Hold to talk, release to type
Hold a key in any app, speak, release — your words appear at the cursor in under half a second.
Speak naturally, your computer writes it down.
100% local · works offline
hold key to dictate
HOLDF12
Your voice stays on your machine
We don't even have a server. Which means even if we wanted to, we couldn't take your voice.
Recording stays on-device
The mic stream goes straight into the local engine, never the network.
The AI model is local too
Downloaded once. Works offline forever after.
Zero telemetry by default
Crash reports? You opt in inside Preferences.
No cloud mode, no stored audio
Transcripts land in the folder you pick. That's the whole pipeline.
Verifiable build: signed and notarized binaries. Bring any firewall tool and watch the traffic yourself — you won't see a request.
Designed for people who actually use a keyboard
Not a kitchen sink. A few things, done well.
AI cleans it up
A local AI strips fillers, breaks lines, light translation. Not a single character leaves your machine.
Tuned for your machine
Native speed on Mac, fast on Windows. No need for top-tier hardware.
Learns your words
Project names, people, jargon — say it a few times and NookType remembers how to spell it.
FAQ
Things you'll probably ask
Yes. Apart from the initial model download and checking for updates, day-to-day use makes no network calls. You can verify it with a firewall tool — you won't see a single request.
It's planned for later — no firm date yet, and not a promise.
v1.1 will add Bring-Your-Own-Model: a local model file you've downloaded, or your own self-hosted endpoint.