Data policy
Plain language. Last updated 2026-11-08.
The short version
- No email needed. Your account is a random 16-digit number, and we store only a salted hash of it.
- Your Style is yours. Writing you import to build your style is stored: about 12 representative samples plus a numeric fingerprint, along with any text you accept or 👍. It's used only to rewrite drafts for you, never to train shared models without opt-in, and you can export or delete it any time.
- Anonymize comparisons never reach our database. When you paste your own writing so that Anonymize can push away from it, the server fingerprints it in memory and discards it. Nothing is stored.
- We don't keep your originals. Drafts you paste into the editor, the Stylometry Score page or Anonymize are processed and then dropped. They aren't written to disk.
- Learning from accepted text is optional per voice. If it's on, we store the text you accept, never your draft. Turning it off deletes it.
- Delete means delete. Deleting a style removes its style profile, samples, fingerprint, accepted text and feedback in one database transaction. Deleting your account removes everything you own.
- Share cards hold numbers. A Stylometry Score card stores the scores and style traits. It never stores your text.
What we store, exactly
| Data | Stored? | Where | How long |
|---|---|---|---|
| Account number | Salted SHA-256 hash only | server DB | until you delete the account |
| Only if you add one for receipts | server DB | until you remove it | |
| Your Style: writing samples (≈12) + fingerprint | Yes | server DB | until you delete the style |
| Anonymize comparison samples / fingerprint | No | in memory during one request | discarded at the end of the request |
| Drafts / originals | No | in memory during one request | discarded at the end of the request |
| Voice style profile | Yes | server DB | until you delete the style |
| Accepted text | Only if learning is on | server DB | until you delete it or the style |
| 👍 / 👎 feedback | Yes: the rated passage | server DB | until you delete the style |
| Usage counters (words per day) | Yes: numbers only | server DB | rolling |
| Payment records | Amount, product, status, processor reference | server DB | as tax law requires |
| Anonymize credit codes | Hash only; the code is shown to you once | server DB | until the credits are used up |
| Server logs | Method, path, status, timing. No request bodies. | server | 14 days |
Who else sees your text
To rewrite text, we send the paragraph being rewritten, your style's style profile and samples (including recent accepted or 👍 text) and up to three 👎 passages to the frontier-model provider configured for the service (by default Anthropic's Claude API). We don't send Anonymize comparison samples, your account number or your email. Scoring runs on our own judge, so scoring alone never sends your text to a third-party AI.
Payment processors (Stripe for cards, NOWPayments for crypto) see what they need to take a payment. A crypto purchase of anonymize credits involves no account at all.
Training
Corrections you flag ("that's not me") and published history may be used to improve your own persona's voice model. We don't use them to train shared models unless you opt in, and the opt-in is off by default.
Requests
Everything is self-serve: export a persona, burn a persona, delete your account. We can't locate an account from a name or an email, because we don't hold that link.