GoCalma vs LocalRedact

GoCalma is open-source browser software for detecting and redacting sensitive information in PDF files without sending document contents to a server. It identifies more than 30 personal-information types and validates formats across more than 10 European ID formats. Users can choose redaction, masking, hashing, highlighting, or synthetic replacement by information type, then review candidates with the strategy, pattern, and confidence behind each flag. OCR for scanned or image-heavy PDFs runs locally. Listed profiles include Swiss DPA, EU GDPR, Healthcare, Financial, and Full Scan. An encrypted .gocalma recovery key can restore the original document later with the matching key. The repository says key payloads use AES-GCM encryption and redaction removes PDF author, title, and timestamp metadata. Its audit summary records a design-level concern involving raw key storage when no password is set. The Vercel demo uses regex-only detection because model files exceed its 250 MB limit. Local setup requires Node.js 18 or later, npm, and a modern desktop browser. The software is distributed under the MIT License.

LocalRedact is a free, open-source PDF redaction tool that follows plain-English instructions using deterministic rules. It removes text beneath redacted areas and blanks image pixels where those areas overlap. Detection can target personal, financial, government, health, and credential data, as well as quoted values and selected page or layout instructions. Processing happens locally: the project says it uses no LLM, cloud API, telemetry, or model download, and reads text and coordinates with PyMuPDF. By default, it removes common PDF extras such as metadata, embedded files, JavaScript, links, and saved form state. After saving, it reopens the document and checks whether detected values remain extractable. Interfaces include a Python API, two command-line tools, and an optional HTTP service. Scanned PDFs can use locally installed Tesseract; image-only pages fail closed without OCR by default. The package requires Python 3.10 or later and is available for Windows, macOS, and Linux. The project calls the software alpha and warns that detection is best effort, not a compliance certification. It recommends visually reviewing high-stakes output.