Using OpenAI models not hosted by OpenAI for transcription

[BrynLoftus]

Can we get an option to point transcription to a locally hosted model?

I want to be able to potentially do two things:
- Point it to an azure/copilot openai model, as many clients already have this setup
- Point it to a locally hosted model. I know we can do this, in some ways, with the new AI server but that requires that the AI server has the hardware for it, but we may want to point it to larger hardware that is running multiple services (not just 3cx).

 

[Charles_3CX]

Hi

Currently, there is no support for Azure OpenAI or arbitrary locally hosted transcription endpoints. Supported options include OpenAI directly, Google Cloud, 3CX Cloud Transcription, or the 3CX AI Server.

The V20 Update 10 AI Server can run on separate hardware and serve multiple PBXs, but it uses the 3CX-managed AI Server stack. It cannot point transcription to an existing Azure OpenAI or other OpenAI-compatible endpoint.

 

[BrynLoftus]

Exactly- is there any potential of fixing that? For customers that don't have a relationship with OpenAI, but do use Microsoft (very common), it would be vastly easier from a billing/management standpoint to be able to use Azure OpenAI.

 

[Dwyte]

Hey, echoing Bryan's question in pointing to different models, we are stuck on what should be a really easy one.

For our OpenAI account to be HIPAA compliant we need to us the US only OpenAI API. Its OpenAI, just a different URL

https://api.openai.com/v1 is hardcoded in, but does not work. We need to send requests to https://us.api.openai.com/v1.

Is there any way to use a custom endpoint or put in a parameter or anything to change the default OpenAI API URL?