Skip to main content

Technology Stack

Overview​

This document outlines the complete technology stack for Aurix. Each choice is driven by our requirements for offline-first operation, user privacy, simple deployment, and powerful local AI capabilities.

Core Framework​

Electron (v28+)​

  • Why: To build a cross-platform desktop application using familiar web technologies.
  • Benefits:
    • Native OS integration (global hotkeys, system tray, notifications).
    • Secure credential storage via safeStorage for optional cloud features.
    • Direct file system access for local audio and document storage.

Electron Forge​

  • Why: Provides a modern, integrated tooling experience for Electron development.
  • Benefits:
    • TypeScript and Vite support out of the box.
    • Hot module reloading for a fast development cycle.
    • Integrated packaging and distribution tools.

Frontend Stack​

  • React 18+: For building a component-based, declarative user interface.
  • TypeScript: For end-to-end type safety, improving code quality and maintainability.
  • Vite: For a lightning-fast build tool with near-instant hot module replacement.
  • Tailwind CSS: For a utility-first CSS framework to build a consistent design system rapidly.
  • shadcn/ui: For a collection of re-usable, accessible components that we can own and customize directly.

Voice & AI Processing​

Voice Processing​

  • Primary Engine (Online): Web Speech API for high-accuracy, low-latency cloud-based transcription.
  • Secondary Engine (Offline): Vosk for reliable, private, on-device transcription.
  • Hybrid Approach: The application will automatically switch between engines based on internet connectivity and user preference, ensuring the best experience in any environment.

AI Workflow Orchestration​

  • Framework: LangGraph (JavaScript/TypeScript version).
  • Why: It is lightweight, runs entirely in-process, supports complex state management and cycles (for feedback loops), and is perfectly suited for AI-native workflows without requiring an external server.

AI Model Integration​

  • Local (Primary): Ollama for managing and running local LLMs (e.g., Llama 3, Phi-3). This ensures complete user privacy and no API costs for core functionality.
  • Cloud (Optional): Support for user-provided API keys for cloud models (e.g., GPT-4, Claude 3) for users who need enhanced power for specific tasks.

Data Storage & Sync​

  • Primary Storage: SQLite (via better-sqlite3) for robust, zero-configuration local database storage. It provides ACID compliance and full SQL capabilities.
  • Optional Sync: PouchDB/CouchDB for users who opt-in to cloud features. This provides a battle-tested, conflict-free architecture for multi-device synchronization.

Packaging & Distribution​

  • Tool: Electron Forge's built-in makers (or electron-builder if more advanced features like delta updates are needed).
  • Formats:
    • Windows: NSIS installer.
    • macOS: DMG with notarization.
    • Linux: AppImage.

Key Architectural Decisions​

  • Process Model: A strict separation between the Main Process (handles Node.js APIs, file system, AI work) and the Renderer Process (UI only). All communication occurs via secure, promise-based IPC through a contextBridge.
  • State Management: Zustand for lightweight, simple state management in the React UI. The main process state will remain the ultimate source of truth for application data.
  • Privacy First: All core functionality, including voice transcription and AI processing, must work entirely offline. Cloud features are strictly opt-in.
  • Error Handling: The application is designed for graceful degradation. Features should fail independently without crashing the entire app. For example, if a diagram fails to generate, the text document should still be produced.