Technology & privacy
What on-device actually means
The phrase is used loosely enough to be meaningless. These pages explain the machinery, and how to check any app's claim for yourself.
"Private", "local", "on-device" and "secure" are marketing words until somebody tells you where the audio goes. These explainers are the technical half of that answer, written for people who are not speech engineers.
They also contain the checks: how to watch an app's network traffic, what to ask a vendor, and which questions have answers you can verify without taking anyone's word for it.
What Is On-Device Dictation? A Plain-English Guide
On-device dictation turns speech into text using a model that runs on your own computer. Here is what that means in practice, and how to check an app's claim.
Read itWhy Apple Silicon Is Good for Local Speech AI
Unified memory removes the copy that makes local AI slow elsewhere. What that means for dictation, and why several local apps refuse to support Intel Macs.
How to Choose Dictation Software for Confidential Work
A decision framework for when someone else has to approve it: which requirement you actually have, what architecture answers, and what only a certification answers.
Does Dictation Upload Your Audio? How to Check
Four ways to find out whether a dictation app sends your voice to a server, from a two-minute test anyone can run to watching the network traffic yourself.
Local Dictation History and Audio Retention Explained
On-device does not mean nothing is stored. What a dictation history contains, where it lives, why it is useful, and how to decide how long to keep it.
What Is Local Speech-to-Text? A Plain-English Guide
How a speech model turns sound into words, why it can now run on a laptop, and what "local" means precisely enough to check. No machine learning background needed.
Why No-Account Dictation Matters
Most dictation apps want an email address before your first word. Why that is usually unavoidable for them, why it is not for a local app, and what it changes.
Offline vs Cloud Dictation: The Real Trade-Offs
Cloud dictation is more accurate on hard audio and needs a connection. Offline is predictable, private and bounded by your hardware. Which constraint is yours?
How ONNX Helps Run Speech Recognition Locally on Mac
ONNX is a portable format for trained models. Why that matters for local dictation, what a runtime does, and why you should not have to care — but might want to.
Parakeet Speech Recognition on Mac: A Plain-English Guide
What Parakeet is, why it suits live dictation on Apple Silicon, how it differs from Whisper, and what it is not good at. No machine learning background required.
Private AI on Mac: What Runs Locally and What Does Not
A clear-eyed inventory of what a 2026 Mac can genuinely do without a network — speech, small language models, images — and where the marketing outruns the machine.
How System-Wide Dictation Works on macOS
Why a dictation app needs accessibility permission, how text reaches every application without plugins, and why the clipboard sometimes gets involved.
What Is Voice Activity Detection in Dictation Apps?
VAD decides where speech starts and stops before the speech model sees the audio. It is also why the first and last words of a dictation go missing.
Voice Data Privacy: Questions to Ask Any Dictation App
Six questions with answers you can verify, in the order that matters. Use them on us as well — most of them are answerable without the vendor's cooperation.
Whisper vs Parakeet for Local Dictation
Two families of open speech models with different design goals. What each is good at, why latency matters more than accuracy for dictation, and how to choose.
15 pages in this section. All of them work without JavaScript.