Open-AutoGLM is a phone agent framework that enables AI-powered automation of mobile device interactions through natural language commands.
The tool addresses the challenge of automating tasks on mobile devices by combining multimodal screen understanding with intelligent action planning. It uses Android Debug Bridge to control devices, vision-language models to perceive screen content, and planning capabilities to generate and execute operation sequences. Users describe tasks in natural language—such as "open a shopping app and search for shoes"—and the agent automatically interprets the intent, understands the current interface state, plans the next actions, and executes them. The system includes safeguards for sensitive operations and supports manual takeover during login or verification scenarios. It also provides remote ADB debugging capabilities over WiFi or network connections.
The tool suits developers building mobile automation workflows who want to leverage vision-language models rather than traditional UI automation approaches. It is designed for Android devices and offers models optimized for Chinese mobile applications as well as multilingual variants supporting English and other languages. The framework integrates with Midscene.js, a vision-model-driven UI automation SDK that supports JavaScript and YAML-based workflow syntax across multiple platforms.
Development activity shows consistent engagement with the project through community channels including WeChat and social media. The team maintains multiple model variants tailored to different language and use-case requirements, indicating ongoing refinement of the core capabilities. The project includes active developer incentive programs and encourages community contributions through structured submission processes. Documentation spans both Chinese and English, reflecting a focus on accessibility across language communities.
