lspahija/AIUI — reverse-engineered prompt

Build me a simple browser app for natural voice conversations with AI. I want it to work on desktop and mobile browsers, where the user can open the page, start talking right away, have their speech turned into text, send that to an AI model, and hear the answer spoken back so it feels like an ongoing back and forth conversation.

Please make it support OpenAI with a default model that is affordable, but let me switch to a stronger model through environment settings. I also want text to speech voice options, language selection with English as the default, and a clean interface that feels easy to use without typing. It should be something I can run locally in Docker with an API key and a few env vars, then open in the browser on localhost.

Keep the setup straightforward and make sure the app is usable end to end. If anything is unclear, look up current docs online and choose sensible defaults.