Privacy policy

Retail Online LLC, 1209 Menaul Blvd, Albuquerque, NM 87107, USA.

For anything concerning personal data: privacy@spamguard.pro.

If you run a website using SpamGuard, we are a processor acting on your instructions for the submissions your visitors make. You are the controller of that data, and the Data Processing Agreement governs it.

If you visit this website, or write to us, we are the controller of that small amount of data.

When a visitor submits a comment, registration or contact form on a site running SpamGuard, that site sends us: the submission text, the name and email address entered, the visitor’s IP address, the browser identification string, and the referring page.

If the visitor has JavaScript enabled, a small set of measurements about how the form was filled in is sent too: elapsed time, keystroke count, correction count, paste count, a sample count of pointer movement, and two statistics describing typing rhythm. This set contains no text, no key identities and no pointer coordinates. It cannot be used to reconstruct anything that was typed.

Separately, and only if the site owner presses the button that offers it, the site can send us its existing backlog: the comments a human classified before SpamGuard was installed. That is the spam folder and the moderation queue, and a matching sample of the comments the site approved and published. The approved ones are included because a spam folder on its own is a single class of example, and a detector trained on nothing but spam learns to see spam everywhere; the sample is capped at the size of the spam side so neither can dominate. This is a bulk transfer of historical submissions, up to 2,000 of each kind, often years old, and it includes their text along with the name, email address and website each author entered at the time. It never happens automatically, it is refused outright unless “help improve detection” is on, and the screen states the counts and says plainly that the text leaves the site before the button can be pressed.

We store a one-way cryptographic fingerprint of the submission and a set of derived numbers — link counts, character ratios, language-independent structural measurements. We do not store the submission text.

There are two exceptions, and both require the site owner to have explicitly switched on “help improve detection”, which is off by default. Submissions the owner personally corrects are stored in full so the model can learn from the mistake. An imported backlog, described above, is stored in full for the same reason.

Of what a protected site sends us, we keep the submission text only in those two cases. The author’s name, email address and website are never stored at all — they are used to compute the verdict and discarded with the request.

Records are deleted automatically: 30 days on the free plan, one year on Pro, two years on Business.

Text kept for training is an exception to that, and is kept indefinitely. That covers both a correction and an imported backlog. Deleting it would remove the example from the model and undo the improvement it paid for. If you want it gone, ask — see “Your rights” — and it is removed on request; there is simply no clock that removes it on its own.

On servers in the European Union. Our database is hosted in Frankfurt and our application in the EU.

No submission content is sent to any third-party artificial intelligence provider. Detection runs on models we operate ourselves. This is an architectural fact, not a policy commitment we could quietly change: sending content elsewhere would require rebuilding how the service works.

For the site owner: performance of the contract between us.

For a visitor to a protected site: the legitimate interest of the site owner in keeping their forms usable, balanced against the visitor’s interests. We designed the data minimisation described above specifically so that balance comes out favourably — we take what is needed to judge a submission and no more.

Our infrastructure providers, acting as sub-processors under contract: Railway (application hosting, EU) and Supabase (database, Frankfurt).

Nobody else. We do not sell data, we do not share it for advertising, and there is no analytics or tracking on this website.

Access, rectification, erasure, restriction, portability and objection, under the GDPR.

If you commented on a site using SpamGuard, address your request to that site first — they are the controller and their WordPress installation can export and erase what we recorded, using the plugin’s built-in integration with WordPress privacy tools.

You can also write to privacy@spamguard.pro directly. You have the right to complain to your national supervisory authority.

This website sets no cookies and runs no analytics.

The plugin sets no cookies on the sites that use it.