Dictation privacy

Local versus cloud dictation, without the privacy theater.

“Private,” “offline,” and “on-device” are often used as if they mean the same thing. They do not. This guide gives you a precise way to inspect speech processing, text cleanup, retention, accounts, and activation.

Source-reviewedReviewed July 26, 2026

1. Separate the stages

A voice workflow can include recording, speech recognition, punctuation, filler removal, rewriting, history, and synchronization. A product can run one stage locally and another in the cloud.

Ask where each stage runs instead of accepting one umbrella word such as “AI” or “private.”

Where is raw microphone audio processed?
Where is the first transcript produced?
Where are cleanup and rewriting performed?
Is any transcript history synchronized or retained?

2. Treat retention and processing as different questions

A zero-retention cloud mode still sends audio to a server for processing; it changes what happens after processing. A local app can still make narrow network connections for model downloads, licensing, updates, or optional services.

A credible product should disclose both the sensitive data path and the supporting network connections.

3. Match the architecture to the material

Cloud services can provide cross-device access, broad language support, team features, and continuously updated models. Local software can reduce the data path and keep working after setup when the network is unavailable.

The right choice can differ between a public brainstorming note and a confidential legal, medical, financial, or source-code discussion.

4. Read the exception list

Look for the phrases that qualify the headline: first-use model download, account authentication, privacy mode, sync setting, model-improvement opt-in, license verification, or unsupported language.

IraVoice discloses its one-time model download and its trial, license, and update check-ins with api.iravoice.com, which carry a salted hash of a hardware ID rather than your words. On IraVoice Local, audio and text are not part of any connection. The optional IraVoice Cloud plan, coming soon, is the hybrid exception, and it is named as one: when Cloud recognition or AI polish is on, audio and text will go to IraVoice Cloud, where Groq and Anthropic process them; IraVoice’s servers never store them, Groq does not retain them, Anthropic deletes API inputs and outputs within 30 days, and neither trains on them.

Decision guide

A five-question privacy check

Use the same checklist for every dictation product, including IraVoice.

If you need	Choose	Why
Where is raw audio transcribed?	Device or named cloud service	This determines the primary sensitive data path.
Where is text cleaned up?	Device, cloud, or optional mode	Formatting can create a second data transfer after transcription.
What is retained?	Audio, transcript, history, or nothing	Processing location alone does not answer retention.
What settings change the answer?	Language, privacy mode, sync, or account tier	The default and the most private configuration may differ.
What still needs the network?	Models, login, licensing, updates, or sync	“Works offline” rarely means the installer never connects.

Primary sources

Use the current documentation.

This guide was reviewed July 26, 2026. Product versions, settings, policies, and availability can change.

Apple: Dictate messages and documents on Mac ↗
Apple’s configuration-specific on-device processing guidance.
Claude Code: Voice dictation ↗
Anthropic’s disclosure that Claude Code voice streams audio for cloud transcription.
Wispr Flow privacy policy ↗
Cloud transcription, Privacy Mode, retention, and model-improvement controls.

Questions

What readers usually ask next.

Does zero retention mean audio stays on my device?

No. Zero retention can describe what a cloud service keeps after it processes audio. It does not necessarily mean the audio was processed locally.

Can an on-device app still use the internet?

Yes. An app may connect for initial model downloads, activation, updates, crash reporting, or optional sync. Check the exact disclosure.

Does IraVoice make any network connections?

Yes. The speech model downloads once on first run, and the app contacts api.iravoice.com for trial and license check-ins, with a salted hash of a hardware ID, and for update checks. On IraVoice Local, audio and text are never sent. IraVoice Cloud, coming soon, will send audio and text to IraVoice Cloud for Cloud recognition and AI polish, where IraVoice’s servers never store them, Groq does not retain them, Anthropic deletes API inputs and outputs within 30 days, and neither trains on them; both can be switched off.

Mac, Windows, and Linux

Try private, on-device dictation for yourself.

Hold one key, speak, and the words are typed where your cursor is. The 3-day free trial starts with your first dictation.

Download free trial $79 once