Filler words like "um", "uh" and "er" are now removed by a fast, exact step of their own — so they disappear from your dictation even when the on-device cleanup model has an off moment. • When the cleanup model's result fails OTO's safety check, you now keep the filler-free text instead of falling back to the raw transcript. • Filler removal recognises the language you actually spoke, and does nothing when it isn't English. • Deterministic cleanup and the on-device model now share one Settings switch and run in one predictable order for every dictation. Everything still runs entirely on your Mac. Your words never leave your device.

Transcripts that read the way you meant them. OTO now tidies up every dictation automatically with a new on-device cleanup stage: filler words disappear, spoken numbers become digits, and punctuation and capitalization get fixed. It runs entirely on your Mac, on top of whichever voice model you use, and never sends your words anywhere. Your words stay your words — if the cleaned-up result ever strays from what you actually said, OTO throws it away and keeps the raw transcript. And there's a real off switch in Settings if you'd rather have it untouched. NEW • Automatic transcript cleanup, on-device, for every voice model • Cleanup now works everywhere you dictate, including code editors and terminals • Recognizes the language you actually spoke, so non-English dictation is left alone • Edit transcripts in History — repeated corrections are learned and added to your dictionary • An optional "Clean up" mode for rewriting dictation with a local or bring-your-own-key model FIXED • Recording no longer cancels itself while the microphone warms up • Sentences following a pause are capitalized correctly again • Repeated-phrase loops are fully collapsed • WhisperKit transcripts now get paragraph breaks, matching Parakeet Please note: Modes now apply in code editors and terminals too. If you have a Mode set up with a cloud provider, text dictated into those apps is now sent to that provider like any other dictation. If you dictate secrets or credentials, switch that Mode to a local provider or set it to none.

Search your voice history — from anywhere • Search everything you've ever dictated, right from the terminal: oto search "that thing I said" finds it instantly • AI assistants connected via MCP (Claude, and any MCP-compatible tool) can now search your history too, and filter by kind — dictations, meetings, or transcribed files — all on your device, like always • Every MCP capability now has a matching CLI command; oto -h shows the full set More reliable meeting recordings • Fixed a bug where recording with a Bluetooth microphone could silently lose the other side of the call — a watchdog now guards the whole session and recovers within seconds if audio stops arriving • The combined playback file no longer cuts short when your mic and system audio recorded at different sample rates • If part of a recording did fail, its session now says so honestly instead of reporting everything was fine • Restored automatic volume balancing and echo reduction in two-sided transcripts Plus: connecting a CLI or AI tool while OTO is running can no longer launch a second copy of the app.

This is a big one. oto now goes everywhere you work, and understands you better than ever. • Command Line Tool: Drive oto from your terminal: transcribe files, start and stop recordings, browse history. Install it in one click from Settings → Advanced. • AI Assistant Integration (MCP): Let AI tools like Claude transcribe through oto, fully on-device. A ready-to-paste config lives in Settings → Advanced. • Major Accuracy Upgrade: Transcription is noticeably more accurate across the board, thanks to an upgraded on-device engine and higher-fidelity audio capture. Soft consonants, word endings, and quiet recordings all come through more reliably. • New English Model: Parakeet V2 (English) is now available as an optional download for the best possible English accuracy. • Screen Context (opt-in): oto can read the names visible on your screen when you start recording, so people and products are spelled exactly as you see them. Fully on-device: one frame, read and discarded. Nothing stored, nothing sent. • Echo-Free Meeting Recordings: In long-form recordings, your mic no longer picks up what your speakers are playing. The other side of the call stays on their side of the transcript, even without headphones. • Readable Long Transcripts: Long recordings now break into natural paragraphs instead of one wall of text. • Smoother Waveforms, Teams recording support, a long-form history filter, and dozens of refinements throughout. Your voice. Your hardware. Your rules.

- Long-form meeting recording — new mode for extended sessions with speaker diarization, recovery support, and a red status indicator while recording - Redesigned Settings — full SwiftUI rewrite with translucent sidebar, cleaner navigation, and a unified design system - Batch file transcription — drop multiple audio files and process them in a managed queue with per-item status - Quick re-paste — access recent transcriptions directly from the menu bar - M4A, MP3, and AAC fix — these formats now decode correctly (were previously producing silence) - Quieter speech detection — VAD threshold lowered so soft recordings are no longer rejected - Retranscribe with model picker — re-run any past recording through a different model from History

- **Settings Migration Safety** — Prevents all user settings (preferences, dictionary, hotkey config) from being reset on app update by adding migration-safe Codable decoders to preference structs

- **Parakeet Quality Overhaul + ITN** — Closes the quality gap with trailing silence padding, VAD tuning, hallucination filtering, filler word removal, and a full Inverse Text Normalization engine (spoken numbers → digits, currency, phone numbers, percentages) - **Unified WaveformEngine** — Consolidates three independent waveform renderers into a single behavioral source, eliminating 546 lines of duplicated animation logic - **USB Mic Silent Recording Fix** — Replaced AVAudioConverter (which silently zeroed multi-channel USB mic output) with manual channel mixing + linear interpolation resampling

- **Mini Player** — slim 100x34pt pill overlay with 6 snap points, floats on any display with automatic routing to notch on built-in screens - **Recording File Retention** — configurable auto-cleanup for audio files and transcription text (7/30/90/Forever), runs silently at launch - **Persistent Stats Accumulator** — dashboard totals survive history deletion via write-through accumulator with one-time migration - **Parakeet Model Warmup** — model preloads at startup and stays resident, eliminating cold-start latency on first transcription - **File Transcription Queue** — live progress panel for batch file jobs with per-item status, retry, and cancel - 5 Various bug fixes

1. AUHAL render callback crash — the main fix, affects USB/Bluetooth/AirPods/built-in during stop or device switch 2. Notch overlay hide state guard — overlay could get stuck visible 3. Diagnostic log crash report attachment — new flow to attach a .ips crash file when exporting diagnostics

- AirPods & Bluetooth mics now work reliably — no more Error 35 or retry loops - Continuity Microphone (iPhone mic) recording fixed — no more 0.1s empty files - Clipboard preserved when auto-paste is on — you never lose what you had copied - Loopback/Audio Hijack aggregate devices now record from the correct mic - Intel Mac crash fixed for large Whisper models - MacBook users with external monitors can now record normally (clamshell false positive fixed) - Input device name now shown in transcription history - Model stats UI enlarged for readability - Settings split into focused Appearance and Audio cards

- **Contextual Tips** - Smart post-onboarding tips surface contextually after real usage (e.g., auto-paste suggestion after 3 transcriptions, file transcription tip after 5). Non-intrusive floating banners with one-tap actions. - **Streamlined Onboarding** - Onboarding condensed from 6 screens to 4 (Welcome → Microphone → Try It → Ready). Success overlay with auto-advance replaces the old manual completion flow.

- **AI Enhancement Global Hotkeys & UI Redesign** - Revamped AI settings with a cleaner card-based layout. Press Control+Option+E to cycle through your prompts during recording, or assign custom hotkeys to jump to specific ones. - **Double-Press ESC to Cancel** - Recording cancellation now requires a deliberate double-press of the ESC key, preventing accidental cancellations while keeping the escape key available for other apps. - **What's New Tab** - A new "What's New" settings tab displays formatted release notes directly inside the app, giving users a polished, in-app changelog organized by version. - **In-App Update Checker** - oto now automatically detects new App Store versions and surfaces a subtle sidebar button and local notification so you never miss an update. - **Hotkey Preferences Fix** - Existing users who kept the original default hotkey (Command+Option+Space) no longer have their preference silently reset on update. - **Notch Loading Indicator** - Pressing the hotkey before the model is ready now shows a "Loading..." message in the notch overlay.

- **Auto-Format on Paste** - Transcripts are automatically normalized into clean, punctuated paragraphs and lists before clipboard/paste with O(n) single-pass processing. - **Global Recording Hotkeys** - New Carbon-based global hotkeys (⌃⌥F for auto-format, ⌃⌥T for language toggle) work system-wide during recording without requiring window focus. - **Multi-Monitor Support** - Recording panel and notch overlay now appear on whichever screen has the mouse cursor. - **Full Accessibility Overhaul** - Comprehensive VoiceOver, Voice Control, Dynamic Type, Reduce Motion, and keyboard navigation support across all UI components.

**Accessibility Foundation for App Store Compliance**: Comprehensive accessibility infrastructure including Dynamic Type, Reduced Motion support, and VoiceOver foundations for all 100+ UI components **Auto-Paste with Accessibility Permission**: - New Auto-Paste section in Preferences enables automatic paste after transcription using App Store-compliant Accessibility permissions

- **Enhanced Onboarding Tutorial** - Improved first-run experience with better navigation, error handling, and "How to Cancel" guidance

- Audio-Reactive Recording Animation - Notch waveform bars respond to actual audio levels during recording - Bouncing Wave Processing Animation - Processing state shows smooth horizontal wave bouncing across bars - System Default Audio Input - New "System Default" option in audio input dropdown - Scrolling Settings Title - Title/subtitle now scroll with content in Preferences and Model tabs - Simplified Menu Bar - Streamlined menu with essential items only - Standard macOS Menus - Full Edit, View, Window, and Help menus when Settings window is open - Crash Recovery Fix - Successful transcriptions no longer incorrectly marked as "interrupted" - Memory Leak Prevention - Stale session cleanup and orphaned temp file removal - Floating Sidebar Settings - Modern floating card layout with shadows and rounded corners - Model Grading System - Labeled rating display with dot visualization for model performance - Performance Optimizations - Shadow rasterization and cached high contrast mode

- Recording Preservation: Recordings no longer lost when starting a new recording while transcription is in progress - Model Tab Reliability: Downloaded models immediately selectable without restart; fixed multiple "Active" states - Model Preloading Fix: Resolved infinite "Model is loading" state after engine/language changes - Recent Languages Fix: Recent language pills now correctly persist when switching languages - Smart Tap Edge Case: Recording GUI no longer gets stuck when recording is too short (less than 0.5s)

- Smart Tap Activation -> New permission-aware activation system with true double-tap detection - Crash-Resilient Recording -> Recordings now survive app crashes with automatic recovery on next launch - Clamshell Mode Support ->External microphones work properly when MacBook lid is closed - Flexible Hotkey Routing -> With Input Monitoring: any key combo including modifier-only. Without: ⌘/⌃ requirement

## Highlights - **Stats Calculation Fix** - File transcriptions no longer penalize efficiency metrics (#102) - **Separate Live vs File Stats** - Dashboard now properly distinguishes live recordings from file transcriptions - **Ready Indicator Fix** - Fixed persistence bug when starting new recordings (#100) --- ## Bug Fixes ### Stats Calculation - Fixed efficiency multiplier dropping after transcribing audio files - Efficiency and WPM now only use live recording stats (not file duration) - Time saved includes both live time saved + full file typing time - File-only users now see time saved (100% of typing time) - Added File Transcriptions card to dashboard (shows only when files transcribed) - Keystrokes saved includes both live and file words - Added cached single-pass computation for better performance ### Dynamic Notch Ready Indicator - Fixed ready indicator persisting when starting a new recording - Properly dismisses ready indicator and clears state before showing waveform - Reduced auto-dismiss timeout from 30 seconds to 15 seconds - Added safety clear of ready indicator state when recording starts

- Apple Native Transcription - macOS Tahoe 26+ support with 33 languages - Smart Ready Indicator - Notch overlay with Cmd+V auto-dismiss (#97) - Dynamic Notch System - Native notch overlays with waveform animations - Updated UI - Modernized settings with native macOS styling --- New Features Apple Native Transcription (macOS 26+) - Native Speech framework integration - 33 languages with proper BCP-47 locale mapping Smart Ready Indicator - Notch overlay showing transcription is ready to paste - Auto-dismisses on Cmd+V detection - 5s fallback timeout when Input Monitoring permission missing Dynamic Notch Overlays - DynamicNotchKit integration - Ultra-slim waveform animations Updated UI - Modernized Settings with glass effects - Updated segmented controls Quick Trigger / Modifier-Only Hotkeys - Double-tap modifier keys to trigger recor in it.

Push-to-Talk Mode Record hands-free with our new Push-to-Talk option! Simply hold your hotkey to record and release to transcribe - perfect for quick notes and rapid dictation. Switch between Push-to-Talk and Toggle modes in Preferences to match your workflow. Improved Recording Reliability - Fixed an issue where consecutive recordings could get stuck - Faster window response during rapid recording sessions - Better handling of short recordings Ollama Integration Fix - Fixed an issue where local Ollama model selection would reset after refreshing the model list

Fresh New Look • Redesigned app icon featuring a modern waveform design • Updated menu bar icon to match the new visual identity Redesigned First-Run Experience • All-new guided onboarding that walks you through setup step-by-step • Interactive demo lets you try voice recording before you start • Clearer microphone permission flow with helpful guidance Settings Improvements • New quick reference card in Tutorial tab for easy hotkey lookup • Replay onboarding anytime with the new "Replay Tutorial" button Under the Hood • Performance improvements and bug fixes

Your transcription model now preloads automatically when oto launches, eliminating the wait on your first transcription of the day.

Enhanced AI Workflow Control Press Tab during recording to cycle through your AI enhancement prompts or disable them on the fly. A visual indicator now shows which prompt is active, giving you instant feedback and complete control over your transcription workflow. Improved User Experience • Visual enhancement indicator appears automatically when recording starts • Helpful tips added to AI Integrations settings • Better support for Parakeet transcription engine • Polished interface with consistent styling throughout Bug Fixes • Fixed enhancement indicator not updating when settings change • Improved visual consistency across settings tabs

Fixed bug where parakeet model was not using custom vocabulary