MindFrame 🤖

MindFrame is a powerful Android-based AI assistant designed to navigate and control your device using natural language. It combines the reasoning capabilities of Large Language Models (LLMs) via Ollama with Android's Accessibility Services to "see" and "interact" with the mobile UI.

🚀 Features

• Multi-Turn Sessions: The agent maintains a continuous conversation history. It remembers previous tasks and screen states within a session, allowing for complex, multi-step dialogue.
• Memory Management: Includes a "Memory Aware" toggle in settings to control context persistence:
  • ON: Continuous memory across multiple tasks (default).
  • OFF: Fresh LLM context for every new task while preserving the UI chat log.
• Reasoning-Action-Verification Loop: The agent observes the current screen, thinks about the next step, executes an action (click, type, scroll), and verifies the result before proceeding.
• Self-Learning Skills: Successful task execution logs are automatically synthesized into reusable SKILL.md recipes, allowing the agent to learn and repeat complex workflows.
• Privacy-First Design: Includes a built-in privacy system that automatically blinds the agent when sensitive financial or banking applications are in focus.
• Multi-Provider Architecture: Seamlessly switch between LLMs via Ollama and OpenRouter from the settings menu.
• Premium Floating Overlay: A glassmorphic chat interface (Android 16+ blur) with dedicated controls for:
  • + New Session: Wipes agent memory and clears the chat.
  • - Minimize: Collapses to a floating bubble.
  • X Exit: Completely stops the service and cleans up memory.

🛠️ Architecture

• AgentOrchestrator: The central brain managing the multi-turn session lifecycle.
• AccessibilityDriver: The eyes and hands, capturing a compacted UI tree and performing physical gestures.
• SkillRegistry: A collection of low-level tools (Search Web, Set Alarm, Type, Click) and high-level recipes.
• OverlayService: The foreground service managing the floating UI and modern system insets.

📋 Prerequisites

• Android 16+ (SDK 36+) - Uses the latest Window Insets and Blur APIs.
• LLM Provider: Either an Ollama or OpenRouter API key.
• Permissions: Accessibility, Overlay (System Alert Window), and POST_NOTIFICATIONS.

🤖 CI/CD

This project uses GitHub Actions for Continuous Integration. On every push to main:

• The project is built using JDK 17.
• A Debug APK is generated and uploaded as a build artifact.
• You can download the latest successful build directly from the "Actions" tab in the GitHub repository.

⚙️ Setup

1. Clone the repository.
2. Build and install the APK using Android Studio.
3. Grant Permissions: The app will guide you through granting Accessibility, Overlay, and Battery Optimization permissions on first launch.
4. Configure LLM:
   • Open Settings from the floating overlay.
   • Tap Active LLM Provider to choose between Ollama API or OpenRouter API.
   • Enter your provider-specific API Key, Base URL, and Model Name (e.g., openai/gpt-4o-mini).

🧪 Usage Examples

• "Tell me the weather in Tokyo." (Web Search -> Summary)
• "Set an alarm for 7:30 AM labeled 'Gym'." (Direct System Action)
• "Open WhatsApp and send 'Hello' to John." (Multi-step UI Interaction)

🛡️ Privacy & Security

• Encrypted Storage: API keys and sensitive settings are stored using EncryptedSharedPreferences.
• Blacklist: Banking apps are hardcoded to be invisible to the agent to ensure financial data never leaves the device.

---

Created by abishaziz with ❤️ and Antigravity AI.