Local dictation processes your voice on your own Mac, so audio stays on the Mac. Cloud dictation sends audio to a server, so you rely on the vendor's policy and a connection. Some apps mix both. You can check which one you have by turning Wi-Fi off and dictating.
What local and cloud mean
Turning speech into text takes a speech model. In a local app that model is stored on your Mac and runs on its chip. In a cloud app your audio is sent over the internet to a server that runs the model and sends text back.
Many apps do both. An app might recognise speech on the Mac and then send the text to a cloud service to tidy it up, or let you pick local or cloud per task. So the useful question is not "is it local?" but "which steps leave my Mac?"
What changes for you
The practical differences are few, and they are the same for any app that works this way.
| Question | Local, on your Mac | Cloud, on a server |
|---|---|---|
| Where does the audio go? | It stays on the Mac. | It is sent to the vendor or a provider it uses. |
| Does it work offline? | Yes, once the models are on the Mac. | No. It needs a connection. |
| What do you have to trust? | The app and your own Mac. | The vendor's privacy policy, its providers and how long audio is kept. |
| What is the setup? | A one-time model download. | Usually nothing to download. |
| What limits languages and speed? | The models stored on your Mac and your chip. | What the vendor runs on its servers and your connection. |
How to find out what your app does
Vendors describe this in different places, and the wording is easy to skim past. These four checks take a few minutes.
- Read the privacy policy or security page and look for the words "processed", "servers" and "retained". Note whether it names providers.
- Look for a data or privacy setting that lets you opt out of your data being used to improve models.
- Turn off Wi-Fi, open a text field and dictate. If text still appears, recognition ran on the Mac. If it fails, the app likely needs a connection.
- If the policy and the offline test disagree, ask support in writing which steps run on the device.
What Pepys Dictate does
Pepys Dictate recognises speech and formats the text on your Mac. There is no account and no cloud transcription, and by default nothing you say or type is uploaded. By default it goes online only to download its models once, to activate a licence key and to check for updates, plus an opt-in usage event with no content if you turn it on. It works offline.
The models take about 1.5 GB of disk space, or 2.5 GB with the higher-accuracy model. An optional advanced setting can send a transcript to an AI provider you configure yourself. It is off by default.
When cloud is a fine choice
Local is not automatically better for everyone. A cloud app can cover languages, devices and team controls that a single Mac app cannot, and some cloud vendors publish compliance reports and let you opt out of training.
If you use a cloud app, read what it says about retention, pick the opt-out if there is one, and keep sensitive material out of it if its policy does not meet your bar.
Last updated October 5, 2026.