AndroidClaw AI Agent

AndroidClaw is an intelligent Android automation assistant powered by Large Language Models (LLMs) such as Google Gemini, OpenAI GPT, and local Ollama instances. Unlike traditional automation tools, AndroidClaw "understands" your natural language commands, "perceives" your screen content, and "acts" by making logical decisions to complete tasks.

🚀 Key Features

• Multimodal AI Integration: Supports Gemini , OpenAI API, and local LLMs via Ollama ，I only tested Gemini.
• Screen Perception: Leverages Android Accessibility Services to "read" the UI hierarchy and understand the current context.
• Three-Tier Action Execution:
  • Intent: Fast, direct system calls for alarms, URLs, and settings.
  • Click: Precise UI interaction by simulating taps on specific coordinates.
  • Shell (Root): High-level system control via SU commands for advanced automation.
• Recursive Reasoning (Agentic Loop): Remembers previous steps to complete complex, multi-stage tasks (e.g., "Go to WeChat and clear my cache").
• Safety First:
  • Global Emergency Stop: A red floating button that stays on top of all apps to instantly kill the AI process.
  • Human-in-the-Loop: Confirmation dialogs for all sensitive Click and Shell operations.
• Voice-Enabled: Integrated Speech-to-Text (STT) for hands-free commanding.
• Modern UI: Built entirely with Jetpack Compose for a smooth, responsive chat-like experience.

🛠 How It Works

1. User Command: You provide a task via text or voice (e.g., "Set an alarm for 8 AM" or "Open YouTube").
2. Screen Analysis: The app captures the current UI Tree (Accessibility Data).
3. AI Reasoning: The prompt (containing your goal + screen data + history) is sent to the LLM.
4. Action Selection:
   • If it's a simple task, the AI returns an Intent.
   • If it's complex, the AI returns a Click or Shell command.
5. Closed-Loop Execution: After a click, the app waits for the screen to refresh, re-scans the UI, and asks the AI for the next step until the task is marked as "finished."

📋 Prerequisites

• Android Version: Android 8.0 (API 26) or higher.
• Accessibility Service: Must be manually enabled in Settings > Accessibility for screen perception.
• Overlay Permission: Required for the Emergency Stop button.
• Root Access (Optional): Only required if you intend to use sh (Shell) commands.
• API Key: A valid key from Google AI Studio (Gemini) or OpenAI.

⚙️ Configuration

Open the Settings dialog in the app to configure your AI provider:

Field	Description
Provider	Choose between Gemini, OpenAI, or Ollama.
Base URL	The API endpoint (e.g., https://generativelanguage.googleapis.com/v1beta).
API Key	Your private secret key.
Model	The model name (e.g., gemini-3-flash-preview or gpt-4o). For gemini, you only need to enter the URL

Note for Ollama users: Use http://10.0.2.2:11434/v1 as the Base URL if running on an Android Emulator.

⚠️ Safety & Privacy

• Privacy: This app sends your screen's UI text data to the selected AI provider. Avoid using the Agent on screens containing sensitive information (passwords, banking).
• Control: The AI cannot perform a Click or Shell action without your explicit permission via the on-screen confirmation dialog.
• Emergency Stop: If the AI begins performing unexpected actions, tap the STOP floating button immediately to freeze all processes.

🏗 Installation

1. Clone the repository.
2. Open the project in Android Studio (Ladybug or newer).
3. Build and install the APK on your device.
4. Grant Accessibility and Display over other apps permissions.

---

Disclaimer

This project is for educational and research purposes only. The authors are not responsible for any data loss or device damage caused by the AI's autonomous actions.