Most text based bots are frustrating to talk to out loud. The awkward pauses ruin the flow of conversation and they struggle if you do not speak perfectly. Building a truly natural voice assistant means moving away from standard prompt completions and embracing real time audio streaming. In this talk I will show you how to use the Gemini Live API to build agents that actually listen and respond instantly.

You will leave this session knowing exactly how to build these voice experiences yourself. We will start by setting up the WebSockets needed for fast audio streaming. Then we will look at how to connect external tools so your agent can fetch data and take actions while speaking.