what I have done in my case is to have an agent configured in an express application that grabs the model from the vapi api assitance endpoint and use that model to drive the agent configuration and connecting directly with the LLM odel I am using ( gpt-4o