bytedance/ui-tars-desktop ? reverse-engineered prompt

Build me a desktop app for a multimodal AI assistant that can actually use my computer and browser like a person would.

I want a clean chat style interface where I can tell it to do things like open apps, click around websites, fill forms, search the web, and help with everyday tasks. It should support both a local computer operator and a remote operator, so I can control my own machine or connect to another one when needed. If possible, also include a simple command line mode for power users and make it easy to plug in different AI models and tools.

Please make it feel polished and easy to use, with clear status updates while it is working, and a way to see what actions it took. If you need current docs for any model or browser automation pieces, look them up online.