GitHub - monjurulkarim/privateredact: Redact PII from documents - 100% offline, local LLM, no cloud. · GitHub
/" data-turbo-transient="true" />
Skip to content
Type / to search
Sign in<br>Sign upAppearance settings
You signed in with another tab or window. Reload to refresh your session.<br>You signed out in another tab or window. Reload to refresh your session.<br>You switched accounts on another tab or window. Reload to refresh your session.
Dismiss alert
{{ message }}
monjurulkarim
privateredact
Public
Notifications<br>You must be signed in to change notification settings
Fork
Star
main
BranchesTags
Go to file
CodeOpen more actions menu
Folders and files<br>NameNameLast commit message<br>Last commit date<br>Latest commit
History<br>3 Commits<br>3 Commits
docs
docs
LICENSE
LICENSE
README.md
README.md
View all files
Repository files navigation
created<br>2026-08-09T12:32
updated<br>2026-08-09T12:45
PrivateRedact
Redact personally identifiable information from documents — 100% offline, no cloud.
A local-first desktop app that finds PII with a hybrid regex + local-LLM pipeline, lets you review every redaction, and exports a securely redacted file. No account. No upload. Nothing leaves your machine.
Why
Most redaction tools fail in one of two ways:
Cloud redaction requires uploading the exact document you're trying to protect. Solving a privacy problem by first handing the file to a server.
Naive desktop tools either make you black out text by hand (slow, error-prone) or auto-black entire lines (unreadable). And because they black the rendering rather than the text, the layer underneath can often be selected, copied, or removed to reveal what was "redacted."
PrivateRedact is the third option: accurate, automatic, value-only redaction that runs entirely on your machine. It redacts the value (555-1234), not the label (Phone:), and never the whole line.
Features
Value-only spans — redacts just the sensitive value, not the label or surrounding text.
Proportional boxes — no whole-line blacking, even on wide text runs.
Reads anything — text PDFs, scanned/image PDFs (OCR), DOCX, TXT, and images.
Hybrid detection — regex for known formats (SSN, cards, IBAN, DOB) + a local LLM for semantic PII (diagnoses, names, addresses, codenames).
OCR + optional vision model — Tesseract under the hood, with a vision model double-check on low-quality scans.
Secure rasterized PDF export — pages are rasterized and rebuilt, so there's no recoverable text layer underneath. Can't be "un-redacted."
Format-preserving DOCX — edits the OOXML in place; all Word formatting is kept.
Post-export leak scan — re-OCRs the output and reports exactly what (if anything) survived. Never claims "clean" on a failed scan.
Metadata stripping — PDF Info/XMP, DOCX core/app props, image EXIF (default on).
Compliance presets — HIPAA · PCI DSS · GDPR · HR · Employment · Legal · Financial · Vendor · Names-Only.
Encrypted PDF I/O — password-protected inputs (you're prompted) + optional encrypted output.
Signed + notarized macOS build — opens on customer Macs with no Gatekeeper warnings.
Free trial vs. license
PrivateRedact is free to load, detect, and review. You can run your real documents through the entire pipeline — extraction, detection, and the review overlay — at no cost. This is the trial.
Exporting a redacted document requires a license. When you click export, the app prompts for a Gumroad license key. One-time purchase, use forever — no subscription.
Download (free trial): GitHub Releases
Get a license: Gumroad
Demo
A full walkthrough — load, auto-detect, review, export, and the leak-scan proof:
https://youtu.be/XtSbOfy0b18
Install
Go to Releases and download the build for your OS:
macOS (Apple Silicon): PrivateRedact--mac-arm64.dmg — signed + notarized.
Windows (x64): PrivateRedact--win-x64.exe — currently unsigned (see troubleshooting).
Open the installer and drag PrivateRedact to your Applications (macOS) or run the installer (Windows).
Launch it. On first use, it offers a one-click install of the local AI engine (Ollama) — click-gated, never silent. Everything runs offline afterwards.
Intel Macs / Linux: not yet supported (Apple Silicon only for now). See roadmap.
System requirements
macOS (Apple Silicon) or Windows (x64).
~8 GB RAM minimum for the smallest useful model; more is better. The app picks the model size automatically based on available memory.
No runtime dependencies. PDF rendering runs in-process; Ghostscript is not required. Regex-only mode works without any model.
Roadmap
Honest about what's next:
Signed Windows builds (currently unsigned — SmartScreen warns on first launch).
Intel/x64 macOS + Linux AppImage targets.
Prompt-injection mitigation in the LLM layer (today, document text reaches the model verbatim — fine for your own documents; hardening is required before e-discovery / adversarial third-party docs).
Keep-text-layer export as default, Ollama pull pause/resume , multi-language...