Dictation that never leaves your machine.

Hold a key anywhere, speak, let go. The text lands where your cursor is — in any app, with no account, no upload and no subscription.

Apple silicon, macOS 14 or later. Free to use every day — see pricing.

Press. Speak. Release. Done.

There is no window to switch to and nothing to click. The shortcut works in whatever you are already typing in.

1. Hold the key

   Right ⌘ by default, or any combination you like. The microphone opens instantly, so nothing you say is clipped.

2. Say it

   A small bar shows the level and the clock, so you always know the microphone is open. It says nothing else during a meeting.

3. Let go

   Transcription runs on the Neural Engine. Typical end-to-end time is around a tenth of a second.

4. It lands

   Text goes to the cursor — and Tirona reads the field back to check it actually arrived, rather than assuming.

Dictation, and the parts around it.

It gets names right

Teach it a correction once and it always wins. Point it at a folder of your notes and it learns the names you actually write — matched by how they sound, so “Ferrara” becomes “Ferreira” without touching the rest of the sentence.

Modes

How you want words cleaned up depends on what you are writing, not on who you are. Name a mode, say what it should do differently, and let it switch itself on in the apps you choose.

Files, not just live speech

Drop in a voice memo, an interview or a meeting recording. Long files are transcribed a window at a time, and on macOS it can split the recording by speaker and label who said what.

It knows whether it worked

Every insertion is read back and checked. When an app cannot be typed into you get the text and a plain reason rather than silence — and Tirona never presses Return on a write it could not confirm.

Nothing to trust us with.

Most dictation apps stream your microphone to a server. Tirona does the recognition on your own hardware, which means there is no audio to leak, no account to breach and no bill that scales with how much you talk.

On-device speech

NVIDIA Parakeet on the Apple Neural Engine. Downloaded once, then it works on a plane.

No account, no check

No sign-up, no email, no telemetry, and no license check — key verification happens on your machine. After the one-time model download, Tirona makes no requests of its own at all.

The same tool on the desk and in your pocket.

macOS

• Global hold-to-talk shortcut, works in every app
• Verified insertion, with a plain reason when an app refuses
• Transcribe recordings you have, with speaker labels
• Menu-bar only if you want no window at all

Android

• A floating bubble above the keyboard, in any app
• The same model, on-device
• Hands-free continuous capture
• The same filler removal, name correction and modes

Settings do not sync between the two. There is no account to sync them through, which is the trade: your corrections live on the device you taught them to.

If it runs on your device, it’s free.

Not a trial and not a demo — the full-accuracy model, no daily limit, no account. Your Mac and your phone already have the hardware to do this; charging you for their electricity would be strange.

Get started.

Apple silicon Macs on macOS 14 or later. The speech model downloads on first launch and then never needs the network again.

The obvious ones.