OMG-Agent

Open-sourced Mobile GUI Agent
Open-source, universal Mobile GUI Agent framework

English | 中文

OMG-Agent is an open-source Mobile GUI Agent desktop client that drives AI to automatically operate Android phones via natural language instructions.

Core Features:

• Supports Mobile GUI models like AutoGLM and GELab-Zero
• ADB Real-time Screenshot + AI Task Execution
• Bilingual Interface (English/Chinese), Dark/Light Themes
• Supports OpenAI-compatible API
• Supports Android Emulators

⚠️ Disclaimer

This project is for learning, research, and technical exploration only. Strictly prohibited for any commercial use. When using this tool, please comply with relevant laws and regulations, as well as the terms of use and service agreements of mobile phone manufacturers and applications. Users are solely responsible for any actions and consequences arising from the use of this project, which are unrelated to this project and its developers.

# Install ADB
scoop install adb  # Windows
brew install android-platform-tools  # macOS
apt install adb  # Linux (Ubuntu)

git clone https://github.com/safphere/OMG-Agent.git
cd OMG-Agent
pip install -r requirements.txt
python run.py

1. Enable "Developer Options" and "USB Debugging".
2. Install ADBKeyboard.
3. Connect phone via USB and allow debugging authorization.

1. Click "Refresh Devices".
2. Click "Start Screen".
3. Enter task (e.g., "Open WeChat and send a message to John").
4. Click "Execute".

These models are specifically trained for Mobile GUI tasks and are recommended.

For users without Android phones, refer to the Emulator Setup Guide.

• Environment Setup
• Model Configuration
• Development Guide

OMG-Agent/
├── omg_agent/
│   ├── gui/           # GUI Interface
│   └── core/
│       ├── agent/     # AI Agent Core
│       └── config.py  # Config Management
├── assets/            # Assets
├── docs/              # Documentation
└── run.py             # Entry Point

Safphere is an open-source community composed of algorithm engineers and university geeks, focusing on technical exploration and knowledge sharing in the AI field.

⭐ If you find this project helpful, please Star us!

This project is licensed under the Apache License 2.0 with Commons Clause.

• ✅ Learning, research, and personal use allowed
• ✅ Modification and secondary development allowed
• ❌ Commercial use prohibited
• ⚠️ Please credit the source when using

See LICENSE for details.

• Open-AutoGLM — Zhipu AI Mobile GUI Model
• gelab-zero — StepFun Mobile Agent Framework
• ADBKeyboard — ADB Input Method

© 2025 Safphere