A dictation shortcut has to answer two questions: is the microphone recording, and what is happening to my text? Vext 1.5.1 makes both answers visible in one place. The Press-to-Paste HUD keeps the same pill on screen from microphone startup through transcription and optional AI polishing.

This release builds on the live transcript, retained recordings, and agent tools introduced in Vext 1.4.0. The focus is the time between pressing your shortcut and getting usable text into your app.

Red means the microphone is live

Hold your dictation shortcut. The pill first shows that the microphone is starting, then turns red only when it is actually capturing audio. That is the signal to begin speaking. Live text continues to appear while you talk.

Previously, lowering system audio could delay capture and cut off the beginning of a sentence. In 1.5.1, audio ducking no longer delays recording — the red state reflects actual microphone readiness, not just a key press. One caveat: Bluetooth microphones can still take a moment to wake, so wait for the red signal before speaking, or use a built-in or wired mic.

Follow the same pill to the finish

Release the shortcut and the pill stays. It shows “Transcribing…”, then, when the selected style needs AI, “Polishing” with that style — say, Email. An elapsed clock makes the wait visible, and the menu-bar icon shows a processing state too.

No spinner chases your pointer through this flow. The live text stays in the pill, so your eyes rest in one place while Vext finishes the take.

Press Esc to skip slow polishing or model loading and paste the transcript as spoken. That skips the AI pass — it can’t skip transcription itself, so speech still in progress won’t paste early. If a take fails, Vext keeps its audio and offers one-click Retry right in the pill: try that recording again instead of repeating everything.

Use the style that fits the app

The dictation styles are As spoken, Clean up, Email, Message, Bullet points, Formal, and Custom. Per-app styles let you choose, for example, Message for a chat app and As spoken for a terminal. These are the existing dictation styles, now part of a clearer processing flow.

As spoken leaves the transcript alone, with no AI reshaping. Other styles request different edits, so review the output when exact wording matters. And when you’d rather not wait for polishing, Esc always gets you the original transcript.

Reliability and everyday controls

The smaller changes remove friction from everyday use:

  • Right-Cmd hands-free and Fn meeting toggles no longer activate on combinations such as Right-Cmd+C or Fn+Delete. Menu hints reflect your own bindings.
  • Paying customers keep their paid access while offline.
  • The local AI model downloads only when a style or translation actually needs it.
  • Native pickers, localized counts and durations, and sidebar-based settings navigation make controls more consistent.
  • Export dictations, notes, and meetings directly from the app — TXT and Markdown, with SRT and VTT for meeting transcripts. Onboarding links to System Settings after a permission is denied, and the Buy button shows the price.

Process locally by default

Dictation runs locally by default. Parakeet TDT v3 runs through FluidAudio on Apple Silicon, with Apple Speech as another option, and polishing uses a local MLX model. Prefer the cloud? Add your own API key for OpenAI-compatible models — they’re off by default.

Choose from 29 translation targets, or keep the original language; spoken-language support depends on your speech engine. With 1.5.1, the journey from microphone to paste is easier to follow — you always know where you are.