Privacy

This site is run by one person. It is not a company, it has no advertising, and it has no commercial interest in anyone's data. What follows is a plain description of what is actually stored and where it goes — written to be accurate rather than to be broad enough to cover anything.

What the website stores about every visitor

You do not need an account to use anything here. So that your work is not lost, the first time you open a page the server gives your browser a cookie called tb_id. It holds a random number, lasts two years, and is only ever read by this server. Against that cookie the server records:

• When you first and last visited, how many separate visits you have made, and which of the apps you opened.
• Your IP address, both the first one seen and the latest, and the IP address of each visit.
• What your browser reports about itself: its user-agent string (and the browser, operating system and device type guessed from it), preferred language, time zone, screen size, and the site that linked you here, if any.
• The tools, machines and settings you save in an app, so a cleared cache does not cost you your library.

The site owner can see this list and may add a short private note to a visitor, for example to recognise a returning one. It is not used for advertising and is not shared with anyone.

If you create an account

An account is optional. If you make one, the server also stores:

• Your email address and, if you give one, a name.
• A hash of your password. The password itself is never stored and cannot be recovered from the hash, by anyone, including the site owner.
• A second cookie, tb_session, that keeps you signed in for up to thirty days. It is deleted when you sign out. The server records the IP address and browser each sign-in came from, so a session can be recognised and revoked.

Signing in attaches the browser you are using to your account, so the library you built there becomes the account's and follows you to other computers you sign in on.

Analytics

Page views are also counted with Umami, self-hosted on the same server as the site. Umami sets no cookies. No data is sent to Google Analytics or any other third-party analytics service, no advertising identifiers are collected, and no profile is built across other websites.

If a page breaks in your browser, the page tells this server which page it was (without anything after a ? in its address), the error message, the file and line it came from, and your browser's family (such as “Chrome on Windows”), so it can be fixed; no IP address, cookie or account is stored with it.

The inbox-triage application and Gmail data

inbox-triage is a private, single-user application. The only mailbox it is authorised against is the site owner's own. It has no sign-up, no other users, and no mechanism by which a third party's mail could reach it.

It requests three Google scopes, and uses them for exactly this:

• gmail.modify — to read messages in order to judge their importance, and to apply labels such as AI/Urgent back onto them. It is not used to delete mail.
• gmail.compose — to create draft replies for the account owner to review. The application does not send mail; a draft becomes a sent message only when a human presses send.
• calendar.events — to copy the owner's own calendar, including dates read out of mail such as deliveries and return deadlines, into the owner's Google Calendar. Granted separately, and only if the owner connects it.

Classification runs on the owner's own server. Most messages are decided there and their contents never leave it. Messages that the local stage cannot decide confidently are sent to Anthropic's API for classification, and the text of a thread is sent there when a draft reply is being prepared. Anthropic does not train models on data submitted through its API. No message content is sent anywhere else.

What is retained on the owner's server is not a second copy of the mailbox. For each message it is the classification result: a label, a score, the reasons for it, and the message identifier. From order, shipping and receipt emails it also keeps the details those emails exist to convey, such as the vendor, the amounts, the order reference and the tracking number. It also keeps the text of any draft reply prepared for the owner. The personal parts of those records are encrypted in the database. Google credentials are held as refresh tokens, encrypted, in that same database, with the key kept in a root-only file outside it.

Access can be withdrawn at any time from Google Account permissions. Revoking it immediately stops all further access, and the stored classification data is deleted on request.

What is never done

• Nothing is sold, rented, or shared with data brokers.
• There is no advertising on this site and no advertising network embedded in it.
• No personal data is used to train any machine learning model.
• No profile is built across other websites.

Deleting your data

Ask, and your account, or the visitor record for your browser, is removed along with everything attached to it. There is no retention period to wait out and no copy kept for analytics.

Contact

contact@zanbur.tech