NARRATO
Speech to textOffline

Speak. It lands where
you’re already typing.

Hold a shortcut, talk, and your words drop straight into the focused field, in any app. Transcribed by Parakeet on your Mac, nothing sent to a server.

Try Narrato freeHow privacy works →

Hold, speak, release

No window to copy out of. The text arrives where your cursor already is.

1
Hold the shortcut
⌥Space
From any text field, in any app. A compact recorder shows it’s listening.
2
Say what you mean
Speak naturally. Parakeet transcribes on-device as you go, never streaming your audio.
3
Release, it’s typed
AT YOUR CURSOR
Cleaned-up text lands where you were typing, ready to send.

Both gestures are always live: hold the shortcut to record and release to insert, or tap once to start a hands-free session and tap again to stop. No mode to switch.

Dial in how much cleanup you want

Four levels, all on-device, from verbatim to crisp. Narrato only ever removes fillers and stumbles. It never rewrites the words you actually said, and when audio is unclear it flags the word rather than quietly guessing.

CLEANUP · OFF

UmIIthinkweshouldshipit ,andtestit

Cleansnothing, verbatim

Verbatim. Every word exactly as you said it.

Lands cleanly in any app

It inserts straight into the focused field and keeps a copy on the clipboard as a safety net. Flip it and see what changes.

Transcript ready
NARRATO CHECKS
A text field focused?
YES
Inserted into the selected field
…and kept on the clipboard too
NO
Copied to your clipboard
paste it yourself, nothing lost

Clipboard Context: copy while you talk

Describing something on screen? Copy it mid-sentence. Narrato pins every copy, text or screenshot, to the exact word you were saying and drops it into the transcript right there. Built for feeding an AI chat real context without breaking your sentence.

1
Copy mid-sentence
⌘C
While dictating, grab anything from any app: a paragraph, an error message, a screenshot. Keep talking.
2
Pinned to the word
Each copy is timestamped against your speech and anchored to the exact word you were saying.
3
Inserted in place
IN YOUR TRANSCRIPT
When the transcript lands, every capture drops in right where you referenced it, as raw context.

Like the rest of Narrato, this happens on your Mac; captures go into the text you insert and nowhere else. Watch it happen on the homepage →

ASR ENGINE · MODEL CARDnarrato://engine/asr
MODEL

Parakeet TDT 0.6B v3

NVIDIA’s ASR model, transcribing on your Mac through MLX.

A FastConformer encoder with a TDT decoder, 4-bit quantized to ~607 MB, fast enough to feel instant, fully offline, with punctuation and capitalization built in, across 25 languages. See how privacy works →

parakeet-tdt-0.6b-v3 ↗NVIDIA/NeMo ↗
PARAMETERS0.6B
ARCHITECTUREFastConformer · TDT
QUANTIZATION4-bit · ~607 MB
RUNTIMEMLX · Apple GPU
LICENSECC-BY-4.0
EXECUTION100% on-device
WAIT AFTER YOU STOP TALKINGflat wait, any length
10-sec note~0.45 s
1 min & up~1.2 s
A rolling tail window keeps the wait flat, a second or so whether you dictate for one minute or thirty. Single session capped at 30 minutes. Measured July 20, 2026 on an M3 MacBook Pro (16 GB, macOS 15), models warm.

Stop typing what you could say.

Dictate into any app on your Mac, offline, with your voice never leaving the device.

Try Narrato free