Droidrun

Droidrun is an open-source mobile agent framework for controlling Android and iOS devices with LLM agents. You describe a task in plain language through natural language commands, and the agent reads the screen, then taps, swipes and types its way through any app until the job is done. It ships as a Python package and a command line tool, so you can run it against a phone on your desk or wire it into a bigger automation pipeline.

The project grew out of a simple problem. Traditional mobile automation depends on hardcoded UI selectors that break every time an app updates its layout. Droidrun throws that approach out and hands the decisions to a model, which adapts to whatever the screen shows. The same core powers a hosted service called Mobilerun Cloud, where you rent cloud phones and run agents without keeping a device on your network.

The limit is that it's a developer tool. There's no point-and-click app builder for non-technical users, and you need to be comfortable with a terminal, Python or an API. Local runs also expect a real device or an emulator set up with ADB, and the framework itself doesn't provide the hardware. Getting real value out of it means writing code or leaning on the cloud tier.

Getting Started

1. Install the package with uv tool install mobilerun or pip install mobilerun, using Python 3.11 through 3.13.
2. Run mobilerun setup to install the Portal app on an Android device and enable its accessibility service, or use the iOS Portal flow on an iPhone.
3. Run mobilerun configure to pick a provider, choose an auth method and select a model.
4. Send your first task with mobilerun run "Open settings and turn on dark mode".
5. Add --vision for screenshot analysis or --reasoning for longer multi-step jobs, then fold the same calls into a Python script or the cloud API.