Hello! I am trying to create a simple assistant that will ONLY perform the following tasks: * read out a short, pre-defined greeting * allow the user to speak to leave their message * (optional) end the call once the user has stopped leaving their message I want to do this so that users can leave "voicemails" and I can leverage the VAPI call recording features. Problem is that I can't create an assistant without selecting an LLM and no matter how I prompt the LLM I can't seem to keep it from responding to the user and interrupting their "voicemail". Thanks in advance for your help here!

You can't fully remove the LLM, but you can minimize its role and use assistant hooks to create a voicemail-like flow. Set a pre-defined greeting with

firstMessage

, and use a

customer.speech.timeout

hook to end the call after silence. Here’s an example configuration:

json
{
  "firstMessage": "Please leave your message after the beep.",
  "hooks": [
    {
      "on": "customer.speech.timeout",
      "options": { "timeoutSeconds": 30 },
      "do": [
        { "type": "say", "exact": "Thank you. Goodbye!" },
        { "type": "tool", "tool": { "type": "endCall" } }
      ]
    }
  ]
}

This setup lets users leave a message and ends the call after 30 seconds of silence, closely mimicking voicemail behavior. For more, see the [Assistant Hooks documentation](https://docs.vapi.ai/assistants/assistant-hooks). Source: - [Assistant Hooks documentation](https://docs.vapi.ai/assistants/assistant-hooks)

thanks for the recommendation, I'll look into this. Are there separate configuration options to "silence" the connected LLM and keep it from responding to the user as they are leaving their voicemail?

Message marked as helpful by @Harry! 🎉

There is no configuration to fully "silence" the LLM after the greeting, but you can minimize its responses by using a fixed

firstMessage

and handling all other actions with assistant hooks. For example:

json
{
  "firstMessage": "Please leave your message after the beep.",
  "hooks": [
    {
      "on": "customer.speech.timeout",
      "options": { "timeoutSeconds": 30 },
      "do": [
        { "type": "tool", "tool": { "type": "endCall" } }
      ]
    }
  ]
}

This setup prevents the LLM from generating further responses during voicemail, but does not guarantee absolute silence in all edge cases. See [Assistant Hooks documentation](https://docs.vapi.ai/assistants/assistant-hooks) for more details. Source: - [Assistant Hooks documentation](https://docs.vapi.ai/assistants/assistant-hooks)