mobile-use
An open-source AI agent that controls a real Android or iOS device with natural-language commands, tapping through app interfaces the way a person would.
mobile-use is an open-source AI agent that automates a real Android or iOS device using natural-language instructions, from sending a message to navigating a multi-step task inside an unfamiliar app. It understands accessibility-tree data to navigate interfaces, which gives it limited effectiveness in games that do not expose that data, and it can also scrape information out of an app and structure it into a format such as JSON from a plain-language description.
It reports being the first project to complete 100 percent of a well-known Android automation benchmark, and it supports configuring different large language models for the agents that power it, including several major providers as well as other OpenAI-compatible or local providers.
It is written in Python, built on LangGraph, and released under the Apache License 2.0.