bytebot-ai/bytebot ? reverse-engineered prompt

Build me an AI agent that can control its own computer to automate tasks for me. The idea is that it should have its own self-hosted, virtual Linux desktop environment that runs in Docker. I want a simple web interface where I can give it instructions in plain English, like "research the best coffee makers and create a comparison document."

I should be able to watch a live stream of the agent's desktop to see it work—moving the mouse, typing, and opening apps. It needs its own filesystem to download and process files, especially PDFs that I can upload. It's important that it's not just a browser agent; it should be able to use any desktop app like a text editor or email client. To handle logins, it should be able to use a password manager. The backend can use TypeScript, and you can build the UI in Next.js. Make it flexible so I can plug in my own API key from OpenAI, Anthropic, or Google Gemini.