X-PLUG/MobileAgent

Implements multimodal vision-language models (GUI-Owl series: 2B-235B parameters) optimized for GUI perception and grounding across desktop, mobile, and browser environments using Qwen3-VL backbone. The agentic framework layers planning, reflection, memory management, and tool/MCP calling on top of vision capabilities, enabling end-to-end task automation across platforms. Achieves state-of-the-art on 20+ GUI benchmarks including OSWorld and AndroidWorld through semi-online RL fine-tuning and native multi-platform support.

8,242 stars. Actively maintained with 25 commits in the last 30 days.