# Transcription and offline use
URL: https://jarvis.ceo/docs/pantry/transcription-and-offline
Status: now

How Jarvis turns speech into text: on-device models that work fully offline, with an optional cloud model you bring your own key for. Accuracy, speed, and privacy.

Jarvis transcribes on your Mac by default. That keeps it fast, private, and able to work with no internet at all.

## On-device by default

A local speech model runs on your Mac. Smaller models are faster; larger models are more accurate. You can pick the trade-off in settings.

> [privacy] On-device transcription means your audio never leaves your Mac, and dictation keeps working on a plane or behind a firewall.

## Optional cloud, your key

If you want a cloud transcription service for extra accuracy, you can plug in your own key. The audio then goes from your Mac to that provider, not through us.

## Accuracy and speed

- About 240ms from key release to the first word.
- Around 3.1% word error rate on technical English with a mid-size local model.
- Larger models trade a little speed for more accuracy.

Related: Privacy and your data (https://jarvis.ceo/docs/pantry/privacy), Dictation: getting started (https://jarvis.ceo/docs/dictation/getting-started)