Avante

Local LLM Writing Assistant | Swift, SwiftUI

A private, local-first macOS application that uses on-device LLMs for real-time writing analysis. Built with Swift and SwiftUI, Avante demonstrates advanced macOS development techniques and innovative AI integration.

Features

• Privacy-First: All analysis happens locally on your device
• Real-Time Analysis: Live writing feedback as you type
• Smart Context Management: Novel recursive summarization algorithm overcomes context-window limitations
• Advanced Metrics: Novelty, Clarity, and Flow analysis with visual highlighting

• Focus Mode: Distraction-free writing environment

• Native File Management: Seamless integration with macOS file system
• Undo/Redo Support: Full document state management

Architecture

Core Technologies

• SwiftUI: Modern declarative UI framework
• FoundationModels: Apple's on-device LLM framework
• Combine: Reactive programming for real-time updates
• NaturalLanguage: Text processing and tokenization

Key Components

Analysis Engine

• Recursive Summarization: Breaks down large documents into manageable chunks while maintaining global context
• Context Window Management: Intelligent handling of local model limitations
• Real-Time Processing: Live analysis with debounced input handling

UI Architecture

• MVVM Pattern: Clean separation of concerns
• Reactive Bindings: SwiftUI + Combine for responsive UI
• Custom Layout Managers: Advanced text highlighting and metrics display
• Native File Integration: Document-based app architecture

Getting Started

Prerequisites

• macOS 26.0+ (Tahoe or later)
• Apple Silicon Mac (for optimal LLM performance)

Installation

1. Clone the repository
2. Open avante.xcodeproj in Xcode
3. Build and run the project

Usage

1. Open a text file or create a new document
2. Start typing to see real-time analysis
3. Use the metrics sidebar to view detailed feedback
4. Toggle focus mode for distraction-free writing

Technical Highlights

Recursive Summarization Algorithm

The application implements a novel approach to handling large documents by breaking them into manageable chunks while preserving global context. This allows local models to understand document-wide patterns despite their limited context windows.

Real-Time Analysis Pipeline

• Input Debouncing: Prevents excessive API calls
• Context Preservation: Maintains document coherence across chunks
• Metrics Aggregation: Combines local insights into global understanding

Performance Optimizations

• Lazy Loading: Efficient memory management for large documents
• Background Processing: Non-blocking analysis operations
• Smart Caching: Reduces redundant computations

Project Goals

This project demonstrates:

• Advanced SwiftUI Development: Complex UI patterns and state management
• AI Integration: Seamless on-device LLM integration
• Performance Engineering: Optimized for real-time processing
• macOS Best Practices: Native app architecture and user experience

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Built with Apple's FoundationModels framework
• Inspired by modern writing assistant applications
• Demonstrates advanced SwiftUI and Combine patterns

Note: This project is designed as a technical demonstration of advanced iOS/macOS development capabilities, particularly in the areas of AI integration, real-time processing, and native app architecture.