Skip to content
unformation

Changelog

Changelog

Every release, including security fixes and policy changes, is listed here.

  1. v0.2.2, Sample document in your language

    • The sample document follows the page language: English pages load an English agreement, Turkish pages the Turkish one. The English sample is read as English, so smart detection offers the smaller English model (~110 MB) for it instead of the multilingual one.

  2. v0.2.1, Split names and PowerPoint templates

    • First and last name in separate columns: when a person's full name is known, from smart detection or from your dictionary, the two parts are replaced where they stand alone in a CSV column, a spreadsheet cell, a JSON value or a table cell. Running text is not affected.
    • PowerPoint: text on slide masters and layouts, notes and handout masters, SmartArt and chart titles and labels is scanned, so a company name in the template footer no longer survives.

  3. v0.2.0, Leak sweep

    • Document properties are cleared in full: author, title, subject, category, company and manager, also in files written by LibreOffice or openpyxl.
    • Author names on Word comments and tracked changes, PowerPoint comments and Excel person lists become Author 1, Author 2; account e-mail addresses are removed.
    • Smart detection: a name or company found once is replaced everywhere in the file, including headers and capitalised signature blocks. Fragments of words and column headers such as IBAN or TCKN are no longer mistaken for names.
    • Dictionary and remembered pairs match capitalised spellings with Turkish İ/ı and names joined by a non-breaking space.
    • Excel: national IDs, tax numbers and mobile numbers stored as numbers are replaced (the cell becomes text); cached formula results that hold a detected value are cleared and recalculated on open; comments are found in every file layout, threaded comments included. Pivot caches are not scanned.
    • HTML: form values, mailto:, tel: and sms: links and data inside script blocks are scanned; comments are removed from the copy.
    • De-anonymizer: labels match as whole words, so an unknown “Person 11” is no longer turned into the name behind “Person 1”.
    • Tool: each step opens at its top with keyboard focus in place; long addresses wrap in the mobile preview; files that could not be processed are listed on the result screen.

  4. v0.1.0, Initial release

    • Anonymize DOCX, PPTX, XLSX, PDF, TXT, MD, CSV, JSON and HTML entirely in the browser; up to 10 files, 50 MB each.
    • Office files edited in place at the XML level so formatting, headers, comments and tables are preserved; PDFs returned as DOCX + Markdown.
    • Three detection layers: patterns with checksums (email, phone, IBAN, card, Turkish national ID and tax number, URL, IP, plates, addresses), personal dictionary, optional in-browser model (~110 MB EN, ~180 MB multilingual).
    • Review list with per-item toggles, replacement rules (numbered placeholders, fixed tokens, custom text) and consistent mapping across a job.
    • Mapping table export (CSV / JSON) and a de-anonymizer to restore names in an AI's answer.
    • Text mode for pasting snippets; Markdown output for every job.
    • Offline support through a service worker; installable as a PWA.
    • English and Turkish interface with localized routes.
    • Security headers (CSP, HSTS, X-Frame-Options DENY), self-hosted fonts, cookieless analytics, no accounts.