mobilerun

, so the idea is to provide the groundwork/library for the hard parts, so developers don't have to repeat these steps: Parse Android UI structure in an LLM-friendly way (identifying elements, states, descriptions). Provide robust primitives for interacting with any UI element (taps, swipes, text input). Create reusable components for mobile agent perception and action. The vision: create repeatable tasks across Android apps just by giving your agent a goal, without worrying about the specific UI steps. To better showcase the power of direct UI interaction, I made a few demos: Droidrun explainer: Video Link: https://x.com/droid_run/status/191244635898166087 Droidrun posts 'Hello world' on X: Prompt Idea: Open X -> Press post -> New Post -> Insert 'Hello world' -> Post Video Link: https://x.com/droid_run/status/1910714055251157141 Droidrun finds cat videos on TikTok & comments: Video Link: https://x.com/droid_run/status/1911769757956665478 I’d like feedback on how this tool fits into your mobile automation or AI agent projects. Try it out (GitHub: https://github.com/droidrun/droidrun) and let us know how it performs on your end. Our website if you need more information: droidrun.ai We are a team of four members and just open-sourced this after getting great initial interest (900+ waitlist signups after 72 hours!).