Privacy Policy

Last updated: October 25, 2026

Thiruttham (“we,” “us,” “Thiruttham”) is a Tamil writing assistant — a Chrome extension and web editor at thiruttham.com — operated by Midhunan Vijendra Prabhaharan, an individual/sole proprietor (no separate registered company).

What we don’t store

We never use your text to train AI models, and we keep as little of it as we can. A rule check (the fixed rules, including the marks shown while you type) runs in your browser and sends nothing to us. When you are signed in and press Check in the extension or the web editor, the text is sent to our correction engine and to the AI model (see “How your text is processed”). To avoid paying for the same work twice (which keeps checks fast and keeps our costs — and your price — down), the AI’s result for each sentence is cached, encrypted at rest, for up to 30 days and then deleted automatically. That result holds the suggested changes, which quote the words being changed; it is filed under a one-way fingerprint of the sentence, not under the sentence itself. Nobody reviews this cache; it exists only to return the same result faster when the same sentence is checked again. On a Pro plan or a trial, for a word the fixed rules cannot decide on their own, we may also put a second, smaller question to an AI model through the same provider and under the same terms (the sentence, that word and the change the rules propose); only its yes-or-no answer is cached, in the same way and under a one-way fingerprint, and the answer holds no text. The one other exception is if you explicitly choose to “report a correction” (see below) — that one sentence and the correction shown are kept longer, for our own review, with your active consent at the time.

Drafts stay on your device. The web editor saves your drafts in your browser, on your device only, so your last one is waiting when you return, and with each draft the corrections of its last check (so a reload does not lose them); they are not sent to our servers and are not part of your account. Anyone who can use your browser can open them, and clearing your browser’s site data deletes them. Your text leaves your browser only when, signed in, you press Check, report a correction or protect a word.

What we do collect

• Your protected words, kept in your account and in your browser. The words you ask us never to change are stored in your account, and the extension also keeps a copy of that list in your own browser, so your names stay protected when you are offline. It is cleared when you sign out.
• Your settings, kept in your browser. Your language, theme, spoken-word suggestions and Tamil-typing choices are stored in your browser. If you also use the extension, thiruttham.com and the extension pass these four choices to each other inside your own browser so they match; they are not sent to our servers. "Continue in the web editor" hands the text you choose from the extension to the editor the same way: inside your browser, not through our servers.
• Account information. If you sign in (via Google), we receive your name, email address, and Google account ID, used to identify your account and manage your subscription.
• Usage events, without your name, email or account. We record whether a correction was accepted or rejected, when checks happen, and similar counts (for example how many corrections were shown) — never the text itself. For each check, the event says where it ran (the editor, a page’s check box or the extension, which also says whether the text box was on Gmail, WordPress, Blogger or another site, decided on your device without sending the address; on the typing page it also counts that the box was used and that a visitor took their text to the editor or the sign-in page, never the text), whether the AI read the text and, if not, why (for example not signed in, or the day’s allowance used), your plan, and sizes only as ranges (how many words, how many corrections). Each event carries a random identifier kept in your browser (on the website) or in the extension, so we can tell that two events came from the same browser or install; it is not linked to your account, and clearing your browser’s site data or removing the extension resets it. When a page or the extension fails, we count that it happened, where, and the kind of error (for example a type name) — never the error’s message or details, which can quote text. If you cancel a subscription, the reason you pick from a short list (or that you picked none) is one of these events; there is no box for your own words. These events help us understand whether the product is useful and where it’s failing.
• How you first arrived, kept in your browser. On your first visit to a public page, the website stores in your browser the page you landed on (without anything after the “?” in its address), its language, the campaign tags in the address if there were any (utm_source, utm_medium and utm_campaign), and the name of the site that linked to it (its host name only, never the page). It is attached to a few events (choosing Add to Chrome, signing up, starting the free trial, starting a checkout, paying, and installing the extension) so we can see which pages bring writers. It stays until you clear your browser’s site data.
• Daily usage numbers, tied to your account. For each day you use the AI check we keep a few numbers against your account: which plan you were on, how many checks, how many words were read, how many corrections were shown, and roughly what the check cost us. Never the text itself. We use it to keep the AI affordable and to set fair limits. On the Free plan, to count your daily AI corrections fairly, we also keep one-way fingerprints of the corrections already shown to you that day, so the same fix is not counted twice; they are replaced the next time you use the AI check on a later day. Daily allowances reset at midnight India time. All of this is deleted with your account; encrypted backup copies made before then age out on our backup schedule.
• Combined daily totals, with no account attached. For each day we also keep those usage numbers added up per plan, plus the day’s highest cost for a single account in that plan, with nothing that says whose they are. We keep them, after an account is deleted too, to see how costs change over time. While few people use Thiruttham, one of these groups can be close to one person’s numbers.
• Correction counts, with no account attached. Separately from the usage events above, we keep daily counts for each type of correction (the rule that made it): how often it was shown, accepted, dismissed, reported or undone, and how often “Don’t change this word” was used; and, by reason, how often a suggestion from the AI was blocked by our safety checks before you saw it. Only counts: never your text, never the words you protect, never the blocked suggestion. The web editor sends them, and so will the extension from its next release, with nothing that links them to you, your account or your device. While few people use Thiruttham, one day’s count for a rarely used correction can be close to one person’s actions. We use them to find corrections that are wrong, so we can fix or switch them off. Because nothing links them to you, they cannot be removed when you delete your account.
• A list of switched-off corrections, downloaded, nothing sent. The website and the extension download from us a short list of corrections we have switched off, so a correction found to be wrong stops appearing without waiting for an update. The request contains nothing about you or your text.
• Trial and subscription steps, tied to your account. The dates of the steps on your way to a subscription: your trial starting, checkout starting, a payment succeeding or failing, and a subscription ending, with whether you chose the monthly or the yearly plan. No amounts and no text. We use them to understand how people come to pay and where they stop. They are deleted with your account; encrypted backup copies made before then age out on our backup schedule.
• Reported corrections (opt-in only). If you use the “report this correction” feature, the one sentence involved and the correction shown are stored, with your explicit consent at that moment, so we can investigate and improve the correction engine. Nothing else from your document is shared.
• Billing information. Subscription and payment details are handled by our payment processor (Razorpay); we do not store your card or bank details ourselves.
• Your email address, if you join the launch list. We keep it to send you launch updates, and send one welcome email through Resend. Write to us to have it removed.

How your text is processed

Checking your writing requires sending it to a third-party AI language model to generate corrections. We only use language model providers whose terms contractually exclude using your submitted text to train their models. As of launch, this is Google’s Gemini model, accessed through OpenRouter under a Zero Data Retention (ZDR) routing guarantee — OpenRouter states that a provider under ZDR “will not store your data for any period of time” and “cannot train on your data.” We do not keep the text you send for a check, apart from the encrypted 30-day results cache described above (which quotes the words being changed) and any correction you choose to report.

Third parties we work with

To operate Thiruttham, we rely on: Supabase (account/authentication, database), Cloudflare (hosting, network infrastructure), a language model provider (correction processing, see above), Razorpay (payment processing), Resend (transactional email), PostHog (the usage events above), Google Analytics (page visits on this website and the few events named above), and Sentry (error monitoring on our servers: it receives the type and location of an error, never your text). Each processes only what’s needed for its function, under its own privacy terms.

No session recordings. We do not record sessions (the clicks, scrolling and screens of a visit) at the moment: session recording is switched off in PostHog, and Microsoft Clarity, which we used for this before, is no longer loaded. Recordings made before we switched it off stay with Clarity under its own retention settings. If we turn recording back on, we will update this page first.

Your choices

• You can add words to your personal dictionary so they’re never flagged — this is stored against your account, not shared.
• You can turn off checking on specific sites (“pause on this site”).
• You can request access to or deletion of your account data at any time by emailing support@thiruttham.com. We delete your account and the data tied to it, such as your personal dictionary. Payment and invoice records may be kept for as long as the law requires for accounting and tax, and Razorpay keeps its own transaction records under its own terms. Cancel any active subscription first, or tell us in the same email and we will cancel it for you.

Children’s privacy

Thiruttham is not directed at children under 13 (or the relevant age of digital consent in your jurisdiction) and we do not knowingly collect data from them.

Changes to this policy

We’ll update this page if our practices change materially, and note the date above.

Contact

Questions about this policy: support@thiruttham.com.