Privacy Policy

Last updated: 12 November 2026

Voizey AI ("Voizey", "the app") is published by Nexa Limited, Hong Kong. This policy explains what the app does with your information.

Summary

Voizey does not collect, store on our servers, share or sell any personal data. We run no servers for the app, and it contains no analytics, advertising or tracking.

Voice and text

Speech recognition and text clean-up use Apple's on-device frameworks (Speech and Apple Intelligence / Foundation Models). Your audio and the resulting text are processed on your device and are not sent to Nexa Limited or to any third party by Voizey. Translation uses Apple's on-device Translation framework with language packs downloaded from Apple.

No third-party AI service. Voizey does not send your audio, text or any other personal data to any third-party AI service (for example OpenAI, Google, Anthropic or similar), and contains no such service. All AI features run on your device using Apple's frameworks and the bundled or downloaded on-device model described below.

Keyboard and Full Access

The Voizey keyboard asks for "Allow Full Access" only so that the keyboard can exchange dictated text with the Voizey app on the same device (through the shared device Keychain). The keyboard does not transmit anything over the network, and does not log what you type.

Data kept on your device

Your selected languages, personal dictionary and words the app has learned from your corrections are stored on your device only. You can delete them any time from the personal dictionary screen inside the app (Auto-learned and Manual tabs, each with a delete button); the app also keeps a diagnostic log file on the device that may contain phrases you dictated; it is never transmitted. Settings and learned data are stored in the iOS Keychain, which iOS may keep after the app is deleted and restore on reinstall.

Purchases

Subscriptions and the one-time purchase are processed by Apple through the App Store. Nexa Limited does not receive your payment details.

Permissions

Enhanced recognition download

On first launch the app downloads an additional speech recognition model (about 0.9 GB) over Wi-Fi from our file server. The download is a plain file request and sends no personal data or audio; recognition then runs entirely on your device. The file is hosted on Cloudflare R2, so Cloudflare may see your IP address as with any download. The model is removed when you delete the app.

Open-source notices

The enhanced recognition uses the Qwen3-ASR model (Qwen Team, Apache License 2.0) and the sherpa-onnx (Apache License 2.0) and ONNX Runtime (MIT License) libraries. Licence texts: apache.org/licenses/LICENSE-2.0, onnxruntime LICENSE.

Children

The app is not directed at children and collects no data from anyone.

Changes

If this policy changes we will update this page and the date above.

Contact

Nexa Limited, Hong Kong · info@nexalimited.io