Hi all 👋 I'm using a custom LLM on my Vapi assistant and I want to trigger the basic phone functionalities provided by default: call end, call transfer and dial. I don't understant how I should generate the chunk for the tool call in order to trigger the action or if there is some configuration to do to enable these functionalities ( I already enabled the check on functions menu) Thanks for your support!

@Marco Falcone Hello there! 👋 Great to hear that you're using a custom LLM with your Vapi assistant. To trigger the basic phone functionalities such as call end, call transfer, and dial, you would typically use specific commands or API calls that are designed for those actions. Since you've mentioned that you've already enabled the check on the functions menu, you're on the right track. The next step is to ensure that your custom LLM can generate the appropriate commands or API calls that correspond to the actions you want to trigger. For example, if you want to end a call, your LLM should generate a command or an API call that is recognized by Vapi as an instruction to end the call. The same goes for call transfer and dialing functionalities. If you're unsure about the exact format or the API calls needed to trigger these actions, you may need to refer to the Vapi documentation or ask support for more info on this. They can provide you with the specific syntax or commands that your custom LLM needs to output in order to interact with the phone functionalities. If you have any specific commands or API calls in mind and need assistance on how to format them for the tool call, please let me know, and I'll be happy to help you further!

@Vapi I tried to send the following chunk for call end:

json
{
  "id": "indigoai-1",
  "object": "chat.completion.chunk",
  "created": 63904351243,
  "model": "ndg",
  "system_fingerprint": "fp_ndg",
  "choices": [
    {
      "delta": {
      "tool_calls": [
         {
           "function": {
              "arguments": "{}",
              "name": "endCall"
           },
           "id": "call_84934067-e51f-4380-a76e-0dfb8f32ff8a_0",
           "index": 0
         }
       ]
     },
     "finish_reason": "function_call",
     "index": 0,
     "logprobs": null
    }
  ]
}

Is this correct?

@Marco Falcone The chunk you've provided seems to be structured correctly for a tool call, but it's important to ensure that the function name and arguments match exactly with what the Vapi system expects. In the chunk you've sent, the function name is

endCall

, and it's being called with an empty arguments object

{}

. However, I don't have the exact specifications for the

endCall

function or other related functions like

transferCall

or

dial

in the context provided. If

endCall

is indeed the correct function name and it requires no arguments, then your chunk should work as intended. If you're not sure about the function names or the required arguments, please refer to the Vapi documentation or ask support for the correct specifications. If

endCall

is not the correct function name or if there are specific arguments needed, you'll need to adjust the chunk accordingly. If you need further assistance with the correct function names or arguments, please let me know, and I'll be glad to help!

@Vapi I need assistance with the naming

@Marco Falcone Please ask support for more info on this.

Please 🙏

@Marco Falcone Try out the following code snippet and let me know how it goes.

py
                if request_payload.function_name == "transferCall" and request_payload.destination:
                    function_call_payload = {
                        "function_call": {
                            "name": request_payload.function_name,
                            "arguments": {
                                "destination": request_payload.destination
                            },
                        }
                    }
                    yield f"data: {json.dumps(function_call_payload)}\n\n"

This doesn't work for me. I try to give you more context. This is the request on call start arriving to my custom LLM

@Marco Falcone available for a quick call?

Sure

check your DM.