Skip to content
BranchesTags

Repository files navigation

Anonymizer

Latest release MIT License macOS Windows

Turn contracts, reports, and scans into shareable Markdown — on your machine, offline by default.

Anonymizer is a local tool for macOS and Windows: PDF / DOCX / plain text → Markdown, with optional PII redaction for English and Finnish. Personal names, emails, phones, IDs, and more become stable placeholders like [PERSON_1] — so you can collaborate, archive, or hand a document to a model without leaking the original identifiers.

  • CLI: anonymize
  • Mac app: Anonymizer.app (drag-and-drop)
  • Windows app: Anonymizer via Setup.exe (Start Menu + Apps & features uninstall)
  • Default: Offline — everything stays on your computer unless you pass --llm

Not a legal guarantee. Detection is probabilistic. Native PDF/DOCX output is best-effort (text-layer search; images/forms/comments may remain). Always spot-check high-stakes output.

Download

Platform Recommended Also
macOS Homebrew · brew install --cask anonymizer-app Latest release — Anonymizer-VERSION.zip
Windows Latest release — Anonymizer-Setup-VERSION.exe Portable Anonymizer-VERSION-windows.zip (keep runtime\ next to the exe)
CLI only macOS: brew install anonymizer Windows: enable “Add CLI to PATH” in Setup, or use the portable bin\anonymize.cmd

On the release page look for:

Asset Platform
Anonymizer-VERSION.zip macOS app (also used by Homebrew cask)
Anonymizer-Setup-VERSION.exe Windows installer (Apps & features)
Anonymizer-VERSION-windows.zip Windows portable

Release assets (from 1.2.0 onward): Mac zip, Windows Setup.exe, and Windows portable zip on the same Releases page. Older tags may be Mac-only.

Why Anonymizer

  • Privacy first — processing is local; nothing is sent over the network unless you explicitly enable a remote LLM (--llm)
  • Real office formats — PDF (including OCR for scans), Word (.docx), and text/Markdown
  • Optional source redaction — redacted .pdf / .docx matching the input type (--format source or both)
  • Modes that match the job — full scrub, identity-only, or plain text extract
  • Stable placeholders — the same person stays [PERSON_1] throughout a document
  • English + Finnish — auto language detection, patterns + neural NER + domain false-positive filters
  • Human in the loop — optional review to keep false positives in clear text
  • No hard-coded company catalogs — patterns, models, and your allow/deny lists only
  • Desktop GUIs — same options on Mac (droplet) and Windows (Setup wizard / Start Menu)

Before → after

Synthetic example only:

Before

Service Agreement between Nordic Widgets Oy and Acme Logistics Ltd.
Contact: Maija Korhonen <maija.korhonen@nordic-widgets.example.fi>, +358 50 987 6543.

After (anonymize · strict mode)

Service Agreement between [ORG_1] and [ORG_2].
Contact: [PERSON_1] <[EMAIL_1]>, [PHONE_1].

Same entity → same tag within that run. Export an optional map file if you need the reverse lookup later (treat it like the original document — it contains PII).

Install

macOS — Homebrew (recommended)

One product: Anonymizer. Terminal: anonymize. Finder: Anonymizer.app.

brew tap arcane-tl/anonymizer
brew trust arcane-tl/anonymizer    # Homebrew 6+ — once per machine
brew install --cask anonymizer-app # Anonymizer.app + CLI (recommended)
# CLI only: brew install anonymizer

anonymize doctor
anonymize --version

Upgrade (formula anonymizer + cask anonymizer-app):

brew update && brew upgrade anonymizer anonymizer-app

If you previously installed the old cask token anonymizer (same name as the formula), remove it so the CLI can link:

brew uninstall --cask --force anonymizer
# if that still errors:
rm -rf "$(brew --prefix)/Caskroom/anonymizer"
rm -rf /Applications/Anonymizer.app

brew install --cask anonymizer-app
brew link --overwrite anonymizer && hash -r

More detail: packaging/homebrew/README.md.

Windows — Setup.exe (recommended)

Download Anonymizer-Setup-*.exe from Releases.

Double-click Setup → Next → Finish
  → Start Menu: Anonymizer
  → Settings → Apps: Anonymizer (uninstall here)
  → optional: CLI on PATH as anonymize
  • Install location: %LOCALAPPDATA%\Anonymizer (per-user, no admin)
  • No separate system Python required for the Setup build
  • Unsigned builds may show SmartScreen → More info → Run anyway

Dev / from source (PowerShell): does not register in Apps & features (use Setup for that):

git clone https://github.com/arcane-tl/anonymizer.git
cd anonymizer
.\scripts\install.ps1 -Yes -FromSource
anonymize doctor
anonymize-gui

Uninstall PowerShell install: .\scripts\uninstall.ps1 -Yes
Full packaging notes: packaging/windows/README.md.

Other options

Platform How
macOS (no Homebrew) curl -fsSL https://raw.githubusercontent.com/arcane-tl/anonymizer/main/scripts/install.sh | bash -s -- --yes then optional ./packaging/macos/install-app.sh
Windows portable Anonymizer-*-windows.zip from Releases — keep runtime\ next to Anonymizer.exe
From source Python 3.11+, pip install -e ".[dev]", then spaCy lg EN+FI models (default; see docs/models.md for sm/md/sv)

60-second tour

# Health check (models, PATH, optional OCR)
anonymize doctor

# Full scrub (default) → contract.anonymized.md
anonymize contract.pdf

# Identity only — keep company names, scrub people & contact details
anonymize standard sopimus.pdf

# Text only — no redaction
anonymize extract report.pdf

# Review redactions before writing (terminal checklist)
anonymize contract.pdf --review
# ↑/↓ move · space check · enter confirm
# Document window (same as desktop GUIs): anonymize contract.pdf --review-window
# Or non-interactive: anonymize contract.pdf --reject ORG_1,PHONE_2

# Delete findings instead of [PERSON_1] tags
anonymize contract.pdf --redact-style remove

# Both: Markdown + redacted source (PDF→PDF, DOCX→DOCX)
anonymize contract.pdf --format both
# Source only (no Markdown): --format source

anonymize examples    # more copy-paste commands
anonymize --help

Desktop GUI
Mac Open Anonymizer from Applications / ~/Applications, drop files (or pick files). Needs CLI on PATH for the droplet.
Windows Start Menu Anonymizer (after Setup.exe). Choose documents → same options as Mac.

GUI options: mode / output style / output format pop-ups, Review findings before saving (default on), open when finished, Lists….

Modes

Mode Command What it does Default output
strict anonymize FILE Full scrub — people, companies, addresses, geo, URLs, plates, IDs, … FILE.anonymized.md
standard anonymize standard FILE Identity PII — person, email, phone, hetu, addresses, IBAN, cards, IP. Keeps companies, Y-tunnus, VAT, URLs, plates FILE.anonymized.md
extract anonymize extract FILE Markdown only — no redaction FILE.md

Aliases: text → extract · normal / pii → standard · scrub / full → strict.

Privacy & security

Mode Document text leaves this machine?
Default (anonymize file.pdf) No — local extract, spaCy, patterns only
--llm + ollama on localhost No
--llm + non-local ollama_url Yes — warned; blocked by --offline
--llm --llm-provider xai Yes — sent to https://api.x.ai (XAI_API_KEY)
Config use_llm: true without --llm No — CLI requires explicit --llm
--offline Blocks remote xAI and non-loopback Ollama
  • No telemetry in this application
  • --map writes placeholder → original JSON (contains PII; mode 0600 when possible)
  • Install-time network: Homebrew / Setup download / spaCy models / optional LLM
  • Native PDF/DOCX is best-effort layout redaction, not a forensic wipe

What gets redacted (strict)

Kind Placeholder
Person [PERSON_n]
Organization / brand [ORG_n]
Email / phone [EMAIL_n] / [PHONE_n]
Street, city, FI postcode [STREET_n] / [CITY_n] / [POSTAL_n]
URL [URL_n]
FI plate / hetu / Y-tunnus / VAT [PLATE_FI_n] / [FI_HETU_n] / …
IBAN, card, IP, VIN (strict) [IBAN_n], …

Dates are off by default (--include-dates to enable).
--entities … overrides the mode preset. YAML config: config.example.yaml.

How detection works (short)

Patterns (IDs, emails, legal-form companies) + heuristics + spaCy NER (EN/FI, optional SV) + domain false-positive filters (contract roles, legal collocations, form labels) + optional LLM proposals + your allow/deny lists. The app does not ship a list of real-world companies or people. Use allowlist_extra / lexicon_extra for keep-clear and FP lexicon (see config.example.yaml). Add domain IDs via YAML custom recognizers (recognizers:) — see docs/plugins.md.

Useful options

anonymize report.pdf -o clean.md          # explicit output
anonymize ./inbox/ --out-dir ./out/       # batch folder
anonymize scan.pdf --force-ocr            # scanned PDF
anonymize doc.pdf --lang fi               # force Finnish NLP
anonymize doc.pdf --map report.map.json   # sensitive reverse map
anonymize doc.pdf --config config.yaml    # allowlist / denylist
anonymize doc.pdf --llm --llm-provider ollama   # optional local LLM layer
Flag Role
-r / --review Terminal checklist: mark false positives to keep clear before write
--review-window Document review UI (used by Mac/Windows GUIs)
--reject LIST Same without a prompt (ORG_1,PHONE_2)
--redact-style placeholder (default tags) or remove (delete text). Review works with both.
--format md (default Markdown only), source (redacted original PDF or Word, same type as input), or both. PDF = black boxes + metadata scrub; DOCX = tags/delete + property scrub. Best-effort (not forensic): images, some forms/comments, wrap misses may remain. Text inputs stay Markdown-only.
--llm Opt-in LLM layer only (YAML use_llm alone is ignored). --offline blocks remote xAI and non-local Ollama URLs.
--keep-headers Keep PDF running headers/footers (default: strip)
-o - Markdown on stdout (progress stays on stderr)
--config YAML: mode, allowlist, denylist, redact_style, format, …

GUIs (Mac + Windows): mode / style / format as pop-ups, Review findings before saving (default on), open when finished. Lists… saves allow/deny to ~/.config/anonymizer/config.yaml (same on both platforms).

Troubleshooting

Message / issue Fix
No Cask with this name brew tap arcane-tl/anonymizer
Untrusted tap (Homebrew 6+) brew trust arcane-tl/anonymizer (note the final r)
“Anonymizer is damaged…” brew reinstall --cask anonymizer-app
anonymize: command not found / skipping link Uninstall old cask token: brew uninstall --cask anonymizer; then brew install --cask anonymizer-app; brew link --overwrite anonymizer && hash -r
Windows: not in Apps & features Install with Setup.exe, not only install.ps1
Windows: GUI won’t open Run anonymize-gui in a console; check %TEMP%\anonymizer-gui.log; see packaging/windows/README.md
Windows: uninstall after install.ps1 .\scripts\uninstall.ps1 -Yes (not appwiz)

Development

python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
python -m spacy download en_core_web_lg   # default quality (also fi_core_news_lg)
python -m spacy download fi_core_news_lg
pytest -q

spaCy models (switch size, add Swedish): docs/models.md.

Regression: tests/test_contract_templates.py, tests/test_realworld_precision.py, tests/test_offline_security.py.
See tests/fixtures/README.md.

Packaging:

Support this project

If Anonymizer is useful to you, you can support its development via GitHub Sponsors:

Sponsor @arcane-tl on GitHub

License

MIT — see LICENSE.

About

Local PDF/DOCX/text anonymizer (EN+FI) → Markdown or redacted originals. Offline by default. CLI + Mac app + Windows Setup.

Topics

anonymizerdocxenglishfinnishmacosmarkdownofflinepdfpiiprivacywindows

Resources

Readme
MIT license

Security policy

Security policy
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages