microsoft/magentic-ui ? reverse-engineered prompt

Build me a local web app called MagenticLite that lets a user give an AI assistant real tasks to do across the browser and their local files. I want it to feel like a simple control center where I can type a goal, watch the agent work, and step in whenever it needs approval before doing anything risky. It should be able to browse websites, fill out forms, look up information, and organize files on my computer, all while keeping the user in control.

Make it run locally with a clean onboarding flow so I can connect a model endpoint, start a task, and see what the agent is thinking and doing. Keep it safe by using a sandboxed browser session so it does not touch the rest of my machine unless I allow it. If you need to look up current docs online, do that too. I want the finished project to be easy to run from the repo and feel polished enough to demo.