Remove Sensitive Data
from Any Document.
MaskBase finds personal data in your PDFs, Word files, spreadsheets, and photos — entirely on your Mac. You review every detection, then export clean copies you can share with anyone: a freelancer, a partner, ChatGPT, or Claude.
please find the Q1 financials below:
Revenue: $2,847,392
Client: Siemens AG (Contract #SK-4821)
Contact: Lisa Weber
Email: l.weber@siemens.com
Phone: +1 (555) 839-4521
Best regards,
Anna Schmidt
CFO, Berger & Partners Inc.
Redact First. Share Freely.
MaskBase strips patient names, MRNs, provider identities, and dates while preserving the clinical context — so the redacted report is still useful to whoever you share it with.
Exposed Data
UNSAFELab result report
Laboratory Report — Comprehensive Metabolic Panel
Patient Maria Gonzalez (DOB 04/12/1971, MRN 00847293). Ordered by Dr. James Okafor at St. Luke's Regional Medical Center. Collected Feb 3, 2026.
Glucose: 214 mg/dL [H]
Creatinine: 2.1 mg/dL [H]
eGFR: 34 mL/min [L]
Potassium: 5.2 mEq/L [H]
CO2: 19 mEq/L [L]
HbA1c: 9.2% [H]
Sanitized
SAFEReady to Share
Laboratory Report — Comprehensive Metabolic Panel
Patient [PATIENT_NAME] (DOB [DOB], MRN [MRN]). Ordered by [PHYSICIAN] at [FACILITY]. Collected [DATE].
Glucose: 214 mg/dL [H]
Creatinine: 2.1 mg/dL [H]
eGFR: 34 mL/min [L]
Potassium: 5.2 mEq/L [H]
CO2: 19 mEq/L [L]
HbA1c: 9.2% [H]
Example: Ask an AI About the Redacted Report
Clinically useful output — zero patient data exposed
Ask the model
Which values are outside the normal range, and what could they indicate?
Useful output
Five values are flagged: Glucose (214 mg/dL) and HbA1c (9.2%) are significantly elevated, pointing to poor glycemic control. Creatinine (2.1 mg/dL) is high while eGFR (34 mL/min) is critically low, suggesting reduced kidney function. Low CO2 (19 mEq/L) alongside elevated potassium may indicate early metabolic acidosis. These results warrant prompt clinical review.
See It in Action
Watch MaskBase scan a document, walk you through the detections, and export a clean copy
Sharing Documents Shouldn't Mean Sharing Secrets
Every document you send out — to an AI tool, a contractor, or a partner — can expose personal data
Feeding Documents to AI?
Contracts, reports, and datasets you hand to ChatGPT or Claude are full of names, IDs, and figures that shouldn't leave your machine.
Outsourcing Work?
Freelancers, agencies, and partners need your documents' content — not your clients' identities.
Manual Redaction?
Scrubbing PDFs, spreadsheets, and scans by hand takes hours — and one missed name is all it takes.
Scan. Review. Export.
Three steps from sensitive documents to clean, shareable copies
Add Documents
Batch & Local
Drop in PDFs, Word files, spreadsheets — even photos or scans of documents. OCR reads them on-device. Nothing is uploaded.
Review Detections
You're in the Loop
A local language model highlights names, IDs, and other personal data in every file. You approve each one, fix false positives, and add anything it missed.
Export Clean Copies
Share Anywhere
Save the redacted files wherever you want and share them with anyone — a freelancer, a partner, or an AI tool. The originals never leave your Mac.
Built for Trust & Control
A privacy tool should have nothing to hide — so this one is open source
Open Source (MIT)
The full source code is on GitHub. Anyone can audit exactly what the app does with your data — a privacy tool you don't have to take on faith.
100% Local, Zero Telemetry
Detection runs on-device with a small language model. No accounts, no analytics, no crash reporting — the code contains zero calls to any vendor server.
Human-in-the-Loop Review
No detector is perfect, so nothing is final until you say so. Review every file, toggle false positives off, and mark anything the model missed.
OCR for Photos & Scans
Snap a photo of a document or drop in a scanned PDF — on-device OCR reads the text and finds the personal data inside it.
Batch Processing
Redact a whole folder in one pass: PDF, DOCX, CSV, XLSX, and images. Review each file, then export all the clean copies at once.
Choose Your Model
Ships with a multilingual PII model (GLiNER), with custom entity types on the fly. Swap in any Hugging Face model — or run pattern matching only.
Frequently Asked Questions
Everything you need to know about MaskBase.
Open Source. On Your Device.
Free Forever.
MaskBase is free and MIT-licensed. No accounts, no license keys, no telemetry — just download it and start redacting.