# OpenGUI

An Android GUI agent framework that lets an AI agent see, plan and operate real Android app interfaces on a physical device, returning structured results for each action.

OpenGUI is a mobile GUI agent framework for Android. It lets an AI agent read a real Android app's user interface, plan its next step, take mobile actions such as taps and gestures, and return structured results, so an assistant can operate phone apps rather than only reason about them.

It is designed to be installed and run as a plugin inside DeepSeek Harness: an installer skill downloads the latest verified release, installs a matching runtime automatically, and starts the host application, while preserving unrelated plugins, settings, workspaces, model settings, credentials and phone authorizations. It supports several recent host releases and, depending on the platform, installs into an isolated managed runtime or reuses one already on the system path; it refuses to downgrade past a credential-store format change rather than risk corrupting existing state. Once installed, a user connects and authorizes an Android phone and can then ask the agent, for example, to open Settings and report the Android version.

It is written in TypeScript and available in English, Simplified Chinese and Japanese.

Website: https://opengui.ai/