We have an inbound voice agent named Fred for a real estate business which handles leads. For every call, Fred asks qualifying questions and once all questions are answered, he says something along the lines of "Your information has been stored and a human agent will follow up shortly." Now, according to our requirements, we want the call to end immediately after this ending line is said. However, the output is not deterministic. Sometimes, it ends correctly, other times it doesn't and keeps the line up and after a few seconds it asks the caller if they are still there. We want this to be deterministic so that the call ends as soon as the final line is delivered. Any ideas?

Hi, Could you please check the end call tool configuration on your end to ensure it has been set up correctly? You can refer to the documentation here for guidance: https://docs.vapi.ai/tools/default-tools#end-call If the issue still persists after verifying the configuration, please share the Call ID so we can review the call logs and assist you further. Best regards, Oshi Raghav Customer Support Vapi

Hi, thanks for getting back to me. I have revisited the configuration, and it all seems good and in accordance with the documentation. Here is a call ID: 019cddcb-5305-7000-9ede-1ca82ada4888 Just to reiterate, it is not an issue with a specific call. Sometimes it triggers correctly while other times it doesn't, which is making it difficult for us to track down the root cause. Our best guess is a problem with hooks, but we are not sure.

Hi, We investigated the call behavior and identified the root cause. The

endCallPhrases

list in your configuration is currently empty, which means the call relies entirely on the LLM to trigger the

endCall

function. This behavior is non-deterministic, so sometimes the model triggers it correctly and sometimes it does not. When it fails to do so, the idle message (“Are you still there?”) is triggered after 10 seconds. Recommended fixes: 1. Add your closing phrases (for example, “A human agent will follow up shortly”) to the

endCallPhrases

field in your assistant configuration. This allows Vapi’s deterministic phrase-matching engine to end the call reliably instead of depending on the LLM. 2. Increase

idleTimeoutSeconds

from 10 to around 20–25 seconds to prevent premature idle prompts. Please try these changes and let us know if the issue persists. We’ll be happy to assist further if needed. Thank you.

Thanks so much, it has worked.