FatihMakes/Mark-XLVIII — reverse-engineered prompt

Build me a cross platform desktop AI assistant, basically a real JARVIS for Windows, Mac, and Linux. I want to talk to it naturally by voice, but also be able to type when I want. It should listen in real time, answer quickly, remember my preferences and past context, and be able to control my computer, like opening apps, managing windows, changing volume or brightness, handling browser actions, and helping with files and coding.

It should also be able to see my screen and webcam when I ask, give a useful morning briefing with time, weather, and real news, do web searches, set reminders, and show a simple HUD with logs, status, and an interrupt button. Pressing Escape or clicking interrupt should stop it immediately and make it listen again right away.

Please include first run setup for the API key, make it feel polished, avoid random terminal popups, and handle reconnects and errors gracefully. Look up current Gemini Live docs online if you need to.