[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-744":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":23,"hasPages":23,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":32,"readmeContent":33,"aiSummary":34,"trendingCount":16,"starSnapshotCount":16,"syncStatus":35,"lastSyncTime":36,"discoverSource":37},744,"ClawGUI","ZJU-REAL\u002FClawGUI","ZJU-REAL","Build, Evaluate, and Deploy GUI Agents — online RL training, standardized benchmarks, and real-device deployment in one framework.","",null,"Python",1281,53,12,1,0,5,9,76,15,18.2,"Apache License 2.0",false,"master",[26,27,28,29,30,31],"agentrl","guiagents","mobile-agent","onlinerl","openclaw","rl-training","2026-06-12 02:00:18","\u003Cdiv align=\"center\">\n\n\u003Cimg src=\"assets\u002FClawGUI-Logo.png\" height=\"140\" alt=\"ClawGUI Logo\">\n\u003Ch1>ClawGUI: A Unified Framework for Training, Evaluating, and Deploying GUI Agents\u003C\u002Fh1>\n\n[![Python 3.12](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPython-3.12-blue.svg)](https:\u002F\u002Fwww.python.org\u002Fdownloads\u002Frelease\u002Fpython-3120\u002F)\n[![License](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-Apache_2.0-green.svg)](https:\u002F\u002Fopensource.org\u002Flicenses\u002FApache-2.0)\n[![Stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002FZJU-REAL\u002FClawGUI?style=social)](https:\u002F\u002Fgithub.com\u002FZJU-REAL\u002FClawGUI\u002Fstargazers)\n[![arXiv](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F📄%20arXiv-2604.11784-b31b1b.svg)](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.11784)\n[![Daily Paper](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F🤗%20Daily-Paper-yellow.svg)](https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2604.11784)\n\n[![HuggingFace Model](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F🤗%20HuggingFace-ClawGUI--2B-yellow.svg)](https:\u002F\u002Fhuggingface.co\u002FSugarVapeur\u002FOpenGUI-2B)\n[![ModelScope Model](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F🤖%20ModelScope-ClawGUI--2B-purple.svg)](https:\u002F\u002Fwww.modelscope.cn\u002Fmodels\u002FSugarFree\u002FOpenGUI-2B)\n[![Project Page](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F🌐%20Project-Page-orange.svg)](https:\u002F\u002Fzju-real.github.io\u002FClawGUI-Page\u002F)\n\n[English](README.md) | [中文](README_zh.md)\n\n\u003C\u002Fdiv>\n\n\n\u003Cdiv align=\"center\">\n\u003Cb>A full-stack framework for GUI agents, covering online RL training, standardized evaluation, and deployment.\u003C\u002Fb>\n\n\u003Ctable>\n\u003Ctr>\n\u003Ctd align=\"center\">\n\u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F38471771-c614-4520-a4a1-0b985546e023\" controls width=\"420\">\u003C\u002Fvideo>\n\u003Cbr>\u003Cb>ClawGUI-Agent controls a real phone\u003Cbr>via natural language\u003C\u002Fb>\n\u003C\u002Ftd>\n\u003Ctd align=\"center\">\n\u003Cvideo src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Fa72f9229-15e4-439e-aeb1-82a05b843fbe\" controls width=\"420\">\u003C\u002Fvideo>\n\u003Cbr>\u003Cb>ClawGUI-RL trains a GUI agent with online\u003Cbr>reinforcement learning\u003C\u002Fb>\n\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\n\n## News\n\n+ 📄 **[2026\u002F4\u002F14]** Our paper is available on arXiv: [ClawGUI: A Unified Framework for Training, Evaluating, and Deploying GUI Agents](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.11784).\n+ 🔥 **[2026\u002F4\u002F13]** ClawGUI is released — train with ClawGUI-RL (GiGPO), evaluate with ClawGUI-Eval, deploy with ClawGUI-Agent. ClawGUI-2B, a 2B agent trained end-to-end with this pipeline, hits **17.1** MobileWorld SR vs. the **11.1** baseline. See [Quick Start](#-quick-start).\n\n## Table of Contents\n\n- [Overview](#-overview)\n- [Architecture](#️-architecture)\n- [Quick Start](#-quick-start)\n  - [ClawGUI-RL — Build](#-clawgui-rl--build)\n  - [ClawGUI-Eval — Evaluate](#-clawgui-eval--evaluate)\n  - [ClawGUI-Agent — Deploy](#-clawgui-agent--deploy)\n- [Roadmap](#️-roadmap)\n- [Acknowledgements](#-acknowledgements)\n- [License](#-license)\n\n\n## 💡 Overview\n\n**ClawGUI** is a research framework for GUI agents, covering the complete lifecycle from **online RL training** and **standardized evaluation** to **real-device deployment**.\n\nBuilding a capable GUI agent involves three tightly coupled problems that are rarely solved together: you need an environment to train the agent online, rigorous benchmarks to measure what it has learned, and a production system to deploy it on real devices. ClawGUI addresses all three.\n\n| Module | Role |\n|--------|------|\n| 🚀 **[ClawGUI-RL](clawgui-rl\u002F)** | **Build** — Train GUI agents online with scalable RL: parallel Docker environments, real Android devices, and GiGPO+PRM for fine-grained step-level rewards |\n| 📊 **[ClawGUI-Eval](clawgui-eval\u002F)** | **Evaluate** — Measure what the agent has learned: 6 benchmarks, 11+ models, 95.8% faithful reproduction of official results |\n| 🤖 **[ClawGUI-Agent](clawgui-agent\u002F)** | **Deploy** — Use GUI agents in the real world: control mobile devices via natural language through 12+ chat platforms, with one-command evaluation built in |\n| 🏆 **ClawGUI-2B** | End-to-end validation: trained entirely with ClawGUI-RL and GiGPO, achieving **17.1** MobileWorld SR vs. the **11.1** baseline |\n\n\n## 🏗️ Architecture\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"assets\u002Fclawgui-framework.png\" width=\"95%\" alt=\"ClawGUI System Architecture\">\n\u003C\u002Fdiv>\n\n\n## 🚀 Quick Start\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FZJU-REAL\u002FClawGUI.git\ncd ClawGUI\n```\n\nEach module is independent with its own environment. Click into each one for full installation and usage instructions.\n\n\n### 🚀 ClawGUI-RL — Build\n\n> 📁 [`clawgui-rl\u002F`](clawgui-rl\u002F) · 📖 [Full Documentation](clawgui-rl\u002FREADME.md)\n\nClawGUI-RL trains GUI agents with online reinforcement learning. It runs dozens of Docker-based Android emulators in parallel or trains directly on physical devices — and replaces standard GRPO with GiGPO+PRM for fine-grained step-level rewards that drive stronger policy learning.\n\n- **Parallel multi-environment** — Dozens of Docker-based virtual Android environments simultaneously\n- **Real-device training** — Physical or cloud Android phones with the same API\n- **GiGPO + PRM** — Fine-grained step-level reward for better policy optimization than standard GRPO\n- **Spare server rotation** — Automatic failover keeps training running without interruption\n- **Episode visualization** — Record and replay any training trajectory\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"clawgui-rl\u002Fassets\u002Fclawgui-rl-framework.png\" width=\"95%\" alt=\"ClawGUI-RL Architecture\">\n\u003C\u002Fdiv>\n\n→ **[Get started with ClawGUI-RL](clawgui-rl\u002FREADME.md)**\n\n\n### 📊 ClawGUI-Eval — Evaluate\n\n> 📁 [`clawgui-eval\u002F`](clawgui-eval\u002F) · 📖 [Full Documentation](clawgui-eval\u002FREADME.md) · [🤗 Dataset](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002Fjohnzqlu\u002Fclawgui-eval) · [🤖 ModelScope](https:\u002F\u002Fmodelscope.cn\u002Fdatasets\u002FMatrix0602\u002Fclawgui-eval)\n\nClawGUI-Eval gives GUI grounding research a reliable measurement baseline. Its three-stage **Infer → Judge → Metric** pipeline covers 6 benchmarks and 11+ models, with a **95.8%** reproduction rate against official results — so numbers across papers are actually comparable.\n\n- **6 benchmarks** — ScreenSpot-Pro, ScreenSpot-V2, UIVision, MMBench-GUI, OSWorld-G, AndroidControl\n- **11+ models** — Qwen3-VL, Qwen2.5-VL, UI-TARS, MAI-UI, GUI-G2, UI-Venus, Gemini, Seed 1.8, and more\n- **Dual backend** — Local GPU (`transformers`) or remote API (OpenAI-compatible)\n- **Multi-GPU & multi-thread** — Parallel inference with automatic resume\n- **ClawGUI-Agent integration** — Pair with ClawGUI-Agent to run the full pipeline via natural language\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"clawgui-eval\u002Fassets\u002Fclawgui-eval-arch.png\" width=\"95%\" alt=\"ClawGUI-Eval Architecture\">\n\u003C\u002Fdiv>\n\n→ **[Get started with ClawGUI-Eval](clawgui-eval\u002FREADME.md)**\n\n\n### 🤖 ClawGUI-Agent — Deploy\n\n> 📁 [`clawgui-agent\u002F`](clawgui-agent\u002F) · 📖 [Full Documentation](clawgui-agent\u002FREADME.md) · [中文](clawgui-agent\u002FREADME_CN.md)\n\nClawGUI-Agent closes the loop from training to production. Built on OpenClaw and powered by nanobot, it lets you control Android, HarmonyOS, or iOS devices with natural language from 12+ chat platforms — and trigger the full ClawGUI-Eval benchmark pipeline with a single sentence, no scripts required.\n\n- **Cross-platform** — Android (ADB), HarmonyOS (HDC), iOS (XCTest)\n- **Multi-model** — AutoGLM, MAI-UI, GUI-Owl, Qwen-VL, UI-TARS via OpenAI-compatible API\n- **One-command evaluation** — Say \"benchmark qwen3vl on screenspot-pro\" and it handles env check → multi-GPU inference → judging → metrics → result comparison\n- **Personalized memory** — Automatically learns user preferences and injects context across tasks\n- **Episode recording** — Every task saved as structured episodes for replay and dataset building\n- **Web UI** — Gradio interface for device management, task execution, and memory inspection\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"clawgui-agent\u002Fassets\u002Fclawgui-agent-logo.png\" width=\"90%\" alt=\"ClawGUI-Agent\">\n\u003C\u002Fdiv>\n\n→ **[Get started with ClawGUI-Agent](clawgui-agent\u002FREADME.md)**\n\n\n## 🎯 Roadmap\n\n- [x] **ClawGUI-Agent** — GUI agent framework for phone control and evaluation via natural language\n- [x] **ClawGUI-RL** — Scalable mobile online RL training infrastructure with GiGPO + PRM\n- [x] **ClawGUI-Eval** — Standardized GUI grounding evaluation suite with 6 benchmarks and 95%+ reproduction rate\n- [x] **ClawGUI-2B** — 2B GUI agent trained with GiGPO, achieving 17.1 MobileWorld SR (vs. 11.1 baseline)\n- [ ] **On-device ClawGUI-Agent** — Deploy ClawGUI-Agent directly on real phones to avoid cloud-based privacy leakage\n- [ ] **Desktop Online RL** — Extend ClawGUI-RL to desktop environments for online reinforcement learning\n- [ ] **Web Online RL** — Extend ClawGUI-RL to web environments for online reinforcement learning\n- [ ] **More Skills for ClawGUI-Agent** — Add more pluggable skills to expand ClawGUI-Agent's capabilities\n- [ ] **Hybrid CLI & GUI Mechanism** — Explore hybrid interaction combining command-line and GUI operations\n- [ ] **Real-time RL** — Integrate real-time reinforcement learning based on the OPD algorithm for ClawGUI-RL and ClawGUI-Agent\n\n\n## 🤝 Contributing\n\nWe welcome contributions of all kinds — new model support, new RL environments, bug fixes, and documentation improvements. See [CONTRIBUTING.md](CONTRIBUTING.md) for how to get started, module-specific guidelines, and PR requirements.\n\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FZJU-REAL\u002FClawGUI\u002Fgraphs\u002Fcontributors\">\n  \u003Cimg src=\"https:\u002F\u002Fcontrib.rocks\u002Fimage?repo=ZJU-REAL\u002FClawGUI\" \u002F>\n\u003C\u002Fa>\n\n\n## 🙏 Acknowledgements\n\nClawGUI is built upon the following excellent open-source projects. We sincerely thank their contributors:\n\n- [**verl-agent**](https:\u002F\u002Fgithub.com\u002Flangfengq\u002Fverl-agent)\n- [**MAI-UI**](https:\u002F\u002Fgithub.com\u002FTongyi-MAI\u002FMAI-UI)\n- [**MobileWorld**](https:\u002F\u002Fgithub.com\u002FTongyi-MAI\u002FMobileWorld)\n- [**Mobile-Agent**](https:\u002F\u002Fgithub.com\u002Fx-plug\u002Fmobileagent)\n- [**nanobot**](https:\u002F\u002Fgithub.com\u002FHKUDS\u002Fnanobot)\n- [**Open-AutoGLM**](https:\u002F\u002Fgithub.com\u002Fzai-org\u002FOpen-AutoGLM)\n\n\n## License\n\nThis project is licensed under the [Apache License 2.0](LICENSE).\n\n\n## 📝 Citation\n\nIf you find ClawGUI useful in your research, please consider citing our paper:\n\n```bibtex\n@article{tang2026clawgui,\n  title={ClawGUI: A Unified Framework for Training, Evaluating, and Deploying GUI Agents},\n  author={Tang, Fei and Lu, Zhiqiong and Zhang, Boxuan and Lu, Weiming and Xiao, Jun and Zhuang, Yueting and Shen, Yongliang},\n  journal={arXiv preprint arXiv:2604.11784},\n  year={2026}\n}\n```\n\n\n## Star History\n\n\u003Ca href=\"https:\u002F\u002Fwww.star-history.com\u002F?repos=ZJU-REAL%2FClawGUI&type=date&legend=top-left\">\n \u003Cpicture>\n   \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\"https:\u002F\u002Fapi.star-history.com\u002Fchart?repos=ZJU-REAL\u002FClawGUI&type=date&theme=dark&legend=top-left\" \u002F>\n   \u003Csource media=\"(prefers-color-scheme: light)\" srcset=\"https:\u002F\u002Fapi.star-history.com\u002Fchart?repos=ZJU-REAL\u002FClawGUI&type=date&legend=top-left\" \u002F>\n   \u003Cimg alt=\"Star History Chart\" src=\"https:\u002F\u002Fapi.star-history.com\u002Fchart?repos=ZJU-REAL\u002FClawGUI&type=date&legend=top-left\" \u002F>\n \u003C\u002Fpicture>\n\u003C\u002Fa>\n","ClawGUI是一个用于构建、评估和部署GUI代理的统一框架，支持在线强化学习训练、标准化基准测试以及真实设备部署。其核心功能包括通过在线强化学习（RL）训练GUI代理、使用标准指标进行性能评估，并能在实际移动设备上运行这些代理。技术特点方面，ClawGUI基于Python开发，提供了一个从训练到部署的全栈解决方案，特别适用于需要通过自然语言控制手机或其他具有图形用户界面的应用场景。此外，它还支持在HuggingFace和ModelScope等平台上分享和获取预训练模型。",2,"2026-06-11 02:39:03","CREATED_QUERY"]