Mobile-Agent
Mobile-Agent is a family of GUI agents from Alibaba's Tongyi Lab that automate tasks on mobile, desktop, and web platforms using visual perception and multi-agent collaboration. It solves the problem of automating complex cross-platform GUI interactions by combining planning, grounding, and execution capabilities. It is for developers and researchers building AI-driven automation for phones, PCs, and browsers.
✨ Key features
- Multi-platform support: mobile, desktop, and browser automation.
- Multi-agent framework with planning, reflection, and memory.
- GUI perception and grounding via GUI-Owl models.
- End-to-end operation and tool/MCP calling.
- SOTA results on 20+ GUI benchmarks.
- Open-source models and code for research.
🎯 Use cases
- Automating mobile app tasks like searching flights and comparing prices.
- Creating and editing documents in desktop applications like WPS Office.
- Performing web research and data extraction across browsers.
- Building cross-platform AI assistants for complex workflows.
- Evaluating GUI agents on benchmarks like OSWorld and AndroidWorld.
❓ FAQ
What platforms does Mobile-Agent support?
Mobile-Agent supports mobile, desktop, and browser platforms, as described in the GUI-Owl-1.5 release.
How can I try Mobile-Agent online?
You can try Mobile-Agent-v3.5 via the ModelScope online demo or Bailian online demo, both linked in the README.
Are the models open-source?
Yes, model weights are available on HuggingFace, including GUI-Owl-7B and GUI-Owl-32B.
What is the relationship between Mobile-Agent and GUI-Owl?
GUI-Owl is a GUI vision-language model, and Mobile-Agent-v3 is a multi-agent framework built on GUI-Owl.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.