A macOS voice transcription application built with SwiftUI that aims to provide seamless speech-to-text conversion with AI-powered text processing capabilities.
VoiceFlow is designed to be a lightweight, floating macOS application that:
- Captures voice input through the system microphone
- Transcribes speech to text using advanced speech recognition
- Processes transcribed text with AI/LLM integration for refinement and enhancement
- Provides quick access via a compact, always-on-top floating window
- Offers seamless workflow integration with copy-to-clipboard functionality
The goal is to create a productivity tool that makes voice-to-text conversion as frictionless as possible for macOS users.
This project is currently in the foundational development phase. Here's what's implemented vs. planned:
- Basic UI Structure: Floating 400x530 window with title bar
- Component Architecture: Separated button components (Copy, Record, Magic Wand)
- Visual States: Button hover effects, loading overlays, and basic state management
- Dark Mode Support: Adaptive colors for light/dark themes
- Debug Interface: Development tools for testing UI interactions
- Clipboard Integration: Basic text copying functionality
- Voice Recording: No actual microphone capture implemented
- Speech Recognition: No speech-to-text processing
- AI Integration: Magic wand processing is currently simulated
- Error Handling: Limited error states and user feedback
- Persistence: No data storage or session management
- Audio Permissions: Basic setup present but not fully integrated
The application features a clean, minimal interface with:
- Prominent "VOICE FLOW" title
- Text display area with scrollable content
- Three main action buttons: Copy, Record, and Magic Wand
- Loading states with progress indicators
- Accessibility support with VoiceOver labels
Built using modern SwiftUI patterns:
VoiceFlow/
├── Views/
│ ├── ContentView.swift # Main application view
│ ├── ButtonComponent.swift # Reusable button base
│ ├── CopyButton.swift # Clipboard copy functionality
│ ├── RecordButton.swift # Voice recording control
│ ├── MagicWandButton.swift # AI processing trigger
│ └── TextDisplayView.swift # Transcription display area
├── VoiceFlowApp.swift # App entry point and window config
└── Info.plist # Permissions and metadata
This was developed as a small hackathon project to explore modern AI-assisted development workflows. The focus was on rapid prototyping and experimenting with LLM-powered code generation tools.
- Claude Code: AI-powered command line coding assistant
- Claude: LLM for code review, architecture planning, and documentation
- Xcode MCP Server: Model Context Protocol integration for Xcode
- GitHub MCP Server: Git workflow automation and PR management
The development process was heavily influenced by the LLM workflow described in this blog post about LLM codegen workflow, emphasizing:
- AI-assisted code generation and iteration
- Automated code review and feedback
- Rapid prototyping with intelligent tooling
- Integration of multiple AI services in the development pipeline
- macOS 12.0 or later
- Xcode 14.0 or later
- Swift 5.7 or later
-
Clone the repository:
git clone https://github.com/nicerice/VoiceFlow.git
-
Open the project in Xcode:
cd VoiceFlow open VoiceFlow.xcodeproj -
Build and run the project (⌘+R)
- Test the UI: Use the debug buttons to simulate text input and loading states
- Copy Feature: Add sample text and test the clipboard functionality
- Button States: Observe how buttons enable/disable based on application state
- Dark Mode: Toggle system appearance to test theme adaptation
This project is open source and available under the MIT License.
- Repository: https://github.com/nicerice/VoiceFlow
- Blog post about LLM workflow: https://harper.blog/2025/02/16/my-llm-codegen-workflow-atm/
- Claude Code: https://www.anthropic.com/claude-code
- Claude: https://claude.ai
- Xcode MCP Server: https://github.com/cameroncooke/XcodeBuildMCP
- GitHub MCP Server: https://github.com/github/github-mcp-server
Built with ❤️ and AI assistance during a weekend hackathon