I got bored one day and built an AI that controls my PC like a human. Just launched it on Product Hunt today.
A while back I had nothing to do, so I started wondering: could an AI actually use a computer the way a person does? Not just answer questions in a chat window, but see the screen, move the mouse, click, type, and get real things done.
That curiosity turned into Rosply. It takes a screenshot, overlays a coordinate grid so the AI can read exact pixel positions, sends that to a vision model (OpenRouter, local Ollama, or even Claude Code as the brain), and executes the action it decides on. Then it loops.
The hardest part wasn't getting the AI to understand a task. It was getting it to recover gracefully when something goes wrong on screen, popups, UI changes, dead ends. Most of the engineering effort went into making it resilient enough to not get stuck in a loop.
It's source-available, runs fully local if you want (your own API keys, no telemetry, no accounts), and works on Windows/Mac/Linux.
Launched on Product Hunt today: https://www.producthunt.com/products/rosply
Would love feedback from this community, especially on the business model side (source-available + one-time purchase vs SaaS) since that's something I'm still figuring out.
Reading this, I found myself less focused on the pricing model and more on what exactly people end up believing they're buying.
The interesting part is that the same product description could plausibly be interpreted in several different ways, and those interpretations don't necessarily point toward the same business model.
That's the part I'd be most curious about before drawing conclusions from pricing experiments.