# ZJU-REAL/ClawGUI

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

Repository: https://github.com/ZJU-REAL/ClawGUI
Canonical: https://ross.abutalabs.com/products/clawgui
Language: Python
License: Apache-2.0
License Family: permissive
Topics: agentrl, guiagents, onlinerl, openclaw, mobile-agent, rl-training
Last push: 2026-06-03T07:10:58+00:00

## Health v2 (maintenance only)
Score: 70/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 85, release rhythm 84, longevity 10
- inputs: {"age_days": 148, "days_push": 91, "days_rel": 104, "gap_med": 1.5, "n_releases_24m": 3}
- flags: young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1338, forks 54 (observed 2026-08-28T04:04:25.825086+00:00)

## What it is
ClawGUI is a unified Python framework for GUI agents covering the full lifecycle: online reinforcement learning training (ClawGUI-RL with GiGPO), standardized benchmark evaluation (ClawGUI-Eval), and real-device deployment (ClawGUI-Agent). It ships a 2B model trained end-to-end with the pipeline that outperforms baselines on MobileWorld.

## Use cases
- train a GUI agent with online reinforcement learning
- benchmark and evaluate mobile GUI agents on standardized tasks
- deploy an AI agent that controls a real phone via natural language
- build an end-to-end pipeline for training, evaluating, and deploying screen agents
- run a self-evolving skills agent on device
- compare GUI agent models on mobile world benchmarks

## When to choose
- you need a full-stack research framework covering RL training, evaluation, and deployment of GUI agents
- you want to train mobile/screen agents with online RL rather than only SFT
- you need to deploy agents on real Android devices
- you want a reproducible benchmark setup for GUI agent research

## When to avoid
- you only need a simple web automation bot without model training
- you need a production-ready no-code RPA tool
- you lack GPU resources for RL training of vision-language models
- you target desktop-only GUI automation with no mobile focus

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, machine-learning, reinforcement-learning, llm-training, benchmarking, computer-vision
- domain: artificial-intelligence, reinforcement-learning, machine-learning, mobile-development
- platform: python
- tags: gui-agents, online-rl, gigpo, mobile-agent, device-control, agent-evaluation, research-framework, ai-agents, android, linux, gpu

## Member repositories
- ZJU-REAL/ClawGUI (main) score 70

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:25.825086+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:43:58.836394+00:00, confidence not recorded.
  - readme: https://github.com/ZJU-REAL/ClawGUI (fetched 2026-08-28T04:04:25.825086+00:00, sha 2068d8a3e4e6)
- Data as of 2026-08-30T08:39:29.467469+00:00.