Privacy policy · effective 28 October 2026

Your pages stay yours.

This policy covers the Leita browser extension (version 1.9.3, and earlier versions where it says so), the web playground and this website. It is written to be read, not skimmed.

The short version

  • The extension reads a page only when you open its find bar on that page.
  • A word, or a phrase of up to three words, that appears on the page is matched in your browser and never leaves your device.
  • Anything else you type is a meaning search: when you search, your query and the page's text (up to 480 passages; up to 960 in PDFs and Google Docs) go to our search service to be scored by an AI model, with no question asked first. Our server doesn't save them.
  • The extension has no accounts, no analytics and no ads. Versions 1.9.2 and later send their search requests without cookies; version 1.9.1 carried the cookies your browser holds for getleita.com (see Cookies below).

What the extension reads

When you open the find bar, with its key or the toolbar button, the extension reads the text of that page, only then and only there. It does not read the values of form fields, plain text areas or password fields, your cookies, your history or the content of other tabs. Text in rich-text editors on the page (a Notion page, an email or chat message you are writing, a comment box) is read like any other page text and, for a meaning search, sent to be scored. Closing the bar removes the highlights.

The find bar is drawn inside the page, so the page's own scripts can see what you type into it, as with any text box on that page.

While Ctrl+F / ⌘F opens Leita (on by default since version 1.9.3, also after the update unless you had turned it off; you can turn it off), a small script on websites listens for that one key. It reads no page text and sends nothing.

Exact searches

A word, or a phrase of up to three words, that appears on the page is matched in your browser, like regular find. Nothing about it leaves your device.

If what you typed isn't on the page (a word that isn't there, a typo) or is longer than three words, Leita searches by meaning instead, once you pause typing: on websites after a pause of 0.22 to 0.6 seconds, depending on how fast you type (1.2 seconds after a space); in PDFs and Google Docs after 0.6 seconds. Because it searches as you type, an unfinished word that isn't on the page can be sent too. Version 1.9.1 waited 0.6 seconds everywhere.

Meaning searches

A meaning search goes out when you search: there is no question first, as in version 1.9.1 (version 1.9.2 asked once before the first one; 1.9.3 no longer does, and ignores the answer 1.9.2 stored). The welcome page, the store listing and this policy say what is sent. A meaning search sends your query and the page's text, in passages of up to 1,200 characters (up to 480 passages from a web page, up to 960 from a PDF or a Google Doc), over HTTPS to our search service: this site's server, hosted by Lovable on Cloudflare. The server asks an AI model (TypeSafe Jev, through the Lovable AI gateway) to score each passage and sends the scores back. A second request then sends the sentences of the best matches, so the bar can highlight the sentence that answers.

On a web page both requests carry only the text Chrome renders: text the page hides with display:none, visibility:hidden or the hidden attribute, text in a closed section, scripts, styles and the defaults of text areas are left out. Text that is rendered but not easy to see, such as labels meant only for screen readers, is sent like the rest. In version 1.9.1 the second request used the matching paragraph's full text, which could include text the page hides inside it.

The page's address and title, and what you have typed into form fields, are never sent.

  • Our server doesn't save your query or the page's text.
  • To answer a repeated search instantly, the server keeps, in memory, a one-way hash (SHA-256) of your connection, the query and the passage together with its score. An entry is never used after 15 minutes and is deleted at the next search after that. The hash cannot be turned back into the text.
  • To prevent abuse, requests are rate limited using a salted hash of your connection's address. It is kept only in memory and removed a couple of minutes after your last search, the next time the server handles a request.
  • Our server logs one line per search: the status, passage counts, timings, the client version (for example ext/1.9.3) and, when calls to the AI service fail, how many failed and its status code. Errors are logged as a status code and a fixed category. Neither ever holds query or page text. The hosting provider (Lovable on Cloudflare) may log the connecting IP address and browser details of each request for operation and security.
  • The Lovable AI gateway and the model provider receive only your query and the passages: not the page's address, not your IP address and no identifier. They process them to produce the scores, under their own terms.
  • If you set your own search service address on the extension's Advanced page, meaning searches go to that service instead, under that service's own policy.

Cookies

The extension has no cookies of its own. Versions 1.9.2 and later send their search requests without cookies: none of the cookies your browser holds for getleita.com goes with them, and a cookie the answer tries to set is not stored.

In version 1.9.1, Chrome attached the cookies it holds for getleita.com to the extension's search requests: __cf_bm (Cloudflare's bot management, set by the first search's response) and, if you have visited this website, __dpl and session-id (see "This website" below). Our code doesn't read them; Cloudflare and Lovable's hosting use __cf_bm and __dpl to protect and route requests, and Lovable's analytics uses session-id.

Google Docs

Each time you open the bar on a Doc, the extension reads the document's text from the open page and also fetches the document's plain-text export from Google with your own Google session. That text stays on your device, except what a meaning search sends as described above.

PDFs

The extension loads the same file again from the address you are viewing (usually from the browser's cache) with your browser's normal session, and extracts its text on your device with a bundled copy of Mozilla's pdf.js. The file itself is never uploaded; a meaning search sends its text as described above. The browser's session storage holds the address and text of up to three recently searched PDFs and, briefly during a PDF jump that reloads the page, your query and results. It is cleared when the browser closes.

Files on your computer

If you allow access to file URLs on the extension's details page, local pages and PDFs are read the same way, only when you open the bar on them.

What is stored on your device

Your Ctrl+F / ⌘F choice, whether the one-time PDF shortcut tip has been shown, the answer version 1.9.2 kept to its meaning-search question (later versions ignore it), an optional custom search service address, and the session items above (with a flag, cleared when the browser closes, that says the key listener in open tabs is up to date). Nothing else. Removing the extension deletes all of it.

Permissions, and why each one exists

  • Access to all websites (shown by Chrome at install as "read and change all your data on all websites"): Chrome can only let an extension react to a key on sites it is allowed to run on, and a find tool has to be one key press away on any page. This access is used for two things: opening the bar on the current page when you press its key, and, while its switch is on (by default since version 1.9.3), a small (under 5 KB) listener for Ctrl+F / ⌘F that does nothing except open the bar. Page text is read only when the bar is open on that page. The same access covers the call to the search service and, in Google Docs, the fetch of the document's text export and, in a PDF, loading the file again to read its text. You can narrow it at any time on the extension's card in chrome://extensions.
  • activeTab: gives the extension the current tab when you click its toolbar button or press its shortcut.
  • scripting: injects the find bar into the current tab at that moment, registers the Ctrl+F / ⌘F listener while that switch is on and, in a PDF, passes a message to Chrome's viewer so the bar knows when the passage is on screen. Nothing else is injected on page load.
  • storage: saves the Ctrl+F / ⌘F choice, the optional custom service address and the per-session PDF text described above.
  • offscreen: runs the PDF text extraction in a hidden extension page, on your device.

This website

This site is hosted by Lovable on Cloudflare. Cloudflare may log IP addresses to keep the service running and secure. The site sets three cookies: __cf_bm (Cloudflare's bot management, 30 minutes), __dpl (Lovable's hosting, 24 hours) and session-id (Lovable's site analytics, 30 minutes). Lovable's analytics record each page view with its full address (including anything after the "?", so campaign tags such as utm_source too), the referring site, your browser's user agent and language, a country guessed from your time zone, and that session id. They also record page-speed measurements (web vitals) and how long a page was in use (engagement time). Errors on the site are reported to Lovable with the error message, its stack trace and the page path. Fonts are served from this site, not from a third party.

The playground works like the extension: a word found in its text is matched in your browser, and meaning searches send the sample or the text you pasted to the search service described above. The playground sends its searches without cookies.

Sharing

No data is sold, shared for advertising or used to build profiles. The extension contains no analytics, advertising or tracking code; its only third-party code is an unmodified copy of Mozilla's pdf.js, which makes no network requests.

Leita's use of information it receives will adhere to the Chrome Web Store User Data Policy, including the Limited Use requirements.

Children

The extension is a general-purpose tool and does not knowingly collect information from anyone, including children.

Changes

If this policy changes, the new version will be posted here with a new effective date.

Your rights and contact

Our server saves no query or page text, so there is none of it to hand over or delete. The website's analytics and hosting logs, kept by Lovable and Cloudflare, can hold visit records and IP addresses. Email us and we'll help you see or delete what we can.

Published by Axel von Essen.

Email vonessen.axel@gmail.com.