Tazo is a model that takes live call audio and a list of options, then returns a probability for each one. It lets voice agents decide the next step directly from the audio instead of from a transcript.