Privacy Policy

1 Who We Are

Scrivox is developed and published by LogicGateGames LLC, a Michigan-registered software company. Contact us at logicgategames@gmail.com.

2 What Data We Collect

We do not operate servers that receive your recordings or transcripts. All AI models run entirely on your device.

Firebase Crashlytics — If the app crashes, Crashlytics sends a crash report so we can fix the problem. A report contains the error's technical details (stack trace), device model, Android version, app version, and a random Crashlytics installation ID, plus the type of build and the name of the app store the app was installed from. It never includes your content, such as your recordings or transcripts, and it contains no name, email or advertising ID. Reports are processed by Google and kept for up to 90 days. We do not sell or share them.

3 On-Device AI Models

Scrivox uses several on-device AI models for speech recognition, summarization, speaker identification, and translation. Every one of them runs entirely on your device:

• Base speech engine — OpenAI Whisper, an open-source model licensed under the MIT License. Downloaded automatically on first launch (~185 MB). Supports 71 languages including English, Spanish, French, and more.
• Indian language packs — Fine-tuned Whisper models for Telugu, Hindi, Tamil, Kannada, and Gujarati, developed by vasista22 (IIT Madras) and licensed under the Apache License 2.0. Optional and downloaded only when you choose to, in Settings or from a recording in one of those languages (~694 MB each).
• AI summary model — Qwen2.5-1.5B-Instruct (Apache License 2.0), used to generate summaries and action items. Optional and downloaded only when you choose to (~1.7 GB).
• Speaker identification — the pyannote.audio speaker-segmentation model (MIT License), used to label who spoke when. A Premium feature: the model is downloaded automatically the first time a Premium transcription needs it (~6 MB).
• Translation & on-device generative AI — powered by Google ML Kit, including on-device translation models (about 30 MB per language, downloaded from Google the first time you translate into or out of that language, or when a summary of a non-English transcript needs one) and, on supported devices, on-device summarization via Google's Gemini Nano through Android AICore. These process your text entirely on the device.

All model files are downloaded over HTTPS — the base engine and speaker model from HuggingFace (huggingface.co), and the Indian language packs and AI summary model from GitHub Releases (github.com); ML Kit translation models are downloaded from Google’s servers, and Gemini Nano is provided through Android AICore on your device. Once installed, all models run on your device, and your recordings never leave your phone. No audio, recordings, or transcript data is ever sent anywhere during transcription, summarization, speaker identification, or translation.

During a model download, the respective server (HuggingFace, GitHub or Google) will receive your IP address as part of a standard HTTPS request. No audio, recordings, or personal data is transmitted. These downloads are governed by HuggingFace's Privacy Policy and GitHub's Privacy Policy. On-device processing by Google ML Kit is subject to the Google Privacy Policy.

4 Microphone Permission

Scrivox requests microphone access solely to record audio when you tap the record button. Audio is:

• Captured only while you are actively recording
• Processed on-device by the Whisper AI model
• Stored as a local file on your device
• Never transmitted over the internet by Scrivox — it leaves the device only if you share it, export it, or save a backup yourself

You can revoke permission anytime in Settings → Apps → Scrivox → Permissions.

5 Advertising (Free Tier)

The free version displays ads served by Google AdMob: banner ads, an occasional full-screen ad after a transcription, and optional rewarded ads you can choose to watch to unlock a translation. Google may collect data to serve personalised ads, including advertising ID, approximate location, and device information. See Google Privacy Policy. Upgrading to Premium removes all ads.

If you are in the European Economic Area, the UK, or Switzerland, Scrivox shows Google’s consent form the first time you open it, using Google’s User Messaging Platform. Your answer determines whether ads may be requested at all and whether they are personalised. You can reopen that form and change your answer at any time from Settings, under “Ad Preferences”; withdrawing consent takes effect immediately, without restarting the app.

6 In-App Purchases

Premium purchases are processed entirely by Google Play Billing. We receive only a transaction confirmation — we never see or store your payment details.

7 Data Retention & Deletion

All recordings and transcripts are stored on your device and fully under your control:

• Delete individual recordings inside the app
• Deleted recordings move to Recently Deleted and are removed permanently after 7 days, or straight away with Settings → Clear Trash
• Uninstalling the app removes all app data. Files you exported or backups you saved outside the app are not removed; delete them separately
• Scrivox does not take part in Android’s automatic cloud backup. Settings → Back up library creates an encrypted backup file, protected by a passphrase you choose, and saves it wherever you pick
• Crash reports, if any were sent, are deleted automatically after 90 days

8 Children's Privacy

Scrivox is intended for adults (18 and over) and is not directed at children. We do not knowingly collect personal information from children.

9 Changes to This Policy

We may update this policy from time to time. The "Last updated" date reflects the most recent revision. Continued use constitutes acceptance.

10 Contact Us

LogicGateGames LLC
logicgategames@gmail.com