WordWarden Privacy Policy
Last updated: 2026-10-24
WordWarden is a voice-dictation app for Windows that runs entirely on your own computer. We built it so that we never need to see, collect or store anything you say or type.
What WordWarden does with your data
| Data | What happens to it |
|---|---|
| Microphone audio | Captured only while dictation is active. Held in memory just long enough to transcribe, then discarded. Never saved to disk and never sent anywhere. |
| System audio (calls, videos, anything you hear) | Captured only while dictation is active and you've turned it on: by holding or toggling the system audio key or the all audio key, or by choosing it as the audio source to start with in Settings. Handled exactly like microphone audio. |
| Transcribed text | Pasted into the app where your cursor is. WordWarden marks it so it's excluded from Windows clipboard history and cloud clipboard sync, then restores your previous clipboard. It isn't logged or kept. |
| Settings | Saved locally in wordwarden_config.json next to the app (hotkeys, device names, language, colors, sound choices, overlay position). They never leave your computer. |
| Diagnostic log | wordwarden_debug.log records technical events (for example "recording started" or error messages) to help troubleshoot. It never contains your speech or transcribed text. It stays on your computer unless you choose to send it to us. |
What we collect
Nothing, unless you send us feedback. WordWarden has no accounts, analytics, telemetry, advertising or automatic crash reporting. We don't receive your audio, transcripts, settings or usage data.
The one exception is a bug report or feature request you choose to send from the Send feedback window (Settings → About, or the tray menu). Nothing is sent until you press Send. A report contains:
- what you typed into the form;
- any screenshots you added, exactly as they look after your markup (anything you black out is removed from the image before it leaves your computer);
- your email address, only if you enter one;
- for bug reports, the end of the diagnostic log, only if you leave that box ticked;
- basic technical details: WordWarden version, Windows version, speech model, whether it runs on your graphics card or processor, language and audio source settings;
- a random install ID. WordWarden makes it up the first time it's needed and saves it in your settings file. It isn't based on your hardware or identity; it only lets us tell that several reports came from the same installation.
We use reports only to fix bugs, consider feature requests and reply to you. They're delivered to us by email through Cloudflare (which runs the small service that receives them) and Resend (which delivers the email), and kept in our mailbox. Ask us and we'll delete your reports.
Internet connections
WordWarden connects to the internet in these cases only:
- Speech model download. The first time you use a model size or language, WordWarden downloads the speech-recognition model from Hugging Face (huggingface.co), a third-party model host. No audio or text is sent. Hugging Face receives a normal web request (such as your IP address) under its own privacy policy. WordWarden turns off Hugging Face's optional usage telemetry. After the download, transcription works fully offline.
- Translation pack (optional). If you choose to have text typed in a language other than the one you speak (other than English), WordWarden offers to download an offline translation model (M2M100) from Hugging Face, once. As with the speech model, no audio or text is sent. Translation then runs entirely on your computer.
- Update checks. WordWarden checks GitHub (github.com) for a newer version shortly after it starts, and when you click Check for updates. This sends a normal web request (such as your IP address) and nothing else: no audio, text, settings or identifiers. If you accept an update, the installer is downloaded from GitHub. You can turn automatic checks off under About in Settings.
- Feedback you send. Only when you press Send in the Send feedback window. See "What we collect" above for exactly what's included.
Accuracy
WordWarden uses OpenAI's Whisper speech-recognition model, run locally via faster-whisper. Like all AI transcription, it can mishear words or occasionally produce words that weren't spoken ("hallucinations"), especially with background noise, music or long silence. Review dictated text before relying on it. WordWarden is provided "as is", without any guarantee that transcriptions are accurate. See the Terms of Use (in the app under About → Legal) for full details.
Recording other people
The system audio and all audio features can transcribe other people's voices, for example on a call. Recording or transcribing a conversation may require everyone's consent where you live. You are responsible for following the laws that apply to you.
Children
WordWarden isn't directed at children and doesn't knowingly process children's data. It doesn't collect anyone's data, apart from feedback a user chooses to send.
Changes
If this policy changes, the updated version will ship with the app and show a new "Last updated" date.
Where we are
WordWarden is developed in Alberta, Canada. If we ever handle personal information, for example when you email us for support, we handle it in line with the privacy laws that apply in Canada and use it only to respond to you.
Contact
Questions: [support email to be added]