Solo-built vision AI agent Rosply lands first sale (€28) two weeks after launch
A self-taught , with no team and no funding, built an AI agent called Rosply. Most automation and s are limited to whatever a platform's API exposes, so Rosply skips APIs entirely: it takes a screenshot of the screen, a vision model decides what action to take, then it clicks and types just like a person would.
That means it can work with any program on screen, even ones without an API. The stack includes a Python agent core, a vision model accessed through OpenRouter, an Electron/pywebview interface, a system, and an MCP server that integrates with Claude Code.
The first sale came in two weeks after launch, totaling €28 in revenue so far. The developer learned that getting that first sale took longer than expected, since most early traffic was driven by curiosity rather than actual , and is now working on making the product's value clearer upfront instead of relying mainly on the demo.
Key points
- Vision-based AI agent controls any on-screen app via screenshots and simulated clicks/typing, no API needed
- Stack: Python agent core + vision model via OpenRouter + Electron/pywebview UI
- Includes an MCP server for Claude Code integration
- First sale landed two weeks post-launch, €28 so far
- Lesson: early traffic was mostly curiosity, not conversion — now clarifying the upfront