Project Echo

The "Black Box" Flight Recorder for Digital Meetings

A privacy-first macOS utility that automatically captures audio from teleconferencing apps (Zoom, Teams, Meet) and generates searchable transcripts using local AI. Everything runs on-device with zero cloud uploads.

No virtual audio device needed! Uses ScreenCaptureKit to record both your microphone and meeting audio.

Features

Core Recording

• ScreenCaptureKit Integration - High-fidelity system audio capture
• Multi-track Recording - Separate tracks for system audio and microphone
• App-specific Capture - Target Zoom, Teams, Meet, and more
• Menu Bar Controls - Quick access to start/stop recording
• Marker Insertion - Tag important moments during calls

Intelligence Layer

• Local AI Transcription - WhisperKit (CoreML) for on-device processing
• Speaker Diarization - Identify who said what
• Smart Summarization - Extract action items and key topics
• Full-text Search - SQLite FTS5 for instant transcript search
• Zero Cloud Uploads - Everything stays on your device

User Interface

• Beautiful Library UI - SwiftUI interface with audio player
• Export Options - Audio and transcript export
• Comprehensive Settings - Customize quality, storage, models
• Privacy Dashboard - Clear data usage transparency

Architecture

+-----------------------------------------------------------+
|                     PROJECT ECHO                           |
+-----------------------------------------------------------+
|                                                           |
|  +-------------+  +---------------+  +-----------------+  |
|  | AudioEngine |  | Intelligence  |  |    Database     |  |
|  |             |  |               |  |                 |  |
|  | ScreenKit   |  | WhisperKit    |  |  SQLite + FTS5  |  |
|  | AVCapture   |->| CoreML        |->|  Recordings     |  |
|  |             |  | Diarization   |  |  Transcripts    |  |
|  +-------------+  +---------------+  +-----------------+  |
|         |                                    |            |
|         +------------------------------------+            |
|                        |                                  |
|                  +----------+                             |
|                  |    UI    |                             |
|                  |          |                             |
|                  | Menu Bar |                             |
|                  | Library  |                             |
|                  | Settings |                             |
|                  +----------+                             |
+-----------------------------------------------------------+

Quick Start

Prerequisites

• macOS Sonoma (14.0) or later
• Xcode Command Line Tools
• ~2GB disk space for Whisper models

Build & Run

# Build the app
./scripts/build.sh

# Launch it
./scripts/run_app.sh

First Launch

1. Grant Permissions (one-time setup)

   • Open System Settings > Privacy & Security
   • Microphone: Add ProjectEcho.app, enable checkbox
   • Screen Recording: Add ProjectEcho.app, enable checkbox

2. Start Recording

   • Click menu bar icon
   • Select "Start Recording"
   • Join your meeting
   • Click "Stop Recording" when done

3. View Transcript

   • Open Library from menu bar
   • Click on your recording
   • AI transcript generates automatically

What You Get

Project Structure

project-echo/
├── Sources/
│   ├── AudioEngine/          # ScreenCaptureKit + AVCapture
│   ├── Intelligence/         # WhisperKit transcription
│   ├── Database/             # SQLite management
│   ├── UI/                   # SwiftUI views
│   └── App/                  # Main entry point
├── Package.swift             # Swift Package Manager
├── Info.plist                # App metadata
└── scripts/                  # Build and utility scripts

Privacy & Security

Local-First AI

• Whisper models run entirely on Apple Neural Engine
• No audio or transcripts leave your device
• No analytics or tracking

Data Storage

• Recordings: ~/Documents/ProjectEcho/Recordings/
• Database: ~/Library/Application Support/ProjectEcho/echo.db

Required Permissions

• Screen Recording: For system audio capture
• Microphone: For your audio track
• File System: Read/write to save recordings

Configuration

General Settings

• Auto-transcribe - Generate transcripts automatically
• Whisper Model - tiny/base/small/medium (speed vs accuracy)
• Storage Location - Where recordings are saved

Advanced Settings

• Sample Rate - 44.1kHz or 48kHz
• Audio Quality - Standard/High/Maximum
• CPU Usage - Optimize for performance

Tech Stack

• Language: Swift
• Platform: macOS Sonoma+
• Audio: ScreenCaptureKit, AVFoundation
• AI: WhisperKit (CoreML)
• Database: SQLite with FTS5
• UI: SwiftUI

Development

# Run tests
swift test

# Build for release
swift build -c release --arch arm64 --arch x86_64

# Debug audio issues
system_profiler SPAudioDataType

Roadmap

• [x] Core Recording Engine
• [x] AI Transcription
• [x] UI Polish & Database
• [ ] Cloud Sync (Optional, user-controlled)
• [ ] Advanced Diarization
• [ ] Real-time Transcription
• [ ] Mac App Store Release

License

Proprietary - All Rights Reserved

Contributing

This is a prototype. For production deployment, additional work needed:

• Code signing for App Store
• Comprehensive error handling
• Unit test coverage
• Performance profiling