Adrishya

Adrishya is a cross-platform desktop AI assistant overlay built using Electron, React, TypeScript, and Tailwind CSS. It serves as a fully functional, private overlay that listens to conversations, reads screen contents using local OCR, and provides contextual AI suggestions.

Designed for privacy, Adrishya is completely invisible during screen-sharing sessions (e.g., Zoom, Teams, Discord, Google Meet) utilizing native OS window protection.

• 🛡️ Screen Share Protection (Invisibility): Uses native hardware-accelerated Content Protection (win.setContentProtection(true)), causing the window to appear completely black or disappear in stream recording/screen sharing.
• ✨ Frosted Glass UI (Glassmorphism): Beautiful semi-transparent window utilizing Windows 11 native Acrylic material and macOS native Vibrancy effects.
• 🔍 Private Local OCR: Captures screen buffers and extracts code, questions, or text locally using a background WebAssembly tesseract.js process.
• 🎙️ Speech Transcription: Continuous voice transcription using Chromium's Web Speech API (Free) or OpenAI/Groq Whisper APIs (for sub-second latency and multi-language support).
• 🚀 Collapsible overlay widget: Drag the header anywhere using frameless dragging. Collapse it into a minimal floating horizontal pill to minimize distraction.
• ⚙️ Multi-Provider Support: Configure API keys and models for Google Gemini, OpenAI GPT, Anthropic Claude, Groq, and xAI Grok on the fly.

For a comprehensive architectural breakdown, process diagram, system demerits/limitations, and detailed specification, please refer to the Technical Specification Document (doc.md).

• Node.js: v18.x / v20.x or higher
• npm: v9.x or higher

Clone the repository and install all dependencies:

git clone https://github.com/IamJayPrakash/adrishya.git
cd adrishya
npm install

Start the Vite dev server and launch Electron:

npm run dev

Compile and package binaries locally for your target OS:

# Package for Windows (.exe)
npm run build:win

# Package for macOS (.dmg, .zip)
npm run build:mac

# Package for Linux (.AppImage, .deb)
npm run build:linux

Output files will be generated in the dist/ or out/ folder.

Adrishya registers global shortcuts at the operating system level, allowing you to control the app even when focusing on full-screen exams, coding editors, or browsers:

1. Launch the application.
2. Navigate to the Settings tab.
3. Select your desired active AI provider from the dropdown.
4. Input your API key (your key is saved securely in LocalStorage on your machine).
5. (Optional) Choose the default model name (e.g., gemini-1.5-flash or gpt-4o-mini).
6. Click Save Settings.

• Opacity: Adjust the opacity slider in the settings tab to make the glass overlay more or less translucent.
• Font Size: Scale message text size from 11px to 18px using the font size slider.
• Themes: Switch between:
  • Light: Frosted glass light theme.
  • Dark: Translucent frosted dark theme (Default).
  • AMOLED: Deep dark layout, perfect for OLED screens.

1. Move the overlay so it doesn't obstruct the text you want to scan.
2. Click the Screen tab, then click Capture Screen.
3. Review the extracted text in the code editor box.
4. Click Explain Code or Solve / Answer to automatically submit the text along with prompt guidelines to your configured AI assistant.

1. Click the Voice tab.
2. Ensure you have configured a Groq or OpenAI key if using Whisper API mode (Local Engine requires no keys).
3. Click Start Voice (or press Ctrl+Shift+V globally).
4. Speak; your transcript will roll onto the screen in real-time, and you can submit questions directly.

Contributions are welcome! Please check out CONTRIBUTING.md for local environment setup guidelines and pull request instructions.

This project is open-source and licensed under the MIT License.