Notas de Versão

O que há de novo no DuckType.

v0.42.0

April 25, 2026

Features

Add a punch-card chart on the Dictations tab that shows your day-of-week × hour-of-day rhythm in one view. Add a theme-river chart on the Performance tab that visualises how your transcription provider mix has shifted over time.
Add a "Show romanization below this?" prompt under transcripts and skill outputs that contain Japanese, Chinese, or Korean when the romanization toggle is off. Click "Open settings" to jump to the toggle (which highlights itself), or "Dismiss" to hide the prompt for good.
Add a folder breadcrumb to the meeting detail header in place of the second back button. Each segment, including the deepest folder, is clickable.
Make transcript undo/redo per-meeting and persistent across navigation. Each meeting now has its own undo stack, so navigating away and back keeps your recent edits available to undo. Cmd+Z in one meeting can no longer reach into another meeting's segments.
Improve editing a transcript selection that spans more than one segment inside a single speaker bubble. The merged text replaces the start segment and the trailing segments are removed in one undoable step. The Edit action is now surfaced for any selection contained in a single bubble.
Improve Cloud skills speed by running them on a lighter, faster model, so post-transcription rewrites (fix grammar, translate, summarize) come back noticeably quicker.
Declutter the home page. The titlebar today/total/saved chips no longer render on the home route, the "last dictation" preview is now a single quiet line under the hero, and the Tips/Milestones panel uses a borderless layout with a plain text toggle.

Bug fixes

Fix DuckType Cloud skills (translate, extract action items, etc.) sometimes returning empty or truncated output.
Fix the home page replaying a count-up on the lifetime "Total words" stat every time you navigate back. The hero number for today still animates, and the strip total still tweens when a dictation lands mid-session.
Fix opening a folder in the meetings list sometimes snapping back out of it.
Fix the fallback recording shortcuts (Start/Stop Recording and Hold to Record, modifier + key versions) being hidden inside an "Advanced" accordion on the Keyboard Shortcuts page, so they're discoverable.
Fix the Settings tab bar overflowing into a horizontal scroll strip when the window is narrow. Tabs now wrap onto multiple rows.
Fix text selection in meeting transcripts painting the colored selection band over the gutter, footers, or the gap where alignment switches sides, so the selection stays visually contained to the words.

v0.41.0

April 25, 2026

Features

Add Reveal in Finder shortcuts directly to each meeting toolbar. The memo, transcript, and AI summary tabs each get a button for their file, and the recording sits next to the Play button. No more digging through the header dropdown.
Add transcript notes to meetings. Pin any moment by pressing Alt+Shift+N, typing in the composer at the bottom of the transcript, or clicking the pin on any speaker bubble. Notes appear alongside highlights above the AI summary, count on each meeting row, and feed into the AI summary.
Add a large "Generate AI summary" prompt at the bottom of the transcript when a meeting has finished recording, so you can kick off the summary without hopping tabs.
Add a typing indicator at the bottom of the meeting transcript while you're speaking, so it's clear the mic is picking up audio before transcription lands.
Add a Zen mode for meetings that turns the page into a note-first editor and hides transcript, recording, and AI surfaces.
Add a dismissible tip recommending headphones for dual-mode meeting recordings.
Show when an AI summary does not cover the whole meeting transcript, and indicate when the transcript has changed since the summary was generated.
Improve transcript editing and selection actions. Selecting text across multiple bubbles now shows the Copy / Highlight / Delete menu. A new Edit action lets you fix transcript text inline (Enter to save, Escape to cancel, change lands on the undo stack). The "add a note here" button now shows a plus icon instead of a pencil.
Refine the meeting Transcript tab empty state with an ambient idle waveform and tighter copy.
Default meeting detail layout to a side-by-side split, with the memo on the left and the transcript on the right.
Add app menu shortcuts for choosing Skills AI, Meetings AI, and Meetings transcription providers.
Make the Auto and Record buttons on the dictations page clearer. The Auto button shows the selected skill in its label, the dropdown splits into "Start now" and "Quick skills" sections, tooltips spell out paste vs feed behaviour, and an amber hint appears when two or more skills are selected.
Show recording time in the dictations list instead of the last-modified time, so the timestamp reflects when you actually spoke. Date headers group by recording date too.
Warn during dictation when DuckType is not detecting any microphone input, and link directly to the audio input settings.
Move the "reveal recording in Finder" button from the meeting tab toolbar into the audio player controls, next to the play and restart buttons.

Bug fixes

Fix "no audio track found" error when opening meetings where no system audio was actually captured. Empty system recordings are no longer saved, and existing empty files in your library now open with the mic track instead of a red error banner.
Fix auto mode recordings so failed segments can be retried. Audio is now written to disk (and compressed to Opus on success) like regular dictations.
Fix Auto Mode so it pastes at the cursor when exactly one skill is selected. Selecting two or more skills still skips paste, since there's no single output to paste.
Pause Auto Mode while a dictation is playing back, so the mic doesn't capture the playback as fresh segments. Auto Mode resumes when playback stops.
Improve skill result pending indicators. Skills show a "Running skills..." indicator under each transcription while they process, whether triggered by Auto Mode or the Run Skill dropdown. Single, chained, and parallel skills all show progress until each result finishes.
Fix meeting summaries and chat with Cloudflare Workers AI so streamed numeric tokens no longer trip the AI SDK validator and cut responses off mid-stream.
Update DuckType Cloud meeting summaries to use the server-reported Cloudflare Workers AI model, and show the active model in meeting settings.
Improve meeting AI model labels and show where summaries end in updated transcripts.
Fix the "Summary ends here" hairline so memos pinned within the covered range stay above the line.
Replace the fake five-part progress bar in meeting summary generation with a duck-themed loading indicator ("Rubber-ducking...", "Hatching ideas...", "Preening the prose..."), so the UI no longer pretends to know how long generation will take.
Improve the meeting summary generation state with clearer progress messaging and a summary-shaped loading preview.
Simplify the "Generate AI summary" dialog. A single reset button replaces the "Reset to template" and "Clear override" pair, the editor shows the real default prompt when no template is selected, and switching templates always loads the new prompt. The stale-summary banner button now reads "Regenerate summary" to match the dialog title.
Fix Meetings sidebar navigation so returning to the app reopens the last meeting you were viewing instead of the list.
Fix "Move to Trash" from the meetings list so it no longer opens the meeting while prompting to delete it.
Stop meeting audio playback when a recording starts, and make the pause button immediate.
Fix broken formatting of the update download progress label in the "Update Available" dialog.
Fix the meeting Summary tab so Notes & highlights scroll with the summary instead of pinning to the top, so long note lists no longer crowd out the AI-generated summary.

v0.40.0

April 23, 2026

Features

Add Sarvam Saaras v3 as a BYOK transcription provider for Indian-language dictation and retranscription.
Faster recording start on Macs with many connected audio devices, especially Bluetooth. Dictation and meeting recording start near-instantly instead of taking up to two seconds.
Change the meeting template and customise the summary prompt from the Generate/Regenerate dialog, so you can fix a wrong-template pick after a meeting without re-recording.
New meetings now default to capturing both mic and system audio. If macOS hasn't granted system audio permission yet, a warning banner asks for consent up front.

Bug fixes

Improve meeting echo cancellation. AEC now auto-detects the delay between system audio and mic capture, and the far-end voice no longer leaks into the mic track when you're silent.
Fix the Ollama model-pull progress bar snapping back to 0% at the start of each layer. Progress now aggregates across all layers and only moves forward.
Fix the update download bar flashing on Settings, General when an update is downloading in the background.
Fix the markdown editor occasionally ignoring the read-only flag on first render, so transcript and memo previews stay non-editable.
Warn Pro users when the meetings transcription or meetings AI provider is explicitly set to a non-DuckType Cloud provider.
Group built-in templates under a separator in the new-meeting picker so your own templates surface first.
Remember the last-selected bring-your-own-key transcription provider. Switching to DuckType Cloud, Local, or None and back to BYOK now restores your previous choice instead of defaulting to Groq.
Block meeting audio playback while a meeting is still recording, so system audio capture can't pick up the playback and bleed it into the recording.

v0.39.0

April 22, 2026

Features

Overhaul shortcut settings. Renamed Push to Talk to Hold to Record, and Hands-free mode to Start/Stop Recording. Both keys in the Start/Stop Recording combo are now independently customisable, and the combo can also be a single key so a short tap of (for example) Right Cmd toggles recording on and off. Modifier-only fallbacks moved into an Advanced section, and Double-Tap to Record then Pick Skill now lives alongside the other Record then Pick Skill options.
Enable the Paste Last Transcription shortcut (Alt+V) by default.

Bug fixes

Turn off Double-Tap to Record by default for new installs. Existing users keep their current setting.
Fix meetings AI provider selection so picking DuckType Cloud no longer shows a false "Provider not configured" error. Error copy for providers that still need setup no longer mentions fields that do not apply (for example, Ollama does not need an API key) and no longer reads like "OpenAI providers".
Fix tooltips not appearing when hovering charts on the Statistics page.
Fix the Record then Pick Skill shortcut so it always opens the overlay on the Skills tab, even if you last left it on another tab.
Fix Reset All in Shortcuts settings so it also turns Double-Tap to Record back off. Previously the toggle stayed on after a reset.
Reorder the onboarding Advanced shortcuts so Hold to Record appears above Start/Stop Recording, matching the primary list. Double-Tap to Record now sits at the bottom of the section.

v0.38.0

April 19, 2026

Features

Add real-time echo cancellation on meeting mic tracks so the other party's voice no longer bleeds in from your speakers. Includes an Advanced audio section with Balanced, Meeting, Whisper, and Custom presets, plus an option to bypass when headphones are detected. Only the mic track is cleaned; the recorded system-audio track is unchanged.
Show today's spoken duration alongside the word count on the home page.
Split the home-page prompt into a testimonial ask (email) and a stats-share path (X or LinkedIn).

Bug fixes

Fix statistics page charts not rendering in shipped builds on the Dictations, Meetings, and Performance tabs.
Fix a privacy-mode bug where the last-dictation preview, meeting names, and other snippets were not blurred during screen shares.
Improve home page typography and fix a layout shift: the tips menu only appears when the tips card is hidden, and the hero uses a single serif focal point.

v0.37.0

April 19, 2026

Features

Add real-time echo cancellation for dictation and meeting recordings, so the microphone track no longer picks up speaker bleed from videos, meetings, or TTS. Advanced Audio settings expose a preset picker (Balanced, Meeting, Whisper, Custom), a bypass-on-headphones toggle, and individual dials for high pass filter, adaptive gain, and noise suppression. Preset changes apply immediately, even mid-recording.
Add meeting templates. Starting a new meeting opens a searchable template picker with four built-ins (1:1, Standup, Interview, Sales call) plus any custom templates you create. Picking a template pre-fills the memo and tailors the AI summary. Manage custom templates on the new Templates page under Meetings.
Add automatic meeting title generation after recording, so the placeholder name updates live in the header when the summary runs.
Add a dedicated AI providers settings tab for language-model providers. Each provider row starts collapsed with a clear Configured badge, expanded when credentials already exist. Deep links from provider pickers auto-expand and scroll to the target row. Skills, Dictation, and Meetings pages keep their provider pickers and reference a configured provider, with a direct link to AI providers when the pick hasn't been set up yet.
Add a Skills tab in Settings so the Skills AI provider and system prompt live alongside Dictation, Meetings, and TTS settings. Dictations, Meetings, and Skills pages gain a settings gear button in the header for one-click access to the matching settings page.
Add file-explorer style navigation to Meetings. Breadcrumbs show your location in the folder tree, and each view shows only the current folder's contents. Folders appear as rows with a meeting count (including subfolders), empty folders have a proper drop-target empty state, and Cmd+Up or Alt+Left jumps up a level. Breadcrumb segments are drop targets, so you can drag a meeting up two levels in one gesture. Search still spans all folders.
Add drag-and-drop import. Drop audio or video files on the Dictations page to transcribe them, or on the Meetings page to import them as meetings. The current meetings folder now persists across navigation and app restarts.
Add an "Open folder" button in the meetings list to reveal the current folder in your system file manager.
Redesign the Statistics page with tabs for Overview, Dictations, Meetings, and Performance. Overview shows a hero streak card and combined totals. A new Meetings section covers total meetings, total meeting time, average length, words transcribed, a meetings-per-day chart (recorded vs imported), day-of-week and duration breakdowns, and an AI engagement summary. Active tab persists in the URL.
Redesign the Home page. The hero shows today's word count in a large display serif, animates up when a new dictation lands, and previews your last dictation. Start actions (Dictations, Meetings) are promoted to large feature tiles with theme-aware color tints; the rest collapse into a tidier list. Selected tabs across Meetings, Settings, Dictionary, and Mic/System/Both are unambiguous on every theme, and the display and body fonts can now change with the theme. Tokyo reads in mono, Bold in condensed Helvetica, Absolutely in Didot.
Share a combined stats link that aggregates your totals across every device you've signed in from. The Share dialog now uses this combined link by default.
Show the monthly reset date alongside cloud word usage in DuckType Cloud settings and the Account page.
Add a dismissible testimonial banner on Home for engaged users. Signed-in users can tap "Share your stats" to open the stats share dialog with a rich tweet snapshot; signed-out users fall back to a plain X post.
Highlight recommended local models with a floating "Recommended" badge, and give the local-model picker a consistent language search across onboarding, Local Models settings, and the DuckType Cloud offline-fallback section.
Preload the selected local transcription model into memory when recording starts. The cold-start load now runs in parallel with audio capture, so by the time you finish speaking the model is usually ready.
Offer cloud transcription when the local model runs slowly. Free users on the offline provider see a dismissable inline prompt above the dictation feed after a few transcriptions where wall-clock time ran well over audio duration. Never shown to Pro users or anyone using their own API key.
Show a heads-up during onboarding when an offline model is selected, explaining the first dictation takes a few seconds while the model loads and that the model will reload if it has been idle.
Default copy-to-clipboard on for new users, and highlight paste-at-cursor as the recommended text-output option during onboarding with a floating "Recommended" badge.
Auto-configure DuckType Cloud as the Skills LLM provider for signed-in users. Onboarding sets it when the user picks the online option, and existing users who finished onboarding without one are back-filled on launch. Meeting summaries and chat now support DuckType Cloud and Cloudflare Workers AI in addition to OpenAI, Anthropic, OpenRouter, and Ollama.
Improve meeting AI setup. Pro users get DuckType Cloud auto-configured for AI summaries on sign-in. The Meetings page shows an in-context prompt when no AI provider is set: free users see an upgrade CTA, Pro users see a configure link. Meetings can now use a different LLM provider than Skills via a new AI provider picker in meeting settings.
Gate DuckType Cloud for meetings behind Pro to avoid quota-exhaustion footguns. Free users can no longer pick DuckType Cloud for meeting transcription or summaries, and see a clear upgrade prompt if their dictation or skills selection would have inherited it.
Label meeting transcript speakers as "Mic" and "System" to match the audio-source switches in the header.
Render GitHub-flavored Markdown (task lists, tables, strikethrough) in meeting chat and other Markdown bodies.
Warn before regenerating a meeting's AI summary, with an option to keep the previous summary as a timestamped backup in the meeting folder.
Drop the YYYYMMDD_N_ prefix from meeting filenames. New meetings save as Meeting name.memo.md / .ai.md / .transcript.md, with a _2, _3, … suffix only when a collision exists.
Replace "Last year" in Statistics time range with calendar-year options covering January through December.
Enable prompt caching on meeting summary and chat flows so regenerating or following up reuses the cached memo and transcript context instead of re-billing it.
Tag @getducktype in the prefilled tweet when posting a stats share on X.

Bug fixes

Fix transcription, meetings AI, and skills sometimes falling back to the slow local model on app launch. The auth client now has a 15 second timeout so session validation can't hang, the cloud LLM path proactively refreshes the signed session token, and the cached cloud-usage short-circuit only triggers after the server has confirmed usage at least once this launch.
Fix DuckType Cloud LLM ignoring the configured temperature when summarizing meetings.
Show a critical in-app dialog with links to logs, email support, and Discord when the database fails to open at startup.
Fix meeting AI summary reliability: the memo no longer shows blank after revisiting a meeting, the refresh button works from any tab, streaming errors surface in the UI instead of failing silently, and generation works across OpenAI, OpenRouter, Ollama, Groq, Mistral, Cloudflare Workers AI, and DuckType Cloud.
Fix the meeting transcript tab rendering blank when you return to it from Summary or Chat while scrolled down. It now re-renders at the preserved scroll position.
Fix statistics page getting stuck on the loading skeleton when the app launches directly on that page.
Fix titlebar stats appearing blank in production until visiting the Statistics page.
Fix the Milestones card on Statistics appearing a moment after the rest of the page. It now animates in alongside the key metric cards.
Fix a bug where renaming a newly-created note in the meetings list would leave a duplicate at the old name.
Fix an extra space between "OS sound settings" and the trailing period in the audio device captions on settings and onboarding screens.
Fix a bug where choosing offline or local models during onboarding could leave the active transcription provider set to DuckType Cloud.
Clicking the "Auto-run skills failed" notification now jumps to the failing transcription's skill results panel on the Dictations page.
Hands-free fn hold now auto-starts the moment you grant accessibility permission, and the "why can't I use fn here?" tooltip no longer gets picked up when copying shortcut descriptions.
Improve meeting chat answers. The assistant sticks to the provided notes, summary, and transcript, avoids re-pasting them back, and when asked to "edit" the summary or memo it replies with a short paste-ready snippet instead of rewriting the whole document.
Compact the meeting detail header while recording so the live transcript and controls get more vertical space.
Move the meeting page header overflow menu down to sit with the other toolbar controls instead of floating on the right of the title.
Polish meeting UI: click anywhere in the memo to start editing, match memo text selection to the app's teal colour, constrain the AI summary to a readable column width, and fix the transcript "Jump to latest" button so it reliably scrolls to the bottom.
Meeting AI summary no longer shows a drag handle or plus icon when you hover over lines.
Meetings count on the home page and Statistics page now reflects every meeting you've ever created, not just meetings currently stored.
Cap meeting memo, summary, transcript, and chat content at a readable width on ultra-wide windows.
Clean up meeting settings copy. The transcription-provider dropdown no longer restates the selection below itself, Local Models sits at the top of the dropdown with DuckType Cloud directly below (locked for free users), and the Auto AI summary description correctly refers to an AI provider instead of a Skills provider.
Add an "Open on website" button to the in-app Release Notes page for a shareable, browser-friendly view.
Unify the DuckType Cloud dictation card with Account settings. Signed-out state offers Register and Log in, and "Manage plan" and email verification link to the in-app Account page.
Move transcription provider settings back into the Dictation settings page. AI providers now holds only language models.
Clarify upgrade banner copy: meetings and AI summaries work for free with local models; Pro adds faster, more accurate cloud transcription and more powerful AI models.
Use the standard tab switcher for Milestones and Tips on the home getting started card.

v0.36.0

April 18, 2026

Features

Add AI Chat for meetings. Ask questions about a meeting while it's ongoing or after it ends. Meeting Summary and Chat now stream directly from the frontend.
Add prompt to regenerate AI summary when stopping a meeting that already has one.
Add a meetings recording retention setting so meeting audio can be auto-deleted after a chosen period.
Add estimated storage sizes for kept dictation and meeting recordings in settings.
Add per-device stats sharing. Each machine now creates its own stats share page with a unique URL, and the website dashboard lets you rename devices.
Improve the home page by moving tips and milestones into a tabbed card with Milestones and Tips tabs. A dropdown lets you close the card for today or permanently, and the page heading dropdown brings it back. Tips now advance on a new calendar day in your local timezone.
Improve settings by showing the "Menu bar icon color" option directly under Appearance when the DuckType bar is in the menu bar. Default for new installs is now Dark (white icon).
Improve system audio permission flow with grant and re-check buttons.
Improve the onboarding welcome screen with a gentle float animation and hover wobble on the DuckType logo.
Improve resizable panels by replacing the drag grip with a subtle line that highlights on hover.
Improve meeting system audio on macOS to use the lighter "System Audio Recording Only" permission instead of Screen Recording. Requires macOS 14.4 or later.
Add an occasional upgrade suggestion for free users after you've gotten real value from DuckType. Dismissable, and never shown to Pro users.
Add Speechmatics as a transcription provider.
Add Gradium as a transcription provider.
Remove Fireworks AI transcription provider. Fireworks is shutting down its audio endpoints on May 26, 2026.

Bug fixes

Fix macOS clipboard restoration after paste-at-cursor so images and other rich clipboard contents can be restored.
Fix microphone fallback when disconnected preferred devices are no longer available.
Fix crash and failed relaunch after applying an update.
Fix "Meeting not found" on cold start when the database is still initializing.
Fix the dictations feed sometimes showing "The pond is quiet" on startup before the database finished initializing.
Fix push-to-talk and hands-free mode not working after granting accessibility permission while the app is running. Previously required toggling the setting off and on.
Fix recording streams not stopping when navigating away from a meeting.
Fix settings page navigation overflowing on narrow windows.
Fix auto-delete removing audio from interrupted recordings before the user has retried transcription.
Fix statistics page loading in laggy, staggered bursts. Charts now mount after the key metric cards animate in, making the entrance and counting animations smooth.
Fix meetings chat failing to reply after an earlier error. Added a typing indicator and a jump-to-latest button when you scroll up during a response.
Fix scratchpad viewport snapping to the bottom when other scratchpads are renamed, created, or deleted.
Fix instant recording standby opening on a disconnected input device at startup. Falls back to the system default or defers until your preferred device reappears.

v0.35.0

April 9, 2026

Features

Add meetings. Record with mic and system audio, get live transcripts, and browse a meetings list with local playback.

Bug fixes

Fix DuckType bar appearing off-center when the macOS dock is on the left or right side.
Fix scrollbar appearing at the content edge instead of the panel edge on pages with centered content.
Fix pending transcriptions showing a misleading loading animation when no transcription provider is configured. Now shows a clear message with a Transcribe button.
Fix "Some dictations failed" banner taking full page width.
Show an error notification with the device name when recording fails to start.