Mouthtype 0.4.0

One window. Mouthtype now lives in a single redesigned window with a new palette and SF Symbols in the sidebar: Home, Stats, Dictionary, Transforms, Style, Appearance and a Settings modal. Saved spellings and Phrases are pages in that window instead of separate windows, and the shortcut recorder is a sheet on it. Leaving a page or closing the window with an unsaved draft still asks first. Set Up Mouthtype and Transcribe Audio File keep their own windows, since the menu bar opens them when no Mouthtype window is up. The Import CSV sheet for the dictionary and Set Up Mouthtype use the new look too.

Transform Selection. Choose it from the menu bar to clean up a selection or run Fix grammar, Make shorter, More formal or More casual on the local Thorough model. A preview panel shows your original with removed words struck through and an editable result, and you choose Accept, Copy result, Retry or Cancel. Accept replaces the text only if the app, field and selection are still the same. Clean up runs without the model. The panel doesn't take focus from the app that owns the selection. Saved phrases and Very Casual are no longer in this menu.

Transforms page. It lists your phrases, each with a kind tag (text, lines, clipboard or date), and the five selection transforms with their model status. New transform opens a sheet with + Date and + Clipboard.

Transcribe Audio File now opens in its own window, from the menu or from Home. It shows the file's name, length and size, the raw transcript with a word count, and Copy and Save Text. A failed import keeps the previous transcript.

Automation. mouthtype:// URLs and a mouthtype command-line script can start, stop, toggle, cancel, copy-last and paste-last, and open a page of the window. The dictation commands are off until you turn on the automation switch, because any web page can open a mouthtype:// link. Opening a page works without it.

Editor hints. With Boost my saved words on, Mouthtype can add up to 40 code names from VS Code, Cursor, Xcode and Terminal (identifiers near the cursor, the file and project names) to the boost for one take. It's off by default and nothing is kept after the take.

Quiet other audio is now on by default. If you turned it off before, it stays off.

Menu bar: the menu is plain text now, with Open Mouthtype, Settings, Paste Last Take, Copy Last Take, Recent Takes, Transform Selection, Transcribe Audio File and Saved Recordings.

The corresponding source also adds a history limit and audio retention to the experimental Linux client. Linux remains source-only and is not a supported release.

Not verified on a device for this release: fresh installation, VoiceOver, and Bluetooth and USB microphones.

Earlier in 0.3.0:

Home. Open Mouthtype from the menu to get a new Home page with your last 50 takes, newest first. Each take has Copy and Recover. Recover works for any take whose audio is still saved. It offers the text again without changing the history entry. Home updates as you dictate.

Stats. The new Stats page shows words today, this week and all time, a 14-day chart, your average and best speed, your current and longest streak, your biggest day, your longest take, and the time you saved compared with typing at 40 WPM. Records show your fastest speed, biggest day, longest streak and longest take, each with its date. Everything is worked out on your Mac from your own history. Takes from secure fields are never counted.

Saved spellings now suggest spellings from your recent takes: names it keeps mishearing, brand names that come out lowercase, and words that get split apart. Each suggestion comes with examples. Choose Add, Remove or Dismiss, and Undo is there right after. Saved spellings also flag saved heard forms that rewrite everyday words.

Settings: with Keyboard navigation on, the page now scrolls to follow Tab and Shift-Tab.

Not verified on a device for this release: fresh installation, VoiceOver, and Bluetooth and USB microphones.

Earlier in 0.2.0:

Transform selected text. Select some text in any app and choose Fix grammar, Make shorter, Make more formal or Make more casual. The Thorough model rewrites it into an editable preview, and you choose Accept, Retry or Cancel. Accept replaces the text only if the app, field and selection are still the same. If they changed, Copy original is there, and Cmd-Z undoes a replacement.

Boost my saved words while recognising is a new setting and it's off by default. It favours your saved spellings while Mouthtype is recognising speech. The first time you use it, Mouthtype downloads an English speech model of about 100 MB. Dictation never waits for that download.

App and website rules can now set a cleanup level as well as a tone, so cleanup can be on for one app while it's off everywhere else.

Formal tone now removes or softens swearing at Basic, Fast and Thorough.

Saved spellings warn you when a heard form is an everyday word, like "win". Those aren't used as boost hints.

Settings: with macOS Keyboard navigation on, Tab now reaches the page controls and Shift-Tab steps back. Sidebar rows show focus.

Appearance: the preview now offers Light and Dark, and Desktop is gone. Glass no longer judges itself against Apple's placeholder picture when your wallpaper is a live one.

Not verified on a device for this release: fresh installation, VoiceOver, and Bluetooth and USB microphones.

Earlier in 0.1.0:

The first stable release of Mouthtype for Apple Silicon Macs on macOS 26 or later. Hold Fn, speak, release, and clean text is pasted where you're typing. Recognition and cleanup run on your Mac.

Since rc.4: if you quit Mouthtype in the middle of a take, the next launch offers to recover it. The docs now say plainly that Tab doesn't reach Settings controls yet; arrow keys move through the sidebar. That is fixed in 0.2.0.

Not verified on a device for this release: fresh installation, Bluetooth and USB microphones, and VoiceOver.

Earlier in rc.4:

RC.4 fixes what the rc.3 reviews found. Recognition and cleanup are unchanged.

Dictation: pressing Esc to cancel a hands-free take no longer swallows the next Fn press. Undo on a card now works while an earlier take is still processing. If cleanup times out, the fallback text no longer brings back tags like [MUSIC]. A {clipboard} phrase now inserts what you copied, not your previous dictation.

Settings: arrow keys move through the sidebar even with macOS Keyboard navigation off, and with it on Tab moves from the sidebar into the page.

Menu: Check for Updates, How to Use and About have icons that match Quit.

Long takes start transcribing sooner, Mouthtype stops watching very large fields for corrections, and removing old recordings no longer leaves stale entries when one file can't be deleted. Cleanup model downloads are pinned to exact versions.

The docs add troubleshooting and known limits.

Earlier in rc.3:

RC.3 fixes what the rc.2 reviews found. Recognition, cleanup and how text is pasted are unchanged.

Pill and cards: a take that finishes while an earlier card is up no longer hides that card's "Your text is safe" line, and the processing pill comes back when that line goes while a take is still being worked on. The recording pill no longer sticks after a card arrives mid-recording. Cards sit above a message pill with a detail line instead of overlapping it. Long Classic messages scroll instead of growing past the screen; with Reduce Motion they jump to the end instead. Large message text is the same size in every font, and the Appearance preview readout scales with Large.

Glass: the glass pill no longer renders smoked on the first press after changing style or size in Settings, and an older wallpaper measurement can't replace a newer one.

Hi-Fi card: text the pixel font can't draw, such as non-Latin scripts, emoji and most symbols, shows on the Classic card instead of as question marks. Windows line endings display correctly. The close button is smaller and easier to hit, the title bar has the ear icon, arrows draw as arrows, and an open card follows readout colour changes.

Menu bar and setup: the menu bar icon retries if its image fails to load, and setup notices a Globe key conflict change within a couple of seconds.

The in-app Read Me is renamed and its links work outside the repository. The docs add a spoken commands reference, third-party notices list each dependency and model license, and the Gemma 4 cleanup model is listed under Apache License 2.0.

Earlier in rc.2:

RC.2 is mostly a visual refresh. Recognition, cleanup and how text is pasted are unchanged; the pill and delivery cards look and animate differently, and update checks now default on.

The menu bar icon is now a pair of lips that stays closed when idle and opens with your voice while recording. Reduce Motion holds a single frame. The app icon has new art, larger lips and Icon Composer layers, so Dark, Clear and Tinted appearances render. The first-run setup is a glass window with step dots, and Settings has a redesigned layout with a glass sidebar. Both use a solid background when Reduce Transparency is on.

Pill size is Automatic, Standard or Large and applies to every style. Automatic picks Large on screens at least 2000 pt wide. Pill style now offers Classic, Hi-Fi, Glass, Matrix and Minimal. Classic stays the default. Glass uses the system glass material and falls back to a solid body only for Reduce Transparency or when the material is unavailable. Matrix and Minimal are compact styles. Delivery cards are drawn for each style and scale with pill size, and Hi-Fi has its own pixel-art card window. Hi-Fi also has a readout colour: Amber, Cool white or Red. Appearance shows a live preview of the pill over a Desktop, Light or Dark swatch.

Update checks are now on by default and no longer ask first. Turn them off in Settings → General. Anyone who already answered the earlier prompt keeps that choice. You still choose when to download and install; automatic installation stays off.

The corresponding source also includes experimental Linux file transcription work. Linux remains source-only and is not a supported release.

Not yet verified for this build: fresh installation, behavior in third-party apps, physical-device acceptance of the new interface, and actual update installation and relaunch over an older copy.

Earlier in rc.1:

RC.1 fixes audio quieting restoration when Core Audio reports the app’s own mute after observation starts. Observed unmute/remute or volume changes still relinquish restoration. Native output checks cover the built-in speakers; other hardware remains unverified.

Daily-use hardening from beta.1: ordinary delivery waits for the previous clipboard read or returns recoverable text instead of overwriting it. Normal Quit preserves an active or queued take before settling an owned clipboard; a failed save keeps the take for retry. Saved Recordings loads on demand in pages, with older takes still reachable. Capture starts before detailed Accessibility context lookup. Missing cleanup assets trigger repair on explicit Download; Basic fallback honors saved spellings and the spoken-list switch. Cleanup diagnostics omit rejected text. These changes retain existing focus and secure-field checks.

Website tone rules now match HTTP/HTTPS domains and subdomains before app/default tone when a supported browser exposes its document URL. Missing URL support falls back to app/default; document URLs are not retained.

Microphone Priority list is opt-in. Choose an ordered list of inputs and separately enable System Default fallback if wanted. Fixed-device choices still report an unavailable input without silently switching. The list applies at the next take, not mid-recording.

Clean Selected Text opens an editable Basic cleanup preview. Accept requests replacement only while the original app, field, range and selected text still match. Unsupported or changed selections keep Copy result available. No AI rewrite, automatic paste or retry.

Paste delivery rechecks focus and secure input after Accessibility reads, before clipboard writes or key delivery. A focus-change abort keeps the transcript recoverable; a secure destination discards it. This closes a reproduced preflight gap, without claiming the intermittent Chrome incident is explained.

Choose how long to keep completed recordings in Settings → General → Saved audio: 14, 30 or 90 days, or No age limit, with a 2 GB or 10 GB size limit. Defaults stay 90 days and 2 GB. Changes apply at the next launch, to completed Mouthtype audio only. Text and timings stay; unfinished recordings and recovery audio are protected. Previously removed recordings cannot be restored.

Microphone readiness now follows the actual capture result, so setup reflects successful starts and input errors. A missing fixed device still requires reconnecting it or explicitly choosing another input. Priority-list fallback is a separate opt-in.

Includes single-file WAV/AIFF/M4A transcription, remembered microphone selection, optional Quiet other audio and optional Live words. Quieting and provisional words remain off by default; the completed take is inserted once. Imported audio is preserved and its raw result is exported only by explicit Copy or Save Text.

The corresponding source includes an experimental Linux daemon/CLI and local Arch test-package builder, with saved spellings, phrase replacements and optional deterministic basic cleanup. Cleanup is off by default and does not add a model. Recovery-only delivery remains the default; automatic typing refuses multiline/control-bearing results and keeps the full transcript recoverable. The source also includes an experimental Omarchy bar plugin with a floating pill; there is no supported Linux desktop release. Physical Arch/Omarchy, Hyprland and microphone acceptance remains open.

Requires Apple Silicon and macOS 26 or later. This is a release candidate; review text before sending. Microphone changes, supported output hardware, provisional display, real file-recognition quality, fresh installation and actual update installation/relaunch still need coordinated testing.
