Native computer-use app for driving autonomous work

I’ve built Vantage, an open-source Windows 11 app where you type a task in a chat box and an AI model carries it out on your actual desktop: it looks at the screen, clicks and types, reads windows through Windows UI Automation, and can run PowerShell. In one test, a single prompt had it open seven Wikipedia articles in Edge, read figures from each infobox, convert them in Calculator’s Length converter, save a file per mountain in Notepad, read system details from Settings, and write a sorted summary, which took 80 steps and about 8 minutes with Gemini 3.8 Flash. I focused on making it read rather than guess, so it reads the window’s actual text instead of relying on screenshots and won’t type a number it hasn’t read. It also does tasks the way you ask, so “use Calculator” means Calculator, and it doesn’t touch your open browser tabs or your clipboard. You bring your own API key for any OpenAI-compatible provider whose model accepts images, and everything is stored locally with API keys encrypted by Windows DPAPI. It’s Windows only and still early, it acts without asking before each step, and how well it does depends a lot on the model, so I’d love to hear which tasks it fails on. The demo, source and installer are linked in the comments.