Skip to content
anonymizer

Web UI

After install, run:

Terminal window
anonymize

A browser tab opens at http://127.0.0.1:<port>. The port is random per session — no fixed port to avoid colliding with other local services.

Main flow

  1. Drop a file into the drop zone, or click to select. Supported: .docx, .xlsx, PDFs with a text layer, scanned/hybrid PDFs when local Tesseract OCR is installed, plain text (.txt, .md, .csv), HTML (.html, .htm), and RTF (.rtf, downloaded as .docx). Files up to 100 MB by default (override via ANONYMIZER_MAX_UPLOAD_MB). See Supported formats for per-format coverage.
  2. Review detected entities — the UI lists what was found, grouped by category. Each entity shows its surrounding context (5 words on each side).
  3. Configure which categories to redact, individually.
  4. Download the anonymized file. The original file on disk is never modified.

For scanned PDFs the progress list includes an OCR recognition step. For normal DOCX/XLSX/text-layer PDFs that step is skipped. If OCR is missing, the UI shows Tesseract installation guidance instead of accepting a scan and producing an incomplete file.

OCR output is still a PDF: the downloaded file keeps the original pages and adds redaction/token overlays. It is not an editable recognized DOCX export.

Manual mask

If the detector missed something (e.g., a name in an unusual format), highlight the text in the preview and click Mask as → [Category]. The selection is added to the entity list with a manual source marker.

If a category over-detects, turn the whole category off in Settings (see below). A per-entity “unmask” undo is not currently available.

Per-category toggles

Each category can be toggled off entirely for the session. For example, if the document already uses pseudonyms in the dates section and you don’t want them re-tokenized, toggle off Dates.

Toggle state is per-session — it resets on next launch. For persistent defaults, see environment variables.

Feedback button

If you find a miss or a false positive, click the Report button in the entity list. See Feedback.

OCR and image warnings

Low-confidence OCR and embedded images that were not OCR-checked are shown as warnings. Treat those as a manual-review queue: printed text in scanned pages is recognized locally, but signatures, logos, handwriting, and image attachments can still require human review.

What never leaves the browser

The web UI runs locally. Files are uploaded over loopback (127.0.0.1) only — no external network requests are made during processing. See What leaves your machine.