can be used commercially legally without having fear to be sued by a callee.

Tracking this. We are looking to do similar in Australia

Hey @Mario, When both Azure OpenAI and OpenAI credentials exist, the system has a precedence order that favors OpenAI over Azure OpenAI. To fix this and ensure Azure OpenAI is used: - Remove the OpenAI credential if you want to exclusively use Azure OpenAI

Okay, it doesnt use the regular VAPI OpenAI provider now, and seems to try to connect with my Azure OpenAI model deployment, but it still fails. The message "Exiting meeting because room was delayed" appears straight a few seconds after I start to speak - if I wait with speaking, the call maintains until I begin to speak. The logs says "Pipeline no available mo..." The call-id`s are f9bba5df-92e0-4ace-9945-721914514ea6 (long call) and e071a88c-4bd2-419d-b32d-899d934bc00d (short call). Here is the setting I use on my Azure OpenAI deployment, together with the setting I use on VAPI and on the Provider Credentials. Is there anything wrong? I notice the Azure OpenAI deployment has its own parameters like temperature and maximal response length, could that interfere with the settings of the VAPI assistant? I also noticed that every time I refresh my Provider Credentials page, the Azure OpenAI section shows "australia" with a mark beside it, and I have to select "swedencentral" manually every time - and only then the mark is rightfully on "swedencentral". Could it be that VAPI doesnt save the filled out credential?! Because I wonder what happens if there are multiple connections, which one selects VAPI then, if the mark isnt placed on the preferred one? Thanks for helping me @Shubham Bajaj! https://cdn.discordapp.com/attachments/1342900897914556467/1343562318101086311/Screenshot_2025-02-24_125741.png?ex=67bdb96a&is=67bc67ea&hm=36a2d7e5c86b96570152825c381df2412031459f208f4892691c65997051dd3f& https://cdn.discordapp.com/attachments/1342900897914556467/1343562318344224808/Screenshot_2025-02-24_125831.png?ex=67bdb96a&is=67bc67ea&hm=bba8e874e61c8829d8518c6ba80d99f486dc0d2bf5d10b90fbfd1e0c048855b9& https://cdn.discordapp.com/attachments/1342900897914556467/1343562318700744705/Screenshot_2025-02-24_125445.png?ex=67bdb96a&is=67bc67ea&hm=257efdbc9552dddcc57b3e2716a9e277d30e421ba54cc8e195603d63e5adf0c1& https://cdn.discordapp.com/attachments/1342900897914556467/1343562319069839441/Screenshot_2025-02-24_131121.png?ex=67bdb96b&is=67bc67eb&hm=900e55b7324a3254899311d45eeb2db87a0984bddd1bd08fddba647820d75721& https://cdn.discordapp.com/attachments/1342900897914556467/1343562319380086805/Screenshot_2025-02-24_131444.png?ex=67bdb96b&is=67bc67eb&hm=3bb5d7f3e06c0ffb5954c091ee9b26bcc10b0b69ba850dd5d5b5cf9cfd143282& https://cdn.discordapp.com/attachments/1342900897914556467/1343562319732674642/vapi_credentials.jpg?ex=67bdb96b&is=67bc67eb&hm=d52188866827b9348d7f0aad438b2e7cc936cb650f3e06c1147f8c9ef677be34&

Mario looking into it.

Anything resolved on this issue?

The issue is occurring because you are not selecting the correct model name in the model section through the API. Once you do so, the problem will be resolved.

Same issue here

@wizzy Can you please ping me to your support ticket? I will take a look.