Summary

Roubao offers a native Android AI assistant for phone automation, eliminating PC dependency. It uses Vision-Language Models (VLMs) and Shizuku for on-device task execution, targeting users seeking accessible automation and developers needing a flexible framework.

How It Works

Built in Kotlin, Roubao runs natively on Android, bypassing PC-based Python/ADB setups. It utilizes Shizuku for system-level permissions (screen capture, input simulation) without root. The core is a "Tools + Skills" agent architecture: Tools are atomic operations, Skills map natural language intents. Execution uses "Delegation" (deep-linking to AI apps) or "GUI Automation" (screenshot-analysis-action loop), powered by VLMs.

Quick Start & Requirements

• Prerequisites: Android 8.0+, WiFi, Shizuku installed/running, VLM API Key.
• Installation: Install Shizuku, install Roubao APK, authorize Roubao in Shizuku, configure VLM API Key.
• Links: Shizuku (Google Play, GitHub Releases).

Highlighted Details

• Native Android: Runs entirely on-device, no PC required.
• Shizuku Integration: System-level control without root.
• Flexible VLM Support: Integrates multiple VLMs (Qwen-VL, GPT-4V, Claude) and custom endpoints.
• UI: Modern Material 3 design, bilingual.
• Security: Encrypted API keys, sensitive page detection.
• Root Mode: Optional enhanced system operations.

Maintenance & Community

Active development focuses on v2.0 (AccessibilityService, UI tree, scripting). Contributions via GitHub Issues/PRs.

Licensing & Compatibility

MIT License permits broad commercial use. Users manage VLM API keys.

Limitations & Caveats

Under active development (v2.0), potential instability. Relies on external VLM APIs and internet connectivity. Root mode requires specific device setup.