What's new in TypeVox

Faster autonomous cleanup

• The transcription engine now releases its GPU and Neural Engine resources after 10 minutes of inactivity by default (instead of 2 hours), so your Mac feels fluid again much faster after each dictation session.
• New 5-minute and 10-minute options in Settings → Advanced → Performance for users who want an even more aggressive cleanup.

Mac stays fluid through every kind of session

After several rounds of progressive optimization, this release closes the chapter on the long-uptime drain. Your Mac stays smooth, and your dictation feels natural.

What changed

Background activity is now truly idle. The dashboard's background animations no longer keep running when its window isn't visible. Internal caches (changelog, regex, mode resolution, transcription metadata) are now properly capped or cleared between dictations so memory doesn't creep up over days of use.

Audio spectrum follows your voice. The visualizer drops back to its baseline as soon as you stop talking, and reacts faster to soft speech. No more residual animation when you're silent, no more frozen bars when you're whispering.

Stronger guard against AI rephrasing. When Apple Intelligence tried to restructure a long continuous dictation into a bulleted summary, TypeVox now detects it and falls back to the clean text instead of inserting the rephrased version. The system prompt has also been tightened to forbid that behavior up front.

Smarter system communication (from previous rounds): permissions checks are dramatically less frequent, keyboard monitoring only runs when your mode needs it, GPU and Neural Engine are released after a configurable idle delay.

Calmer permission checks

• The accessibility permission status no longer flickers between states when your Mac is busy. TypeVox now waits for steady confirmation before signaling a change, which keeps the entire UI smooth and predictable.

Stay current effortlessly

• TypeVox now checks for new versions silently a few seconds after launch. The new version badge in the Dashboard, the sidebar banner and the menu bar dot all light up the moment a release is available, with no need to click "Check for updates" first.

Smoother dictation HUD

• The audio spectrum in the HUD updates only when the level meaningfully changes, removing micro-redraws during dictation while keeping the visual just as fluid.

Under the hood

• Wider safety margin for the accessibility check so charged Macs no longer trigger false alerts.
• Tighter deduplication of internal state changes to avoid pointless redraws.
• Reduced polling cadence on the Settings panels.

Truly idle when idle

• TypeVox now stops every background animation the moment its window is no longer on screen, even when the Settings window is just hidden behind another app rather than fully closed. Your Mac stays as fluid on day five as on minute one.

Under the hood

• Switched the Dashboard background animations from a fire-and-forget loop to a manually paced loop driven by the actual on-screen visibility of the window.
• More resilient detection of when no window of TypeVox is being drawn anywhere on your screens.

Stays light for days on end

• TypeVox now stays just as light after days of uptime as on the first minute. The Dashboard animations truly stop when the Settings window is hidden, so nothing keeps your Mac's compositor busy in the background.

Under the hood

• More robust detection of when the Settings window is actually visible, with multiple safeguards layered together.
• Tightened control over a few looping symbol animations so they only run when their view is on screen.

Effortless on long sessions

• TypeVox now stays light through the whole day. The local AI model frees up memory automatically when you take a break, so your Mac feels just as snappy after a full day of dictation as in the first minute.

Smoother first launch after macOS updates

• After a major macOS update, the model can take a couple of minutes to recompile. TypeVox now shows a friendly progress message that rotates every few seconds, so you know things are moving.

Launch at Login that stays put

• TypeVox now keeps itself registered as a Login Item even after macOS updates that suspend background tasks. Your "Launch at Login" preference holds.

Snappier Settings

• Heavier views like the Dashboard now pause their animations when the window is hidden, freeing up resources for the rest of your Mac.

Under the hood

• Faster audio analysis using Apple Silicon's vector units.
• Smarter caching of language and mode preferences.
• Cleaner history pruning.

Lighter on your Mac

• TypeVox now runs noticeably lighter, especially across long sessions.
• The HUD, Dashboard and Settings views feel snappier and more responsive.

Under the hood

• Smarter resource handling so animations and background tasks rest when they should.
• General performance polish across the app.

Stay in the loop

• Help & feedback in the sidebar — share an idea, ask a question or report something without leaving TypeVox.
• What's new in the sidebar — read the release notes for the version you're running, plus the full history in an accordion.

Updates surfaced

• A subtle dot on the menu bar icon when an update is ready to install.
• A discreet banner appears in the sidebar so you never miss a new version.
• Toggle automatic update checks from Settings → Advanced.

Polish

• Smoother hover and focus across the sidebar.
• Cleaner typography in text fields.

Polish & maintenance

• Quality release behind the scenes.

Workflow

• Smoother AI workflow.

History

• Clearer copy menu in the dictation history.
• A subtle sparkles badge now shows when TypeVox AI shaped your text.

What's new

• AI prompt generator — describe your custom dictation mode in plain English ("a mode that turns my speech into formal email replies") and TypeVox crafts the system prompt for you.
• License activations manager — see and revoke the machines using your license at typevox.app/license/manage. Two activation slots per license.

Local AI by default

• TypeVox now polishes your transcripts with a local language model. Apple Intelligence on macOS 26+, an on-device model otherwise. Cleaner punctuation, better capitalization, no internet round-trip.

Reimagined dictation modes

• Custom modes powered by AI — write a system prompt and TypeVox shapes the output to match.
• Vocal Lab — test a custom mode with your voice before saving.

Productivity boost

• Auto-send after insertion (opt-in per app) — speak, release, message sent. Perfect for Slack and iMessage.
• Configurable shortcuts — pick your push-to-talk modifier key, add a separate toggle shortcut.
• Clipboard preserved — TypeVox restores your previous copy after pasting.
• Trailing space auto-added in flowing contexts.

Brand new Dashboard

• Streaks, time saved, top apps, weekly graph, hourly heatmap.
• Open Settings to see your progress at a glance.

More choice

• Multi-model Whisper — pick the model tier that fits your hardware: small (fast), medium, or large (most accurate).
• Visual app assignment — drag your apps into a grid and set a default mode per app.
• Configuration backup — export and import all your settings, modes, and vocabulary as a single file.

Now in 6 languages

• TypeVox speaks English, French, Spanish, German, Portuguese and Italian. The whole onboarding is fully translated. Switch language anytime from Settings → General.

Auto-update

• TypeVox stays up to date automatically. New versions install silently in the background.

Welcome to TypeVox

• Push-to-talk dictation for macOS. Press your key, speak, release: your text appears in any app.
• 100% on-device transcription with Whisper Large V3 Turbo. Zero cloud, zero account, zero API key.
• Works in any text field thanks to a smart 3-tier insertion: pasteboard, accessibility API, keystroke fallback.
• Smart history of your dictations — copy and search.
• Requires macOS 15+ on Apple Silicon (M1 or later).