HookCut

macOS app that uses AI to find the best highlights in podcast and video transcripts, then exports them as NLE-ready timelines

---

Features

• AI Highlight Detection -- GPT-4o, Claude, or a local Ollama model analyzes full transcripts to find the best one-liners, cliffhangers, hot takes, emotional moments, quotable insights, and humor
• Cloud or On-Device Transcription -- Transcribe with the OpenAI Whisper API, or run fully on-device with WhisperKit (CoreML / Neural Engine) — no key, no upload, no cost
• Fully Local Mode -- Pair on-device Whisper with a local Ollama model for an end-to-end pipeline that never touches the cloud
• Whisper Transcription -- Extracts audio from video files and transcribes with word-level timestamps; long files are automatically chunked and stitched
• Speaker Diarization -- Identifies and labels individual speakers throughout the transcript
• Prompt Templates -- Built-in templates (Podcast Teaser, Best Quotes, Key Takeaways, Funny Moments, Controversial Takes) plus custom prompt support
• Multi-Format Export -- Export highlights as FCPXML (Final Cut Pro), Premiere XML, EDL, SRT subtitles, CSV, or plain text
• Teaser Sequencing -- AI suggests an optimal highlight playback order for maximum impact in teasers and trailers
• Batch Processing -- Queue multiple media files, process them sequentially with per-item progress and cancellation, then Export All to a folder in one click
• Video Preview -- Built-in video player with highlight navigation and back-to-back "Preview All" of approved clips
• Configurable Detection -- Set highlight count, target duration, enabled types, and custom prompt additions
• AI Provider Choice -- OpenAI GPT-4o, Anthropic Claude, or local Ollama, with automatic OpenAI fallback when a Claude key is missing
• Cost Estimation -- Estimates API cost before processing (and shows $0 for fully-local runs)

Getting Started

Prerequisites

• macOS 14 (Sonoma) or later
• Swift 5.9+ and Xcode 15+
• For the cloud path: an OpenAI API key (Whisper transcription + GPT-4o analysis)
• (Optional) An Anthropic API key for Claude-based highlight detection
• (Optional) Ollama running locally (ollama serve) for fully-local analysis — no API key required

Build & Run

git clone https://github.com/markksantos/HookCut.git
cd HookCut
swift build
swift run HookCut

Or generate the Xcode project with XcodeGen and open it:

xcodegen generate
open HookCut.xcodeproj   # then press Cmd + R

After launching, the first-run welcome screen points you to Settings (⌘,). Pick your transcription engine (Cloud or On-Device) and AI provider, enter any needed API keys, and you're ready to import a file.

Tests

swift test

Covers the timecode math, every export format (including XML-escaping correctness), and the cost estimator.

Tech Stack

Layer	Technology
Language	Swift 5.9
UI Framework	SwiftUI (macOS 14+)
Transcription	OpenAI Whisper API or on-device WhisperKit
Highlight AI	OpenAI GPT-4o / Anthropic Claude / local Ollama
Audio Pipeline	AVFoundation (AudioExtractor)
Export Formats	FCPXML, Premiere XML, EDL, SRT, CSV, TXT
Build System	Swift Package Manager (XcodeGen for .xcodeproj)
Persistence	UserDefaults (settings) + JSON session cache

Project Structure

HookCut/
├── Package.swift
└── HookCut/
    ├── HookCutApp.swift               # App entry point, AppState, SettingsManager, PromptTemplates
    ├── Models/
    │   └── SharedModels.swift          # TranscriptionResult, AnalysisResult, Highlight, Speaker, etc.
    ├── Pipeline/
    │   └── AudioExtractor.swift        # AVFoundation audio extraction from video files
    ├── Services/
    │   ├── APISession.swift            # Shared URLSession with rate-limit retry
    │   ├── CostEstimator.swift         # API cost estimation before processing
    │   ├── HighlightDetector.swift     # GPT-4o / Claude highlight detection with structured prompts
    │   ├── SpeakerDiarization.swift    # Speaker identification and labeling
    │   ├── TranscriptionService.swift  # Orchestrates extract -> transcribe -> diarize -> detect
    │   ├── LocalWhisperService.swift   # On-device transcription via WhisperKit
    │   └── WhisperService.swift        # OpenAI Whisper API integration
    ├── ViewModels/
    │   └── AppViewModel.swift          # Main view model coordinating UI and services
    ├── Views/
    │   ├── ContentView.swift           # Main window layout
    │   ├── ImportView.swift            # File import / drag-and-drop
    │   ├── TranscriptView.swift        # Scrollable transcript with speaker labels
    │   ├── HighlightsPanel.swift       # Highlight list with ratings and types
    │   ├── VideoPlayerView.swift       # Built-in video preview with highlight navigation
    │   ├── ExportSheet.swift           # Export format selection and configuration
    │   ├── BatchView.swift             # Batch processing queue + Export All
    │   ├── WelcomeView.swift           # First-run onboarding screen
    │   └── SettingsView.swift          # API keys, AI provider, highlight preferences
    └── Export/
        ├── ExportService.swift         # Export orchestration protocol implementation
        ├── FCPXMLGenerator.swift       # Final Cut Pro XML export
        ├── PremiereXMLGenerator.swift  # Adobe Premiere XML export
        ├── EDLGenerator.swift          # Edit Decision List export
        ├── SRTExporter.swift           # Subtitle export
        ├── CSVExporter.swift           # Spreadsheet export
        └── PlainTextExporter.swift     # Plain text export

Packaging & Distribution

A debug .app is produced by Xcode/XcodeGen builds. To ship a signed, notarized .dmg you need an Apple Developer account and a Developer ID Application certificate:

xcodebuild -project HookCut.xcodeproj -scheme HookCut -configuration Release \
  -derivedDataPath build \
  CODE_SIGN_IDENTITY="Developer ID Application: <Your Name> (<TEAMID>)" \
  build
# then: create-dmg / hdiutil, codesign --deep, notarytool submit, stapler staple

The app already ships hardened-runtime + App Sandbox entitlements (HookCut/HookCut.entitlements) and a PrivacyInfo.xcprivacy manifest, so it's ready for notarization and the Mac App Store once a signing identity is in place.

License

MIT © 2025 Mark Santos — see LICENSE