Privacy
Privacy policy
For the Saywind app and saywind.com. Last updated 29 October 2026.
Saywind is a dictation app for the Mac, published by Peiser Apcar Ltd (“we”, “us”), a company registered in England and Wales. This policy covers Saywind 1.1 and later, and this website. If you still use Saywind 1.0, a short section says what differs.
In short
- Saywind turns your speech into text on your Mac. Your audio and your words never leave it.
- There’s no account and no tracking, in the app or on this website. This website sets no cookies and has no analytics of its own.
- The app goes online for two things only: downloading speech models the first time each is needed, and checking for a newer version at most once a week.
- The app keeps one small note in your Mac’s Keychain: the date it first ran there, so that early-access users keep Saywind free.
- If you send us feedback or email us, we keep what you send to read it, reply and improve Saywind, as set out below.
Your voice and your words
Speech recognition runs on your Mac, on its own chips. Saywind never sends your audio or your text anywhere, so we never see them. When Saywind pastes text into another app, that app handles it like anything you type.
Saywind listens only while you hold your dictation key (you can change it in Settings), or while hands-free mode is on after a double-tap.
The microphone between dictations
A microphone can take a moment to wake up, and your first words would be lost. So between dictations, Saywind keeps the microphone running but muted, using the per-app input mute built into macOS. While it’s muted, Saywind receives only silence: it hears nothing until you press the dictation key, and nothing is recorded or kept. In Saywind 1.1 this is always on.
- macOS shows the orange microphone indicator only while you dictate. Some microphone-monitoring apps may still list Saywind as using the microphone between dictations, because it’s running.
- Saywind does this only with built-in microphones and ones connected by cable or USB, never Bluetooth. A wireless headset with a USB receiver counts as USB.
- Saywind lets go of the microphone while your Mac sleeps, while the screen is locked, and after 10 minutes without keyboard or mouse use.
Smart clean-up
On macOS 26 or later with Apple Intelligence switched on, Saywind can use Apple’s on-device language model to fix changes of mind and turn lists you say into bullets. It runs only on the dictations that need it. The text stays on your Mac, and Saywind sends nothing for it. You can switch it off in Settings › Writing.
What Saywind keeps, and where
Saywind stores these on your Mac only:
- Your dictation history: your most recent 2,000 dictations, with the app each one went into. Insights are worked out from it, on your Mac.
- Your dictionary, snippets and settings.
- The date Saywind first ran on this Mac, and its version. Saywind 1.1 writes them once, the first time it runs, as an item called “Saywind early access” in your login Keychain. It’s how this Mac will keep Saywind free when it starts costing £25, so it stays when you delete the app and still counts if you reinstall. It’s never sent anywhere.
Your history is in ~/. Your dictionary and settings are in Saywind’s preferences on your Mac. The early-access date is in your login Keychain.
Saywind doesn’t keep recordings of your voice. A hidden setting for developers can save them for testing; it isn’t in Settings and stays off unless someone switches it on in Terminal. Recordings kept by Saywind 1.0’s “Keep recordings” setting stay in the folder above until you delete them.
Learning your corrections
After Saywind pastes, it watches the text box you dictated into for up to a minute. If you fix a word Saywind got wrong there, it learns the spelling. This happens on your Mac only. You can switch it off in Settings › Learning (“Learn from my fixes”).
Saywind needs Accessibility access for this, and to notice the dictation key and paste your text.
When Saywind uses the network
Saywind connects to the internet for two things only:
- Speech models. Saywind downloads each speech model from Hugging Face (huggingface.co) the first time it’s needed: the main model on first run (about 630 MB), a small word-spotting model (about 100 MB) the first time your dictionary has a word in it, and any other model you choose. Dictation itself works offline.
- Update check. At most once a week, Saywind fetches a small file,
peiserapcar.com/, to see whether there’s a newer version. The request carries no identifiers and no cookies, and nothing about you or your dictation. You can switch it off in Settings › About (“Check for updates weekly”).saywind/ latest.json
As with any web request, the server that answers sees your internet (IP) address. Hugging Face’s own privacy policy covers its downloads. Our hosting provider, Cloudflare, runs peiserapcar.com and saywind.com for us and may keep request logs for a short time to deliver and protect them.
This website
saywind.com sets no cookies and doesn’t track you. It loads nothing from any other website: no fonts, no scripts and no trackers.
It has no analytics of its own. The only number it keeps is how many times Saywind was downloaded each day: a download adds one to that day’s count and records nothing about you. Cloudflare, which hosts the site, also shows us anonymous totals, such as how many requests it answered each day.
Feedback you send us
If you use the feedback form, we receive:
- your rating and your answers;
- the name you’d like shown and your email address, if you give them;
- whether you ticked “OK to show this on saywind.com”;
- the Saywind version, and whether you came from the app, when the app opened the form.
Why. We use it to understand what works and what doesn’t and to improve Saywind, to reply if you asked something, and to keep spam off the form. The law calls this our legitimate interest in running and improving Saywind. We quote you on saywind.com only if you ticked the box, and that rests on your consent.
Where. In a database that Cloudflare runs for us, and as a copy emailed to our inbox, which Google hosts.
Who else handles it. Cloudflare hosts this website and its database, and forwards email sent to hello@saywind.com. Google hosts our inbox. If we use an AI assistant to help choose quotes for the site, it’s Claude, made by Anthropic, and it sees only feedback you said we could show, never your email address. Cloudflare, Google and Anthropic are based in the US, so your feedback may be handled there. Cloudflare and Google take part in the UK Extension to the EU-US Data Privacy Framework (the UK-US “data bridge”), which UK law recognises as protecting it.
How long. We keep feedback, including the emailed copy, for up to two years, then delete it. Deleted feedback can stay in Cloudflare’s database backups for up to 30 days more. A quote shown on the site stays while it’s shown, and comes down whenever you ask.
Spam. To stop one network flooding the form, the site counts how many forms each network sends in a day. It does this with a one-way code made from the first part of your IP address (the part that names your network, not your own device) and a random value that changes every day. The address itself is never stored. Each day’s codes and random value are deleted just after midnight UTC; until then, a code could in principle be matched back to the network it came from. Deleted codes can stay in Cloudflare’s database backups for up to 30 days.
Your choice. Email hello@saywind.com to see, correct or delete your feedback, or to withdraw permission to quote you. We’ll take a quote down promptly.
If you email us
Email to hello@saywind.com is forwarded by Cloudflare to our inbox, which Google hosts (both US companies, as above). We use your address and your message only to reply (our legitimate interest in answering you), and delete the conversation two years after it ends.
If you use Saywind 1.0
Everything above applies to Saywind 1.0 too, except:
- Keeping the microphone ready between dictations is a setting, Instant start, on by default. With it off, Saywind uses the microphone only while you dictate.
- A “Keep recordings” setting, off by default, saves the audio of each dictation in
~/.Library/ Application Support/ Saywind - Saywind 1.0 checks for a newer version at most once a day, not once a week.
- Saywind 1.0 has no Smart clean-up, and keeps no early-access date in the Keychain.
- Its switches have older names. Learning is “Learn from my corrections”, in Settings under Behaviour. The update check is “Check for updates once a day”, in About. History is deleted with “Delete All History and Recordings…”, in History.
How to delete your data
To delete your history, use Delete all history in Saywind’s Settings › Data and privacy. Or quit Saywind, then delete the folder ~/. Moving Saywind to the Bin doesn’t remove this folder.
To remove your dictionary and settings too, quit Saywind and run defaults delete com.peiserapcar.saywind in Terminal.
The speech models (several hundred MB) are in ~/. Delete that folder to free the space. Other apps built on FluidAudio may use the same folder.
To remove the early-access date, open Keychain Access (search for it with Spotlight) and delete “Saywind early access”, or run security delete-generic-password -s com. in Terminal. This also removes this Mac’s proof that Saywind was installed during early access, so it won’t be free here automatically when it starts costing £25.
Your rights
Under UK data protection law (and EU law, if you live in the EU), you can ask us for a copy of the personal data we hold about you, and ask us to correct it, delete it or stop using it. Email hello@saywind.com, and we’ll reply within one month. If you’re unhappy with how we’ve handled your data, you can complain to the Information Commissioner’s Office (ico.org.uk) or to the data protection authority where you live.
Changes
If this policy changes, the new version will be on this page with a new date.
Who we are
Peiser Apcar Ltd, registered in England and Wales, company number 16157036, registered office: 8 Hestercombe Avenue, London, SW6 5LD. Email hello@saywind.com.