ZJU-REAL

ZJU-REAL / ClawGUI

Public

Build, Evaluate, and Deploy GUI Agents — online RL training, standardized benchmarks, and real-device deployment in one framework.

agentrl guiagents mobile-agent onlinerl openclaw
44
0
100% credibility
Found Apr 13, 2026 at 45 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

ClawGUI is a research framework for training AI agents to control mobile devices via natural language, evaluating their performance, and deploying them for real-world use.

How It Works

1
🔍 Discover ClawGUI

You stumble upon ClawGUI online and watch demo videos of an AI smoothly controlling a real phone just by chatting in everyday language.

2
📱 Connect Your Phone

Plug in your Android phone with USB debugging on, and the setup recognizes it instantly, ready for action.

3
🤖 Link an AI Brain

Connect a smart AI service so your assistant can understand screens and decide what to do next.

4
🚀 Launch Your Assistant

Start the web dashboard or chat interface with one click, and your personal phone helper comes alive.

5
💬 Give Commands

Type simple instructions like 'Open WeChat and message my friend I'm running late' – watch it happen step by step.

6
📹 See It in Action

Live screenshots show the AI tapping, swiping, and typing exactly as needed, completing your task perfectly.

🎉 Phone Assistant Ready

Now your AI handles phone chores anytime via chat, freeing you from manual fiddling – success!

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 45 to 44 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

ClawGUI

ClawGUI delivers a full pipeline to build, evaluate, and deploy GUI agents that control real phones via natural language. Train models online with RL on parallel Docker Android envs or physical devices using fine-grained step rewards; evaluate across 6 standardized benchmarks like ScreenSpot-Pro and OSWorld-G with 11+ VLMs and 95.8% official repro rate; deploy via ClawGUI-Agent on Android, HarmonyOS, or iOS through 12+ chat platforms, complete with one-command evals, personalized memory, and Gradio web UI.

Standout: End-to-end validation via ClawGUI-2B, a 2B model hitting 17.1 MobileWorld success rate (vs 11.1 baseline). Design, build, evaluate, and iterate on LLM agents without stitching disparate tools—rare in GUI agent kits. Integrates ClawGUI-Eval for quick policy checks or hybrid RAG setups on docs.

Target: ML researchers prototyping mobile GUI agents, or devs building production phone automation (e.g., ClawGUI-Agent for Clawgey-style bots). Python-based, cross-platform device support shines for real-device testing.

Low 44 stars and 1.0% credibility signal early-stage: docs solid but expect rough edges in RL scaling or iOS quirks. Solid for experiments, monitor for maturity.

Verdict: Grab if you're serious about GUI agents—unique full-loop beats fragmented alternatives. Worth starring for benchmarks/deploy alone.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.