v1.4

New Features

• On-Device Whisper Transcription — Transcription now powered by OpenAI Whisper running entirely on-device via Apple Neural Engine, replacing Apple Speech for dramatically better accuracy. Choose from 4 model sizes (Tiny ~75 MB, Base ~150 MB, Small ~500 MB, Large v3 ~3 GB), download on demand from Settings, and switch or delete models with one click. 26 languages with auto-detect, real-time download progress with cancel support, and auto-transcribe after recording with live progress in the session list. No audio data ever leaves your Mac
• Window-Level Capture — The "Apps" capture mode now shows individual windows instead of whole apps. Select specific windows to record — pick two Chrome windows and each gets its own output file. Windows listed from all Dock-visible apps across all displays, with a refresh button and hover-to-highlight preview
• Multi-Source Standard Mode — Standard mode now supports recording multiple sources simultaneously, previously only available in composite mode
• Camera + Microphone Pairing — Pair a camera with a specific microphone to combine video and audio into a single file. Useful for external cameras with dedicated mics
• Stream Deck Live Timer — The Stream Deck record button now shows a live recording timer that updates every second, so you always know how long you've been recording
• App Store Update Checker — Lumary now checks for updates on the App Store and notifies you when a new version is available. Check from the menu bar or Settings

Improvements

• Whisper pipeline automatically unloads after 60 seconds of inactivity to free memory
• Whisper special tokens are stripped from transcript output
• Separate microphone audio file now written alongside composite output when auto-transcribe is enabled, ensuring transcription works in composite-only mode
• Inline create, rename, and delete for courses, modules, and lessons with improved navigation and cleaner visual hierarchy
• Welcome screen now persists output directory choice across app restarts and no longer blocks system-initiated restarts when granting permissions
• Camera and Microphone permission buttons now trigger the system permission prompt directly
• Permission checks (Screen Recording, Camera, Microphone) now always appear in preflight regardless of source selection
• Permission issues in the inspector sidebar are now clickable — "Open Settings" links directly to the correct System Settings pane
• Added Accessibility and Input Monitoring quick-links to open their settings panes
• Reduced idle preview CPU usage from ~78% to ~27%
• Composite preview capped at 1080p/10fps during recording to reduce CPU overhead
• Default encoding changed from ProRes to H.264 High Quality for smaller file sizes
• Default preview frame rate increased to 60 fps for smoother preview

Bug Fixes

• Fixed transcription failing for recordings longer than ~60 seconds
• Fixed high CPU usage during composite recording caused by SwiftUI AttributeGraph cycles
• Fixed multi-app capture not including all selected applications
• Fixed composite recording issues with certain source configurations
• Fixed app capture mode not recording the selected application
• Fixed delete session/project missing confirmation dialogs
• Fixed sidebar resize and toggle behavior — sidebar now respects manual hide/show when resizing the window
• Fixed auto-transcribe not showing results until manually refreshed
• Fixed app unable to quit or restart when Settings or other sheets are open
• Added label to Teleprompter toolbar button for clarity
• Added "Open" label to the sessions window button for clarity