PRIVACY · UPDATED AUGUST 2026
Wispr Flow privacy: where your voice actually goes.
Before you let a dictation app listen all day, two questions matter: where the audio is processed, what context features are enabled, and how retention is controlled. Here is what the current documentation says, with the controls kept distinct.
What happens to your audio
Per Wispr Flow's security and compliance FAQ (reviewed July 2026), dictation audio is sent to its cloud for transcription. An account and network connection are therefore part of the product architecture.
Its privacy-control documentation separates two settings that are easy to conflate: Privacy Mode controls whether data can be used for model training, while Private Cloud Sync controls whether audio and transcripts are retained on its servers after processing.
The screen-context story
Wispr Flow documents two context features with different defaults. Accessibility-based text context is enabled by default. Screen OCR, which uses screenshots to derive context, is disabled by default and can be enabled separately. Users can turn either context setting off in the app.
Those settings are separate from Privacy Mode and Private Cloud Sync. Evaluating the product accurately means checking all four controls rather than treating one switch as a complete description of its privacy posture.
The architectural difference: cloud transcription requires audio to leave the device. Context features can add text or screenshot-derived data when enabled. Wispr Flow provides controls for training, retention, and context; LockedIn Flow avoids the cloud transcription path entirely.
What "local" means — and how to verify it yourself
Every dictation app now claims to care about privacy. Don't take anyone's word for it — including ours. You can watch what any app actually does on the network in under two minutes:
- Terminal: run
nettop(built into macOS) and dictate for 30 seconds. Sustained upstream traffic while you talk means your audio is leaving. - Little Snitch: the Objective Development firewall shows every connection in real time, per app and per host. Dictate and watch.
- Airplane mode: turn Wi-Fi off and dictate. If the app errors out, it was never local.
LockedIn Flow runs Parakeet TDT v3 on the Apple Neural Engine. Core cleanup uses local rules on macOS 15+. Optional Apple FoundationModels cleanup requires macOS 26, Apple Intelligence, and an available on-device model. After the one-time ~600 MB speech-model download, dictation itself does not require network traffic. No account is required, and history is encrypted locally.
Side by side
| Area | LockedIn Flow | Wispr Flow |
|---|---|---|
| Audio processing | On-device | Cloud |
| Transcript history | Encrypted locally when enabled | Cloud retention controlled by Private Cloud Sync |
| Accessibility text context | Not used for language context | Enabled by default; can be disabled |
| Screen OCR | Not used | Disabled by default; can be enabled |
| Account | Not required | Required |
| Training control | No vendor training pipeline | Privacy Mode |
| Offline dictation | Yes, after model download | No |
The honest trade-off
Wispr Flow's cloud buys real things: Windows, iOS, and Android support, 100+ languages, and settings that sync across devices. LockedIn Flow is macOS-only (Apple Silicon) and focused on European languages. If cross-platform matters more than privacy, Wispr Flow is the honest pick. If you live on a Mac and want dictation where "private" is something you can check rather than something you're asked to believe — that's why we exist.
FREE 14-DAY TRIAL · NO CARD · NO ACCOUNT
Try dictation you can verify.
Everything unlocked for 14 days. Run nettop while you talk — we welcome the scrutiny.
Download for macOSWispr Flow claims reviewed against its official privacy and security documentation on July 30, 2026. Tell us if a documented setting changes.