# Changelog

## 4.4.0 — 2026-10-01

- Transcription runs on Parakeet Ultra (FluidAudio 0.17.5). It has the same languages, word timings and speed as Parakeet v3 and fewer errors: 2.13% vs 2.27% word error rate on LibriSpeech test-clean, and lower in all 24 FLEURS languages FluidAudio measures. The model (about 600 MB) downloads once, with the first transcription after updating.
- A transcription cut off by quitting, a crash or a force quit restarts the next time Voom opens. Before, the recording stayed on "Transcribing…" with its Transcribe button hidden.
- Sharing a recording before its transcript is ready no longer leaves the share page without one. When transcription finishes, Voom pushes the transcript, title, summary and chapters to the page, and regenerating a transcript updates it too.
- Sharing while a transcription runs, or transcribing in the player while a share uploads, no longer throws away the other's result.
- The share server replaces a recording's transcript and chapters when the app re-sends them instead of adding a second copy.
- The AI model list now offers GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna; Claude Opus 5.5, Fable 5.1 and Sonnet 5.5; Gemini 3.8 Flash; and Grok 4.7. A saved model that was dropped moves to its successor at the same tier.
- Claude requests allow up to 16,000 output tokens and AI requests wait up to three minutes, because current models think before they answer and that thinking counts against the limit.

## 4.3.0 — 2026-09-24

- Meetings are diarized on your Mac with NVIDIA's Nemotron 3 (through FluidAudio 0.17.1), which tells up to eight remote speakers apart and detects people talking over each other. The previous pyannote pipeline remains as an automatic fallback.
- Speakers are assigned per word from the transcription's word timings, so lines split exactly where the speaker changes.
- When recording on laptop speakers, remote voices echoed into the mic are no longer labeled as you.
- Long meetings are streamed in chunks instead of being decoded into memory whole.
- The in-app What's New catches up on 4.0 through 4.2.

## 4.2.0 — 2026-09-22

- Record a single window: a new Window mode with a hover picker that highlights the window under the cursor. Capture follows the window as it moves, resizes or crosses displays, and records its own content even when another window sits on top of it.
- The webcam bubble is composited into window recordings, since a single-window capture cannot include a separate floating panel.
- Choose which camera the picture-in-picture bubble uses, from the control panel or Settings. External and Continuity cameras are listed, the choice is remembered, and an unplugged camera falls back instead of failing.
- Camera frame-rate requests are clamped to what the device actually supports, fixing a crash on cameras that report a fixed rate above the requested one.
- Recording now stops and saves with an explanation when the capture source goes away — a recorded window closing no longer leaves a silently truncated file.
- The microphone is released when a recording ends rather than when the camera session happens to be torn down, so the system microphone indicator no longer stays on afterwards.
- Microphone audio is resampled to the recording's rate before mixing, fixing sped-up, pitch-shifted audio from Bluetooth headsets and other non-48 kHz inputs.
- Audio timing is preserved per sample when moving captured audio onto the recording timeline.
- Voom's own playback is excluded from screen and meeting capture.
- `voom record --window <id>` records a single window from the CLI.

## 3.1.0 — 2026-03-06

- Bring your own AI: connect your own API key from OpenAI, Anthropic, Google, or xAI for AI-powered titles, summaries, and chapters.
- Auto-detects provider from API key prefix — paste a key and the provider and model switch automatically.
- New VoomAI package with direct provider API integration (no middleman services).
- AI settings in both Settings window and inline settings with provider picker, model selector, and connection test.
- Falls back to Apple's on-device model when no API key is configured (zero-config default unchanged).

## 3.0.2 — 2026-03-06

- Meeting recordings now use MeetingRecorder: HD/2K (capped at 2560), 30fps, split-track audio for speaker diarization.
- Fixed meeting recordings using ScreenRecorder at native retina resolution instead of the dedicated meeting pipeline.
- AI title and summary generation now subsamples transcripts to fit Apple Foundation Models context window, fixing failures on long recordings.
- Speaker labels from diarization are now included in title and summary prompts for meeting-aware AI generation.
- Summary card in player view fits content height without internal scrolling; edit via pencil icon.
- Lowered word count thresholds for title (5 to 2) and summary (10 to 3) to handle short recordings.

## 3.0.1 — 2026-03-06

- Web-optimized sharing: videos are re-encoded to H.264 for universal browser playback with smaller file sizes.
- WebVTT captions on share pages with speaker labels visible in the player and fullscreen.
- Auto-generated chapters displayed on share pages with clickable timestamps and seekbar markers.
- Share pipeline shows optimization progress in the share sheet before upload begins.
- Speaker labels and meeting flag included in share metadata.
- D1 migration adds chapters table, speaker column on transcript segments, and is_meeting flag.
- Added WhatsNewKit integration to showcase new features on first launch after update.

## 3.0.0 — 2026-03-05

- Migrated from WhisperKit to FluidAudio for on-device ASR and speaker diarization.
- Split monolithic app into VoomCore, VoomApp, and VoomMeetings packages.
- Split-track speaker diarization: saves separate mic and system audio during meeting recordings for accurate speaker separation.
- "You" identification: mic audio diarized separately so the local user's segments are labeled "You" instead of a generic speaker number.
- Remote speaker separation: system audio diarized independently for cleaner multi-speaker identification without local voice interference.
- Chapters now auto-generate after transcription for both regular and meeting recordings.
- Fixed chapter generation only covering first ~2 minutes by subsampling transcript evenly across full duration.
- Fixed garbled mic audio caused by AVAudioEngine voice processing conflicting with ScreenCaptureKit audio capture.
- Speaker labels displayed in transcript view.
- Added debug mode in settings.

## 2.8.1 — 2026-03-05

- Reduced video file size ~40% with optimized HEVC compression (8 Mbps, B-frames, 4s GOP).
- Replaced AVCaptureSession mic with AVAudioEngine for microphone capture.
- AI title generation now uses full transcript instead of only the first 2 minutes.
- Improved title prompts to focus on primary work topics, ignoring small talk.
- Enhanced AI summaries to 4-8 sentences covering decisions and action items.
- Meeting recordings auto-stop immediately when camera turns off.

## 2.8.0 — 2026-03-05

- Meeting detection: auto-trigger floating panel when video calls begin.
- Full dark theme applied to share settings sheet and all modal views.
- Share sheet primary buttons now use accentRed consistently.

## 2.7.3 — 2026-03-04

- Fixed Open Graph tags to use absolute URLs for share page embeds.
- Added embed player route for share links.

## 2.7.2 — 2026-03-04

- Refined vibrancy materials and card styling across the app.

## 2.7.1 — 2026-03-04

- Added vibrancy and blur materials to app backgrounds and cards.

## 2.7.0 — 2026-03-04

- Modernized codebase with destructive action confirmations.

## 2.6.2 — 2026-03-03

- Updated docs for free tier and added support email with system info.

## 2.6.1 — 2026-03-03

- Wired up tags feature in player detail view badge row and context menu.

## 2.6.0 — 2026-03-03

- Added self-host setup wizard to deploy Cloudflare Worker from within Voom.

## 2.5.2 — 2026-03-03

- Abort stale multipart uploads on failure.
- Added R2 lifecycle cleanup for orphaned uploads.

## 2.5.1 — 2026-03-03

- Fixed header buttons not being clickable.
- Added hardened runtime entitlements.

## 2.5.0 — 2026-03-03

- Fixed notarization by re-signing Sparkle binaries with timestamps.
- Enabled hardened runtime for release builds.
- Auto-set marketing and project version from git tag in release workflow.

## 2.4.1 — 2026-03-02

- Automated Sparkle appcast generation in release workflow.

## 2.4.0 — 2026-03-02

- UI overhaul: fixed selection colors, added animations.
- Integrated Sparkle for auto-updates.

## 2.3.1 — 2026-03-02

- Fixed sidebar navigation and recording data safety.

## 2.3.0 — 2026-03-02

- Full Loom feature parity: recording modes, video editing, folder organization, share enhancements.
- Fixed OG tags on password-protected share pages.

## 2.2.1 — 2026-03-02

- Improved share page: HD thumbnails, centered layout, app icon.

## 2.2.0 — 2026-03-02

- AI-generated titles and summaries for recordings.
- Editable title and summary fields.
- Multipart upload for large recordings.
- Thumbnail Open Graph images for share links.

## 2.1.1 — 2026-03-02

- DMG now includes Applications shortcut for drag-to-install.

## 2.1.0 — 2026-03-02

- New app icon.

## 2.0.0 — 2026-03-02

- Single audio track recording with real-time mic + system audio mixing.
- Custom toast notification system replacing alert dialogs.
- Custom video controls on share page with seekbar, volume, and fullscreen.
- View counter and timestamp copy links on shared videos.
- Increased upload timeout for large recordings.
- Upfront permission requests on app launch.

## 1.0.0 — 2026-03-02

- Screen recording with webcam PiP overlay.
- System audio and microphone capture.
- On-device transcription via WhisperKit.
- Share via link (self-hosted on Cloudflare).
- Global keyboard shortcut (Cmd+Shift+R).
- Auto-transcription toggle.
- First-launch onboarding flow.