exotel/voicebot-quick-starter ? reverse-engineered prompt

Build me a simple voice AI agent that I can talk to in my browser and also over a phone call through Exotel.

I want it to listen to speech, turn it into text, send it to Gemini for a reply, and speak the answer back with a natural sounding voice. Please make the browser version easy to run locally, and also add the Exotel version that can handle live phone audio through a websocket connection. Use Google Cloud for speech to text, Gemini for the brain, and Google Cloud text to speech for the voice.

Keep the setup straightforward with environment variables for credentials, and make it easy to change the agent’s personality in one place. If there are any current docs or setup details that would help, look them up online and use the latest approach. I’d also like basic metrics or logs so I can see that the pipeline is working.