mobile-use

Why we included this project

This is an open-source agent that runs on a real Android or iOS device and carries out natural-language instructions by interacting with the actual UI. Instead of guessing from screenshots, it reads the app's accessibility tree to locate the right buttons and menus, which is how it can open apps, send messages, and move through multi-step screens. The practical payoff is mainly for QA engineers who want UI automation on real hardware, and for anyone who needs to pull structured data out of an app in a repeatable way. It's model-agnostic, so you can wire it to OpenAI, Google, xAI, OpenRouter, or a local provider and keep the cost and privacy tradeoffs in your own hands. One honest caveat from the maintainers is that games work poorly because they don't expose accessibility data, a reminder that this is grounded in real device behavior rather than a demo.