Privacy Policy
Windy Word Privacy Policy · Version 11 · Last updated 1 October 2026
Windstorm Labs LLC ("we", "us") provides Windy Word. This policy explains what we receive, what we do not, and what you can do about it.
1. The short version
Windy Word transcribes speech on your device, using engine files stored on your computer. Dictating does not require an internet connection, and when you dictate locally your audio and the resulting text stay on your machine.
We do receive measurements about how the app is used — how long, how often, how long each dictation runs, which engines and languages you choose, what kind of machine you run it on, and approximately where you are (country, region, and city or metro area, from your network address). Measurements are about usage, not about what you said.
Optional cloud features work differently, and section 4 says how.
We have tried to write this without absolutes. Instead of promising that something can never happen, we describe how the system is built and what we do if something goes wrong (section 9).
2. What the app sends us
For the local app, the app reports:
| Category | Examples |
|---|---|
| Account | Your Windy account identifier; whether your email or phone is confirmed |
| Usage volume | Hours of use per day and per week; number of dictation sessions; total words produced (a count, not the words) |
| Activity state | Whether the app is actively in use or idle |
| Recording characteristics | Length of each dictation (a duration, not the audio); average and distribution of lengths |
| Configuration | Which engines are installed and selected, and the share of your dictation time each one accounts for; preferred and detected languages; translation language pairs used; app version; operating system and platform |
| Settings choices | Which options you have picked in the app — for example recording mode, paste behaviour, theme, widget, alert interval, audio and video quality settings, and which keyboard shortcuts are assigned (the assignment, never what you type with them) |
| Library size | How much space your local recordings and transcripts occupy on your own disk (a byte count and item count — the library itself never leaves your machine) |
| Clone Capture readiness | If you use Clone Capture: how many minutes of usable audio you have captured and coverage percentages — measurements of quantity and completeness, not the audio itself |
| Device characteristics | Processor family and core count; memory; graphics hardware and video memory; architecture; device model identifier (for example "MacBook Air M2"); number of displays; whether running on battery or mains; free disk space at install |
| Approximate location | Country, region, and city or metro area, derived from the network address your app connects from, plus your timezone and the local time of day you use the app |
| Machine condition during a dictation | How busy the processor was, as a percentage, and — on macOS, which is the only system that reports it — whether the machine was running hot enough to slow itself down. Measured around the moment of a transcription so we can tell a slow engine apart from a busy or overheating computer |
| Reliability | Crash and error reports, performance timings, install and update outcomes, permission grants and denials |
| Notices | Whether an in-app notice was shown, dismissed, or acted on |
What is not in that list: your audio, your dictated text, your transcripts, your translations, your clipboard, the names of your files, the identity of the applications you paste into, or the contents of your screen.
This is enforced rather than merely intended: our measurement endpoint rejects events whose fields match content-like names — text, body, message, prompt, transcript, subject, completion and similar. An event carrying your words is refused by the server rather than stored.
You can turn this off. Settings ▸ Help improve Windy Word is on unless you turn it off, and the setup wizard explains it before you start. Turning this off stops collection. Delete my usage data, next to it, erases what was collected: it asks us to erase the usage measurements this installation has sent, and it works whether the switch is on or off. We keep a record that you asked, and of your on/off choice, but not the measurements themselves.
3. Approximate location — what we do and do not derive
We derive an approximate location — country, region, and city or metro area — from the network address your app connects from, together with your device's timezone. "Metro area" means a broad market region (for example, the Albany metro), not a neighbourhood and not a point on a map.
We do not collect GPS or precise location, we do not collect your street address, and we do not collect the identity of your wireless network. Location derived from a network address is approximate by nature: it is often wrong at city level, particularly on mobile networks, and we treat it accordingly.
We use it to understand which places the product is used in, which languages matter where, when during the day people dictate, and to make regional offers and announcements relevant to where you are.
3a. Device characteristics, and why we look at them
We record what kind of machine you run the app on — processor, memory, graphics hardware, architecture and device model — and which engines you actually use on it. The engines have very different speed and accuracy characteristics, and the only way we can recommend the right one for a given machine is to measure which ones perform well on which hardware in real use.
We identify the machine by model (for example "MacBook Air M2"), not by serial number, and each installation is associated with a randomly generated identifier rather than anything derived from your hardware. That identifier can be reset by reinstalling and is deleted when you delete your account.
We also record how busy and how hot the machine was at the moment of a dictation, because without it the speed figures are unreadable. The same engine on the same laptop can transcribe twice as fast as you speak or three times slower, and the difference is usually that the computer was busy doing something else or had heated up enough to throttle itself. Measured on real machines, one engine fell from 4.9x to 1.5x for exactly that reason. Without knowing the machine's condition we would have blamed the engine and "fixed" the wrong thing.
This is a percentage and a coarse state — never a temperature reading tied to you, never a list of what else you were running, and never the names of other applications.
4. Optional cloud features
If you choose to turn on a cloud feature — cloud compute, Windy storage, or cross-device sync — then the content you send to that feature does leave your device, because that is what the feature does. When you use one:
- audio and text you submit are transmitted to our servers, and processed there or by a service provider acting for us (section 8 names them);
- content is encrypted in transit. Not all of our storage is encrypted at rest yet;
- content is retained per section 7 and deleted on request;
- you can stop using cloud features at any time and continue locally.
Cloud features are off unless you turn them on. We say this explicitly because "your words stay on your device" is only true of local use, and we would rather be accurate than tidy.
4a. The browser extension
The Windy Word browser extension is a separate, optional install from the Chrome Web Store or Microsoft Edge Add-ons. The app works without it, and removing it from your browser removes everything in this section.
It runs on every website you visit, and we would rather explain that than bury it. The extension's job is to put a dictate button next to the message box you are already typing in. We cannot know in advance which sites those are — for you it may be a support desk, a hospital portal, or your own company's tools — so the extension asks to run on all https pages rather than keeping a list of approved ones. On each page it looks for the site's own microphone or composer, puts our button beside it, and otherwise does nothing.
The extension has no server of its own, and sends nothing to us. It talks to one place: the Windy Word application already installed on the same computer, over the browser's native-messaging channel — a pipe between two programs on your machine that does not cross the network. The messages it sends are the commands start, stop, state and level. They do not include the address of the page, its title, or its contents.
Specifically, the extension does not collect your browsing history, does not record which sites you visit, does not read pages for us, and does not run any code it downloads.
When you dictate through the button, the recording and transcription are done by the desktop app, so sections 1, 2 and 3 apply to them unchanged. The resulting text goes to the message box you clicked in — either typed in by the app or written directly into the editor by the extension. Your archive is the same local archive the app already keeps.
The permissions the browser shows you at install are used as follows: nativeMessaging to reach the desktop app; scripting and access to https pages to place the button; activeTab to act on the tab whose button you clicked; and storage for your own extension settings, held in your browser.
5. Account data
We hold your name, email address, optional telephone number, confirmation status, password hash (we do not store your password itself), account creation and sign-in metadata, and entitlements.
Your email address is your recovery channel. Confirming by text message does not replace it: password reset and account recovery are available only by email.
6. Communications, and how consent works
Service messages — account, security, billing and material service changes. These are part of operating your account and are not marketing.
In-app notices — from time to time, dismissible.
Marketing email and text messages — sent only if you opt in. At signup this is a separate, unticked checkbox, not something bundled into accepting the Terms. You can withdraw at any time from account settings, by the unsubscribe link, or by replying STOP to a text. Withdrawing marketing consent does not affect your free local licence or your ability to use the app.
We keep this separate on purpose. Under GDPR consent must be freely given, specific, informed and unambiguous, and consent buried inside accepted terms does not qualify. US law requires separate express written consent for marketing texts. A single "I agree" covering both verification and marketing would not be valid in either place.
7. How long we keep things
| Data | Retention |
|---|---|
| Account record | While your account exists; deleted on request, subject to legal retention duties |
| Usage measurements | Aggregated; identifiable form kept no longer than 24 months |
| Crash and error reports | 12 months |
| Cloud content (if you use cloud features) | Until you delete it, or account closure. Copies can remain in our database backups after that |
| Translation history | Until you delete it, or account closure |
| Translation cache (the text of a translation, without your account details, kept to answer repeat requests faster) | Up to 1 hour, in memory only; never written to disk |
| Cloud dictation service records (which engine, how long it took, whether it worked; never the audio or the text) | 13 months |
| Voice Project donations (only if you choose to donate) | Up to 24 months, or until you ask us to delete them |
| Marketing preferences and consent records | Kept as evidence of consent while relevant, and after withdrawal as evidence that you withdrew |
8. Who else is involved
We use service providers to operate the Service, including hosting and content delivery, email delivery, text-message delivery, payment processing for paid features, and error and usage measurement. They act on our instructions and are bound by contract.
Where a cloud feature uses a third-party model provider, the provider handling your content is identified in that feature before you use it. Today that is:
- Groq, Inc. (speech recognition): receives the audio of cloud dictation on the standard tier, and of spoken translation, to turn it into text.
- Your AI helper (its replies in Windy Chat and the other places your helper answers): what you say to your helper is sent, through Windy's own routing service, to AI model providers to get its answers. Today these are Windy's own computers; Groq, Inc.; Cloudflare, Inc. (Workers AI); and OpenRouter, Inc., which passes a request on to other model hosts and which we ask not to let those hosts train on it. For an account owner's own agents, Anthropic, PBC; and, only for models chosen on paid or connected accounts, Cerebras Systems, Inc. and SambaNova Systems, Inc. Providers whose free terms may allow training on messages (Google's free Gemini tier and NVIDIA's trial API) are not used for your messages unless you choose otherwise. Each provider's own terms govern what it does with a request it receives, including how long it keeps it. The notice shown when you hatch a helper says the same in short, and we keep a record that you agreed to it.
We do not sell your personal data. We do not share it with data brokers.
9. If data reaches us that should not have
Systems fail and people make mistakes. Rather than promise this cannot occur, here is our commitment if it does:
- We will stop the flow as soon as we identify it.
- We will delete the affected content, and confirm deletion with any processor that received it.
- We will notify regulators and affected users where the law requires, within the time limits it sets.
- We will publish what happened and what we changed.
This is a description of our process, not an admission that it has happened.
10. Your rights
Depending on where you live, you may have the right to access, correct, delete, port, restrict or object to processing of your personal data, to withdraw consent, and to complain to a supervisory authority. Residents of the EU/EEA and the UK have these under GDPR; residents of India have comparable rights under the DPDP Act; residents of California and several other US states have rights under state privacy laws.
We do not require you to give a reason, and exercising these rights will not degrade your use of the app.
- Access and export: available in the app, or by request
- Deletion: available in the app, or by request
- Marketing withdrawal: account settings, unsubscribe link, or reply STOP
Contact [email protected]. We respond within the period applicable law requires, and within 30 days where no period is set.
For EU/EEA users, our representative is [EU REPRESENTATIVE — REQUIRED IF NO EU ESTABLISHMENT]. For India, our Data Protection Officer is [DPO — REQUIRED UNDER DPDP FOR SIGNIFICANT DATA FIDUCIARIES].
11. International transfers
We operate globally, so data may be processed in countries other than yours, including the United States. Where we transfer personal data out of the EU/EEA or UK we rely on appropriate safeguards such as Standard Contractual Clauses.
12. Children
The Service is not directed at children under 13. We do not knowingly collect their personal data. If you believe a child has created an account, contact [email protected] and we will delete it.
13. Changes
We will post changes here and update the date above. For material changes we will give notice in the app, by email, or both before they take effect, and we will ask for fresh consent where the law requires it.
14. Contact
Windstorm Labs LLC, Eagle Mountain, Utah, USA · [email protected]