roubao

What it is

Android Automation Tool Based on Vision-Language Models. Native Kotlin app for AI mobile automation with a multi-agent architecture, featuring Tools (atomic capabilities), Skills (user intents), and an Agent loop that integrates with VLMs.

How it works

• Core architecture consists of Tools (search_apps, open_app, deep_link, clipboard, shell, http), Skills (intent mapping), and Agents (Manager, Executor, Reflector, Notetaker).
• Uses Shizuku for system-level actions (screencap, input tap, etc.) and supports Delegation via DeepLink or GUI automation when needed.
• VLM Client handles API calls to VLM providers (e.g., Aliyun, OpenAI, Claude).
• Configurable API endpoints and model lists are fetched dynamically; supports local models (Ollama, vLLM).

Getting started

• Pre-requisites: Android 8.0 (API 26) or higher; WiFi; Shizuku; VLM API Key.
• Installation steps include installing Shizuku and the Roubao APK from Releases, then authorizing Shizuku and configuring API Key in Settings.
• Shizuku setup can be via Wireless debugging or ADB as described in the README.
• Example command usage is captured in usage notes under Usage and Examples.

Recent releases

• V1.4.2 (2025-12-18): 更新日志 v1.4.2. Fix ANR during app startup, precompile regex, run app scanning in background, fix fetchModels URL crash, and set default theme to follow system.
• V1.4.1 (2025-12-17): Bug fixes for settings page model list crash due to bad Base URL and improved error messaging.
• V1.4.0 (2025-12-14): New features for independent configuration per API provider (Aliyun, OpenAI, OpenRouter, custom), automatic loading of corresponding configs when switching providers, and UI/UX improvements for app opening actions.
• V1.3.0 (2025-12-12): Root mode support when Shizuku runs with root, display current model during tasks, immediate stop button after command, and navigation to details after stopping.
• 1.1.0 (2025-12-11): Settings for maximum execution steps (5-100, default 25) and custom API Base URL.

Traction

• Stars: 2302
• Forks: 238
• Open issues: 38

Behind the repo

• Repository owner: Turbo1123
• License: MIT
• Language: Kotlin
• Created: 2025-12-09
• Last push: 2026-01-08
• Topics include ai, ai-agents, android, android-automation, kotlin, mobile-agents, mobileagent, phone-automation

Caveats

• License: MIT (as stated in README/Licenses).
• Age and activity indicators are shown by last_push date (2026-01-08) and releases through late 2025.
• Requires Shizuku for system-level permissions; setup described in Getting started.