nitkots/DESK ? reverse-engineered prompt

Build me a local app that lets me use ordinary objects on my desk as computer controls, like a can for game movement, a bottle for a slider, or a cup to act like a button.

I want an overhead camera view where I can start tracking objects, click a little mascot named Cal, and talk or type what I want that object to do. The app should understand the request, confirm unclear cases, and then save that mapping so the same object keeps working until I reset it. It should support things like keyboard input, scrolling, opening apps, brightness control, and a simple DJ style control mode if available.

Keep the vision running locally, make the whole thing work in a browser, and use my own API keys for speech and planning. If you need current setup details for any model or dependency, look them up online.