anonimize.it

Desktop, CLI, and OCR

The same idea across platforms: work with documents without shipping them elsewhere.

• Desktop for Windows and macOS A graphical application for day-to-day anonymization work.
• CLI for Linux A command-line tool for scripts, servers, and headless environments.
• OCR for scans Support for scanned documents when text is only available as an image.

What's new

1. November 5, 2026

   • New “Remove” mask type: the selected fragment disappears from the result with no replacement token. In the dual-pane view on macOS and Windows — strikethrough in the source, empty result preview, and undo. Removed content does not appear in TXT, DOCX/PDF export, the map, or the cache (including OCR scans). Available in the GUI (macOS/Windows) and in the CLI (detect / process --annotations).
   • Built-in product list of about 12,000 banks across the EU/EEA/UK/CH/Balkans (aitm/eu-banks seal, ECB/GLEIF attribution). Bank names from this list are detected automatically in all builds — without maintaining your own list.

2. November 4, 2026

   • By default, the text of repeated headers and footers is no longer included in the text output (TXT export, copy for AI, text preview). Detection and masking still run on the full text, and PDF/DOCX export keeps the layout and masks every occurrence. A setting in the macOS and Windows apps (and the CLI option --include-header-footer-text) brings back the previous behavior with labeled header/footer sections at the end.
   • Recognition of international phone numbers beyond Poland and Croatia: +CC / 00CC numbers with common separators, validated against numbering-plan lengths for EU/EEA countries, the UK, Switzerland, the Balkans and North America (NANP); national numbers starting with 0 only after a phone label (phone, Telefon, tél., telefono, Handy, mobile, fax…).

3. November 3, 2026

   • Masking of monetary amounts: new data type “Amount” (label [KWOTA_n]) for numbers with a currency marker before or after the digits (e.g. 500 zł, 100 €, $500), including amounts written in words next to a currency. Numbers, percentages and shares without a currency are not masked. Available in the rules lists of the macOS and Windows apps; reversible codes supported.
   • Well-known company names from a built-in list (banks operating in Poland, WIG20/mWIG40 companies, large IT and insurance firms), including name variants, Polish inflection and short uppercase tickers, are detected automatically in all builds — without maintaining your own list.

4. November 2, 2026

   • OCR now handles PDF pages built from many image tiles (“mosaic” scans): the whole page is rendered before OCR. On macOS the app clearly warns when OCR could not recognize text on a page.
   • Text output (TXT, copy to AI, preview) for text PDFs and DOCX now ends with separate “Header content” / “Footer content” sections instead of repeating headers and footers on every page in the body. Export in the original format keeps the original layout and masks every occurrence on every page. Does not apply to OCR scans.
   • Croatian OIB (11 digits with checksum, also with the HR VAT prefix) as its own data type on macOS and Windows with reversible codes, and Croatian phone numbers (+385/00385 always; national 0xx numbers when labeled as phone). Detection works regardless of document language.
   • Windows: on the “Decode” tab the two panels scroll in sync paragraph by paragraph, and a paragraph connector (ribbons linking each paragraph with codes to the same paragraph with restored data) can be toggled in Settings. Marking or unmarking a phrase now applies to all its occurrences in one click.

5. October 31, 2026

   • Windows: one simple anonymization model with no mode to choose — everything that is text (Text view, .anon.txt, Copy, also OCR text from scans and images) always gets reversible codes; everything that is a document (result preview, saving DOCX / PDF / scan / image) is always redacted (blacked out, irreversible). Scanned PDFs and images can now be saved as a redacted document. The “Decode” tab is a separate tool: paste or drop text with codes from many documents, see on a deanonymization summary page which documents the codes came from and which codes stayed unknown, and load the source document from there.

6. October 30, 2026

   • Windows and Linux CLI: anonymized text now uses reversible codes instead of plain labels; each document’s result is kept in an encrypted local document history (key stored in the system credential store), so reopening a file brings its previous result back. History page (Ctrl+H) with folders, search (also by code), sorting, deleting and retention settings. Decoding an AI response: paste text, or open/drag a .txt, .md, .docx or .rtf file — codes from many different documents are restored to the original data in one step. The CLI gained `registry` and `restore` commands.
   • Recognition of Polish locality (place) names based on official name registers (TERYT, PRNG) and the SGJP inflection dictionary, including inflected forms, masked as an address.

7. October 28, 2026

   • macOS: a new floating utility panel Window → “File info” shows the file name, engine status and session status; it stays on top and updates live with the active window. This info moved out of the main window so the document view is cleaner. Closing the panel does not stop processing. The Save button and File → Save… / Cmd+S are disabled when nothing has been detected or masked (no entities), so you cannot accidentally export an unchanged document; they become active as soon as a mask appears.
   • A single capitalized word between two parts of the same person’s name on the same line (e.g. a middle name between given name and surname) is now masked as part of that person’s name, even if it was not found separately.

8. October 27, 2026

   • Windows: the document original and its preview (DOCX, PDF, scan, image) appear right after loading, while data detection is still running; search (Ctrl+F) already works on the original. The progress card shows the current stage (reading text or document, scan OCR, detection) with character or page counts, and you can cancel.

9. October 25, 2026

   • Windows: new “AI answer” tab next to Text/Original (Ctrl+R). Paste labeled text on the left (typically an AI chatbot’s answer) and immediately see on the right the same text with original data restored from the current session’s map, highlighted by type. Restoration runs live in memory — no files with personal data are written to disk. Recognizes label variants AI tends to produce (e.g. [Osoba 1], [osoba_1], OSOBA_1) and warns about labels not in the map. Copy and Save (.odtworzony.txt). You can also load a previously saved JSON map without an open document.
   • Windows: text zoom — Ctrl+mouse wheel changes text size in the Text view and in the “AI answer” panels (75–225%); also Settings → General → “Text size” with reset to 100%. The level is remembered.

10. October 23, 2026

    • Windows: preview of the anonymized result for scanned PDFs and images directly in the document view (raster PDF or image); previously only available in the Text view.
    • Detection on large .txt files now runs in linear time — previously ~1 MB files could take over 10 minutes.
    • PDF export with many different entities is about 5× faster (e.g. a 19-page test PDF: from about 30 s down to about 6 s).

11. October 22, 2026

    • The URL rule also masks bare domains (e.g. example.pl without http/www), bare IPv4 addresses and host:port, plus surnames glued to a domain or e-mail. It uses a conservative TLD allowlist to avoid false positives on abbreviations.
    • User-defined permanent mappings in .aitm files: keep lists of phrases (priority: company names) that are always masked. Matching tolerates letter case, legal forms (S.A./SA/sp. z o.o.), spacing and ampersand, and light Polish inflection. Available in the CLI (--aitm, --aitm-dir, --no-aitm) and in the Windows app (choose the .aitm folder, entry counter). Not in the macOS UI yet.
    • An anonymization audit card (for IT/archive, GDPR Art. 5(2) accountability) without originals and without replaced values: only counts per data type, file extension and SHA-256; JSON + HTML. CLI flag --audit PATH and `anonimize audit --map …`; Windows menu “Save audit card…”. Not in the macOS UI yet. The card is also produced when processing is blocked (fail-closed).
    • PDF: file attachments embedded in the document (EmbeddedFiles, FileAttachment annotations) are fully removed from the output, with a warning.
    • Better OCR on low-resolution scans: small embedded images are upscaled before OCR, so names on low-res scans are read more reliably; no slowdown on normal A4 scans.

12. October 21, 2026

    • PDF: scanned pages are recognized with OCR in parallel on several cores (up to 8 pages at once). Blank pages in a scan no longer interrupt processing — they are skipped, and the app warns which pages were blank.
    • PDF: correct text from files with non-standard character encodings (for example MacRoman, /Differences tables, form fields) and from compressed object streams, so Polish letters (ą, ł, ś…) are read and anonymized correctly. Page splitting and export also work more reliably for PDFs with unusual syntax.
    • Better detection of given names and surnames hidden in e-mail addresses and domain names, including under OCR noise, and of given names written in lowercase.

13. October 20, 2026

    • New data type: URLs. Detection and masking of full addresses (http/https/ftp/sftp/ssh/file/git/ws/mailto, www. without a scheme, IP with port) as [URL_n], with an option to turn this type off. Hidden hyperlink targets are cleaned too — in DOCX (HYPERLINK fields, tooltips, links on images) and in PDF (URI actions, links to external files).
    • macOS: a navigator of detected data in the left pane — a tree by pages (PDF, DOCX) or by type; a click scrolls to the place in the document. When a PDF has locked forms, the app does not stop: it saves anonymized text (TXT) with a map and shows a warning.
    • Windows: “Don’t anonymize” applies to every occurrence of the same value in the document, after confirmation.
    • Windows: “Open with → Anonimize.it” in File Explorer for supported formats (TXT, DOCX, PDF, RTF, images), without taking over the default program.
    • Windows: confirmation before replacing the open document with a file dropped onto the window.
    • Strings shaped like PESEL, NIP, REGON, KRS, VIN, IBAN/NRB, ID card numbers, passport numbers, or dates are masked even when the checksum is wrong or the date does not exist on the calendar (e.g. a typo in a PESEL). When context does not resolve the type, the fragment is labeled “ambiguous” instead of being skipped.

14. October 19, 2026

    • DOCX result saving is about 10× faster — on a long test document (~97k characters, ~1,600 detections) from about 26 s down to about 2.8 s; in the app this also speeds up refreshing the result preview after each edit.
    • Windows: edit the specific selected occurrence in the DOCX/PDF preview (not the first one in the document); a mouse selection in the preview opens editing immediately, and a right-click on a highlighted detection lets you change its type, skip it, or remove it.
    • Windows: you can open a document that is currently open in Word (the app reads the last saved version).
    • PDF: text stored in forms and fields (including form appearance content such as checkboxes) is now detected and anonymized together with the page body.
    • Windows: save, copy, error, and warning messages appear in the window footer, color-coded by type, and no longer cover the result.

15. October 18, 2026

    • Windows: original DOCX and PDF preview with highlights on detected data, image preview, and synchronized scrolling of both panes (including the scrollbar).
    • Windows: select a fragment in the document preview and mark it for anonymization from the Mark menu.
    • Windows: the installer installs the WebView2 Runtime automatically when it is missing.
    • Windows: new window layout — a bar with detection count and type legend, search in the panes, and save/copy of the result from the result pane toolbar.
    • macOS: a warning when the name-recognition model has not loaded; saving, copying, and dragging the result then require an explicit confirmation.
    • PDF: an image that cannot be processed safely is left out of the output document.

16. October 17, 2026

    • Windows: the app opens DOCX, PDF (including scans), RTF, and images; OCR (Tesseract) is installed with the application.
    • Windows: data-type legend, colored underlines, and F3 to move between detections.

17. October 16, 2026

    • Windows: a new Windows app with an installer — detecting data in text, manually editing detections, saving the result, a settings page, and a warnings panel.
    • Windows: a pseudonymization map (JSON) can restore original values in labeled text, for example in a reply from an AI assistant.
    • Recognition of inflected Polish given names (e.g. “Annie”, “Pawła”) using the SGJP inflection dictionary.
    • Masking with # characters no longer reveals the length of the original word.

18. October 15, 2026

    • DOCX: headers, footers, footnotes, and document metadata are anonymized as well.
    • macOS: quick preview of the original with colored marks in DOCX, and double-click to jump to a detection.
    • PDF: more reliable text replacement in files with embedded fonts (ToUnicode) and in court PDFs.

Why anonimize.it

• Work with AI safely Mask → work in AI (e.g. Claude Code, ChatGPT) → restore the data. Sensitive values in text become reversible codes; the local document history lets you restore the originals afterwards.
• Linux CLI Mask and restore from the terminal, ready to use in your own scripts.
• Fully offline Data does not leave your computer; the app works without internet. No telemetry and no online activation.