Sürüm Notları

Features

• Add Fast and Accurate models to DuckType Cloud. Fast is the lowest-latency option. Accurate uses GPT Transcribe and holds up in background noise, strong accents, and technical vocabulary. Both support live streaming dictation, and the previous model stays available as Batch (Legacy).
• Default DuckType Cloud to Accurate. Existing installs move across once, and Fast and Batch stay selectable in Settings, during onboarding, on the dictations page, and in the expanded DuckType bar.
• Simplify the dictations header into a single model picker beside the Record button, showing the provider and model you are dictating with. Its menu holds the provider, a slider between the DuckType Cloud speeds, the spoken language, the AI provider behind styles and skills, and a switch for AI Edit.
• Add a provider picker to the meetings header, so meeting transcription and meeting AI are chosen on the page they affect.
• Lighten the dictations header. Auto mode moved into the Record menu with its presets and per-dictation skills. Search, sort, and filter no longer hold a permanent row: search opens from the header, the title bar, or the find shortcut, and sort moved into the overflow menu. Your speaking pace for the day now sits next to the Dictations title.
• Declutter dictation rows. The Style, Skill, and Speech badges no longer sit under every dictation. Restyle, Run skill, and Read aloud moved into the hover action bar, grouped into playback, text, and overflow. The style pill still appears when a dictation used something other than your default style.
• Move dictation lists onto their own page, opened from the dictations overflow menu. Click a list to filter the feed to it, with a breadcrumb showing where you are, and the list is part of the address so the back button works.
• Put search in the middle of the title bar, replacing the word counts and time saved, which are still on the Statistics page. When microphone or accessibility access is missing, that warning takes the slot instead.
• Collapse cancelled dictations into a single expandable card with a bulk delete action, matching failed and empty dictations. The empty dictations notice can now be dismissed.
• Add an optional hidden macOS text input helper for live dictation previews without changing the active keyboard source, with guided macOS authorization and a native drag-and-drop Accessibility setup panel.
• Add accurate, phrase-aware Jyutping romanization for Cantonese dictation.
• Change the default AI Edit trigger phrase to "Run AI". Existing trigger phrases are left alone, and AI Edit is no longer marked experimental.
• Simplify the display menu. Zoom moved into a submenu showing the current level with a "Reset to 100%" item, and colour themes moved into their own submenu with previews. The input device picker, overlay toggle, and screen sharing privacy toggle are gone. Choose your microphone in Settings, Dictation, Audio, or from the recording panel, and screen sharing privacy stays in Settings, General, Appearance.
• Move "Transcribe files" and "Add dictation from text" to the top of the Record menu.
• Show the audio duration and model badges only when Stats for nerds is enabled. The model badge still appears on its own for dictations that are pending, running, failed, cancelled, or fell back to another provider.
• Remove starred dictations. The star button, its menu action, the S hover shortcut, the selection bar button, and the Starred only filter are all gone, and existing star marks are discarded. Use lists to keep dictations you want to come back to.
• Remove reading mode, since the default dictations list is now clean enough without it.
• Remove the Edit previous button.

Features

• Add "Add dictation from text" to the Record menu so you can paste or type text directly into a dictation.
• Add a configurable spoken command to send dictations. Say "enter" as its own final sentence, or choose a custom trigger phrase in Dictation settings.
• Improve DuckType bar feedback for spoken formatting commands and automatic actions applied after dictation.
• Improve meeting audio imports with named progress stages, percentage complete, estimated time remaining, and progress that stays visible when you return mid-import.
• Add OpenAI GPT Transcribe as the default OpenAI model and GPT Live Transcribe for live dictation, with clearer model generation labels.
• Improve long recording and contiguous file-group imports by recommending meetings while keeping unrelated files as separate dictations.
• Add experimental AI Edit with a configurable spoken trigger for corrections and prompted text generation.

Bug fixes

• Fix renaming notes, folders, meetings, and meeting summaries when only the letter casing changes.
• Fix typing spaces in dictation dialogs such as "Convert to meeting".
• Fix dragging audio files into the meeting import dialog and prevent long file names from widening the dialog.
• Fix generated meeting summaries forcing the scroll position back to the bottom after you scroll up.
• Fix meeting titles shifting when the saving indicator appears or changes.
• Fix retranscription leaving spoken formatting commands such as "new paragraph" and "new line" as literal words.
• Fix automatic text insertion fallback duplicating a dictation after an unverified paste.

Bug fixes

• Fix speech playback when replaying or switching dictation entries from a transcription menu.
• Show unavailable local AI models on older macOS versions with clear system requirements instead of hiding them.
• Preserve existing DuckType Cloud batch settings while migrating to its stable model identifier.

Features

• Add a Customize hub for dictionary entries, shortcuts, styles, and skills.
• Add default-on voice commands for line breaks and pressing Enter after live dictation.
• Add Duck Context for drafting text from the focused app with review before inserting or copying.
• Add controls to copy and reveal saved API keys in settings.
• Add a cleanup action for cancelled dictations.
• Add a Duplicate action to shortcut menus.
• Add ElevenLabs Scribe v2 Realtime for lower-latency recording with dictionary keyterms.
• Add a separate default transcription provider for audio files, video files, CLI transcription, and meeting imports.
• Add a Groups option to the dictations filter menu.
• Add a confirmation prompt when quitting during an active meeting recording.
• Add multi-word dictionary entries and phrase hints for supported providers.
• Add broader settings search coverage for meetings, sounds, storage, providers, and shortcuts.
• Add live transcript previews to the dictation feed and DuckType bar while recording.
• Add continuous meeting transcription with speaker-labeled remote participants for supported cloud providers.
• Add an option to show DuckType's identity in macOS Mission Control and Spaces.

Bug fixes

• Default live meeting speaker separation to off while keeping it available as an option.
• Preserve live meeting transcripts when word timing is unavailable or a streaming connection fails.
• Show the microphone selected from the active priority list in the navigation menu.
• Prevent automatic text insertion from duplicating a dictation after it is quickly submitted or focus moves elsewhere.
• Restore dictation recordings when deletion is undone and remove the macOS Finder permission requirement for deleting recordings.
• Recover microphone capture after macOS device changes interrupt the input stream.
• Keep DuckType running on older macOS versions by showing only compatible local AI engines.
• Add a macOS app menu toggle for the Command-Q quit warning.
• Improve Duck Context capture, browser content detection, replies, and input safety.
• Improve ElevenLabs Realtime guidance, finishing latency, model status, shutdown handling, and keyterm limits.
• Improve streaming dictation accuracy around natural pauses and interim transcript revisions.
• Improve DuckType Cloud AI latency with Fireworks-hosted GPT OSS 20B for skills and styles and DeepSeek V4 Flash for meetings, with readable model and hosting labels.
• Improve Cloudflare Nova transcription accuracy with dictionary keyterm hints.
• Improve speaker volume reduction with smooth fades while dictating.
• Improve Home and Customize with clearer examples, consistent motion, and theme-aware duck artwork.
• Improve audio file action labels for original, voice-only, microphone, and system recordings.
• Improve dictation audio controls and collapsed group previews in narrow layouts.
• Improve failed cursor insertion recovery with typing retries and Paste Last guidance.
• Improve imported audio group names, ordering, and reliability when the source device disconnects.
• Improve meeting summaries, active meeting rows, transcript notes, and automatic-finish notices.
• Improve long meeting transcript scrolling and keep transcript bubbles stable while speech is active.
• Improve DuckType bar previews when a meeting indicator is visible.
• Improve Customize navigation with simpler controls and breadcrumbs.
• Improve dictionary and shortcut lists with manual ordering.
• Improve the Soft Pop theme with a distinct main-page surface.
• Fix authenticated Ollama servers for model lists, meeting summaries, and meeting chat.
• Fix paste-at-cursor reliability on busy machines and when restoring rich clipboard contents.
• Fix app-specific writing styles reusing the previously focused app's style.
• Fix Home shortcut card alignment across Playful and Professional styles.
• Fix active recordings switching to a newly connected preferred microphone.
• Fix a macOS crash when quitting from the app menu with the mouse.
• Fix macOS microphone recovery when audio capture starts without returning samples.
• Fix meeting headers wrapping the Back button too early.
• Fix meeting search so matching previous meetings appear while searching.
• Fix onboarding transcription setup when neither cloud nor a local model is selected.
• Fix the provider setup banner opening the old settings page.
• Fix shortcut replacements repeating when expanded text contains the trigger.
• Fix saved-recording transcription so an available provider can be selected when none is active.
• Fix the Models & Providers page showing a saved message before anything changes.
• Fix recording shortcuts getting stuck when another DuckType build is running.
• Fix Stream Deck shortcuts that emit Option+/ on macOS.
• Fix Windows Ctrl+Win shortcuts starting recording from stale key state.
• Fix the collapsed sidebar notification signal clipping.
• Fix date, time, and daily statistics after the system time zone changes.
• Fix Granola imports missing transcripts and allow stale imports to be refreshed.
• Fix collapsed meeting transcript bubbles showing the total instead of hidden word count.
• Fix the dictations feed jumping after deleting an item while scrolled down.
• Fix DuckType Cloud defaults, remove misleading per-minute pricing, label streaming as experimental, and clarify batch versus streaming model names.
• Fix Deepgram and Cloudflare Nova requests to opt out of provider model improvement programs.
• Fix automatic meeting-finished notices and active meeting restoration after restart.
• Fix meeting detail panels closing when a recording starts.
• Fix manual meeting notes not scrolling into view after being added.
• Fix the main content gutter when the sidebar becomes a mobile drawer.
• Fix transcript playback continuing after the DuckType window closes.
• Fix saved audio playback continuing when a new dictation starts.
• Fix deleting a dictation with multiple saved audio files so they move to Trash together with one sound.
• Remove redundant provider and preparation controls from the dictations header.
• Remove the redundant meeting summary regeneration warning.

Features

• Add a new Models & Providers settings page for dictation, local models, AI providers, and meeting settings.
• Add offline fallback for DuckType Cloud and custom API keys, so dictation can switch to a local model when a cloud provider fails.
• Add support for local .gguf AI model files for skills and styles, with the needed engine bundled.
• Add test buttons for custom provider API keys.
• Add Deepgram smart formatting and more Nova 3 language options.
• Add ducktype-cli login, logout, and whoami so CLI users can sign in through a browser without opening the desktop app.
• Add inline renaming and drag-and-drop organization for Dictionary and Shortcuts lists.
• Add optional speaker volume reduction during dictation and an option to show the DuckType bar in screen shares and recordings.
• Add Support page shortcuts for opening logs and app data folders.
• Add a quick new-scratchpad button next to the scratchpad selector.
• Add Home announcements for launches, updates, and product notices.
• Add left and right arrow shortcuts for browsing dictation versions.
• Add style details showing the model used for AI dictation style rewrites.
• Add Cohere Transcribe 03-2026 as a downloadable local transcription model, and mark it as very slow.
• Add Cartesia Ink 2 for live dictation, and note that it does not support DuckType dictionary hints.

Bug fixes

• Fix macOS Quit menu and tray quit actions so they quit immediately, while Command-Q still asks for confirmation.
• Fix speaker volume reduction so it limits system volume to the selected level.
• Fix cloud fallback so provider errors are shown when local fallback succeeds, and live dictation can use the selected local model.
• Fix provider changes so live dictation immediately uses the saved language and model.
• Fix quiet dictation so silence is less likely to become made-up words, while quiet speech can still transcribe.
• Fix a crash when macOS reports an invalid microphone input format.
• Fix Paste Last Transcription so it restores the previous clipboard contents after pasting.
• Fix grouped dictation actions for playback, deletion, and stable row position after deleting the top item.
• Fix failed and empty dictations so they collapse into clearer summary rows.
• Fix Dictionary and Shortcuts pages so external edits refresh automatically and list editing is more reliable.
• Fix offline calendar refresh so stale meeting list entries are cleared.
• Fix re-transcription controls so model settings are easier to find and choices remain available when no provider is active.
• Fix Local GGUF AI rewrites when model paths use a home-relative ~ prefix.
• Fix smart dictation styles so short Email draft and Terminal dictations skip AI rewrites.
• Fix Text-to-Speech settings so "Saved" no longer appears before a setting changes.
• Fix shortcut editing so Command-Enter or Control-Enter saves from either input, and right Option is handled correctly on macOS.
• Improve DuckType Cloud transcription reliability with smarter upload timeouts.
• Improve local model downloads so multi-file progress is more accurate.
• Improve settings pages with clearer layouts and visible Saving/Saved feedback.
• Improve Style settings with clearer behavior controls and smarter local cleanup for short messages.
• Improve provider and model badges so they stay short outside model selection.
• Improve the DuckType bar with a wait state while the microphone opens and a clearer command-palette icon.
• Improve dictation audio settings with simpler sound controls and faster starts when speaker playback reduction is enabled.
• Improve tooltips so the pointer no longer covers content or shortcut labels.

Features

• Add Streaming and Batch dictation modes for Deepgram and Speechmatics, with live dictionary hints and expanded Deepgram model choices.
• Add a Dictionary menu to the DuckType bar so selected text can be saved to an editable dictionary list.
• Add DuckType bar visibility controls, including temporary hide durations, always-visible and dictating-only modes, and show-again controls.
• Add preview-first import support for Granola CSV exports.
• Add a setting to pause other media while DuckType plays audio.

Bug fixes

• Improve dictation insertion speed by moving nonessential finalization work out of the paste path.
• Improve provider guidance for dictionary hints, microphone issues, shortcut badges, and selected-text actions.
• Improve settings access from Notes and Meetings, and make the BYOK provider picker searchable.
• Keep DuckType bar audio controls in sync after recording audio settings change elsewhere.
• Keep the Notes window open after closing its last tab.
• Fix latency tooltips so selecting tooltip text does not drag dictation rows.
• Fix repeated paid-plan DuckType Cloud reminders so dismissals stay stable across window reopens.
• Refresh Skills and tray actions when skills.md is edited outside DuckType.
• Show text-only dictations under a dedicated Text source category instead of grouping them with recorded dictations.

Features

• Add a reading mode that focuses the dictations list on completed dictation text.
• Add selected text support to the DuckType bar for skills, restyles, and voice edits.
• Add playback for saved no-text dictations and shortcuts to text-to-speech provider settings.
• Add a DuckType Config agent skill and CLI commands for managing DuckType skills, dictionary entries, and custom dictation styles.
• Add Anarlog meeting import alongside Granola, with source selection and refreshed import previews.
• Add a macOS quit confirmation prompt for accidental Command-Q presses.
• Add a setting to include all timed calendar events in reminders and the tray meeting list.
• Add Original and Voice playback choices for imported audio, with consistent volume leveling across both modes.

Bug fixes

• Improve imported dictation ordering when file metadata is skewed into the future.
• Fix dialogs overflowing short windows by allowing modal content to scroll.
• Improve the Developer Settings agent skill installer with clearer installed-command details and the standard DuckType segmented control.
• Use clipboard paste when DuckType is frontmost or the insertion target cannot be identified.
• Allow completed dictations to be corrected to empty text.
• Improve tray meeting countdowns so long waits show hours and minutes with clearer start and end wording.
• Fix Granola import to find IndexedDB meeting data inside nested profile folders.
• Fix batch imported dictations to keep source file recording order instead of transcription completion order.
• Keep reading mode timestamps and row actions visible while hiding badges and extra details.
• Keep the Meetings sidebar, Home tile, command palette, and keyboard shortcut on the meetings list after returning from a meeting.
• Fix meeting search so it stays focused on the selected meeting list.
• Fix list search fields resetting after opening an item and navigating back.
• Add row double-click to open or create meetings from list results.
• Repair legacy meetings that had transcripts or recordings but no notes file, so notes, Reveal in Finder, and AI summaries work again.
• Improve Automatic insertion for macOS Screen Sharing, remote desktop, virtual desktop, and remote support apps.
• Show DuckType's built-in app insertion behavior in Dictation settings.
• Fix automatic skill output stats metadata so latency appears reliably.