AgenticDNN - Agentic Deep Neural Network Builder

A beautiful macOS desktop app for building and training deep neural networks with an agentic AI interface. Built with SwiftUI, inspired by Claude Code's aesthetic.

Features

✨ Current Implementation

• Claude-Inspired UI: Dark, modern interface matching Claude Code's visual language
• Welcome Screen: Beautiful onboarding with "Good morning" greeting
• Agentic Terminal: Interactive terminal interface for conversing with AI about DNN architecture
• Split-Screen Mode: Automatic split when building - terminal on left, visual builder on right
• LEGO-Style Builder: Drag-and-drop visual DNN designer with:
  • Input layers
  • Dense, Conv2D, LSTM layers
  • Dropout, BatchNorm, Pooling operations
  • Activation functions
  • Output layers
• Session Management: Track multiple DNN projects in sidebar
• Smooth Animations: Polished transitions and interactions

🎨 Design System

• Color Palette: Dark theme with orange accents (#D97706)
• Typography: SF Pro for UI, SF Mono for terminal
• Spacing System: Consistent 4px-based spacing
• Component Library: Reusable UI components matching Claude aesthetic

🏗️ Architecture

AgenticDNN/
├── project1.swift          # App entry point
├── Theme.swift            # Design tokens & color system
├── AppState.swift         # State management
├── ContentView.swift      # Main app container
├── WelcomeView.swift      # Onboarding screen
├── SidebarView.swift      # Project sessions sidebar
├── TerminalView.swift     # Agentic terminal interface
└── SplitScreenView.swift  # Split layout + DNN builder

Running the App

# Build the app
swift build

# Run the app
swift run

# Or open in Xcode
open Package.swift

Usage

1. Start: Launch app to see welcome screen
2. Create Project: Click "Start New Project"
3. Chat: Describe your DNN in the terminal (e.g., "Build a CNN for image classification")
4. Build: Click split-screen icon to open visual builder
5. Design: Drag blocks from palette to canvas to build your network

Roadmap

Phase 1: Core UI ✅

• App shell with Claude theme
• Welcome screen
• Terminal interface
• Split-screen layout
• Visual DNN builder

Phase 2: Integration (Next)

• Codebase import functionality
• Linear integration for project context
• Claude API integration for real AI responses

Phase 3: Backend

• TensorFlow runtime execution
• PyTorch support
• Model training orchestration
• Export to Python code

Tech Stack

• Language: Swift 6.2
• Framework: SwiftUI (macOS 14+)
• Architecture: MVVM with ObservableObject
• Future: TensorFlow, PyTorch, Claude API

Development

The app is currently in frontend-complete state. All UI components are functional with mock data. Next steps involve wiring up real AI backend and ML frameworks.

License

Built for Factory AI mission - Agentic DNN Builder