Roubao - An open-source AI-powered mobile automation assistant, a budget-friendly alternative to Doubao mobile phones.

What is a meat bun?

Roubao is an open-source AI-powered mobile automation assistant, serving as a free alternative to ByteDance's "Doubao Mobile Assistant." Based on a Visual Language Model (VLM) and developed natively in Kotlin, Roubao requires no computer or root access and supports system-level control via Shizuku. Roubao's core architecture employs a "screenshot-AI analysis-execution" loop, supporting both Delegation (delegating to AI applications) and GUI automation modes, enabling tasks such as ordering takeout, sending messages, and navigation.

The main function of meat buns

• Intelligent AI AgentRoubao can understand screen content based on a visual language model, operate the phone through natural language commands, and automatically plan and execute complex tasks.
• Exquisite UI designRoubao uses a meticulously designed Material 3 interface, supports adaptive dark and light themes, provides smooth animation effects, and offers full bilingual support in Chinese and English.
• Highly customizableRoubao supports multiple visual language models, including Alibaba Cloud Tongyi Qianwen, OpenAI GPT-4V, Claude, etc., and supports users to configure custom API endpoints and local models.
• Security protection mechanismMeatball implements AES-256-GCM encrypted storage of API keys, has an automatic detection and stop mechanism for sensitive pages, and allows for full visibility of task execution and manual interruption at any time.
• Root Mode ExtensionMeatball supports Root mode extensions. When Shizuku is run with Root privileges, more system-level operation capabilities can be unlocked.

The technical principle of meat buns

• Two-layer Agent ArchitectureThe core of Roubao is a two-layer agent architecture, inspired by Claude Code, divided into a Tools layer and a Skills layer. The Tools layer encapsulates atomic capabilities such as application search, DeepLink navigation, clipboard operations, and other basic functions; the Skills layer addresses user intent, mapping natural language to specific tasks to achieve a complete link from "ordering takeout" to executing the operation.
• Cyclic Automated ProcessMeat Bun employs a cyclical automated process of screenshot-analysis-execution, using Shizuku to invoke system-level commands to complete screen capture and simulated clicks. When a user command is received, SkillManager first performs intent recognition. High-confidence tasks are directly delegated to professional AI applications for processing via DeepLink, while low-confidence or complex tasks enter a standard Agent loop, where the Manager plans, the Executor makes decisions, and the Reflector reflects until completion.
• Native Android implementationRoubao rewrote the MobileAgent framework in Kotlin, running entirely locally on the Android device without the need for a computer. It obtains ADB-level permissions through Shizuku, enabling direct execution of actions such as screenshots, clicks, and swipes, avoiding the cumbersome process of connecting to a computer required by traditional Python solutions.

The address of the meat bun project

• GitHub repositoryhttps://github.com/Turbo1123/roubao

Application scenarios of meat buns

• Daily task automationRoubao can help users perform everyday tasks such as ordering takeout, checking the weather, setting alarms, and sending messages. Simply state your needs and the task will be completed automatically without having to manually search for the app.
• Cross-application information processingRoubao can automatically take screenshots and recognize content, helping users send the last photo to Weibo, share links to WeChat, or copy and paste information between different applications.
• Entertainment content acquisitionRoubao can automatically open music apps to play daily recommendations, search and play specific videos on Bilibili, or help users browse short videos and watch popular content.
• Navigation and travel servicesRoubao supports directly waking up Gaode Maps to search for routes, help users hail a taxi, or find the location information of nearby restaurants and shops.