Redacting a PDF sounds simple until you actually have to do it properly. Drawing a black rectangle over a name or an SSN doesn't remove it — the text is still sitting underneath, extractable with a copy-paste or a basic PDF parser. Doing it right means the sensitive content is gone from the file, not hidden under a layer.
I've been working on PDF Redaction Studio Desktop, a native Windows (and Linux) app that handles this end to end: AI-based PII detection, a review step so you stay in control of what gets flagged, and an export that actually strips the matched content out of the PDF.
It just landed on the Microsoft Store, so here's a walkthrough of what it does and how it's put together.
Why a desktop app instead of just a web app
A web version already existed, but a chunk of users — legal, HR, healthcare, anyone handling regulated documents — didn't want to upload files anywhere to get them redacted. The desktop app runs the detection stack (OCR, PII/NER models, face detection) locally, so documents stay on the machine by default. No Docker, no backend to stand up, no account required to just open a PDF and start reviewing it.
The dashboard
Once installed, you land on a dashboard that tracks documents, completed redactions, and page usage — useful if you're processing a batch rather than a one-off file.
Opening a document
Drag a PDF in, or click to browse. Nothing happens with the file until you tell it to.
Detection runs locally
This is the part I care about most. Detection can run against a cloud API or entirely in-browser/on-device using a local model — no network round-trip for the document contents. You pick the model, tune the detection threshold (higher = fewer but more confident matches), and set how many CPU threads to throw at it.
Reviewing what got flagged
After detection, matches are grouped by category — names, SSNs, emails, phone numbers, driver's license numbers, addresses, credit cards, faces — and highlighted directly on the page. You can filter by category, add anything the model missed, and remove false positives before committing to anything.
Only after you're happy with the review does export actually strip the matched content from the PDF — text, embedded objects, and image regions covering faces or signatures all get removed rather than painted over.
What's under the hood, roughly
- OCR and layout parsing for scanned documents, not just digital-text PDFs
- A multilingual PII/NER model bundled for local, offline detection
- Face and signature detection for scanned forms and ID pages
- Custom rule support for anything the built-in categories don't cover — internal codenames, account numbers, whatever your compliance checklist needs
Where it stands today
Windows and Linux builds are out now; macOS is in progress. If you're already comfortable with command-line tools there's also a Snap package for Linux, and the desktop app shares the same detection engine as the web version, so behavior is consistent across both.
The free tier covers 100 pages a month, which is enough to actually try it against your own documents before deciding whether it fits your workflow.
If you try it and hit a document type it doesn't handle well — dense forms, unusual layouts, non-Latin scripts — I'd genuinely like to hear about it. That's the kind of edge case that's easy to miss without real-world documents to test against.
