Cloud-accessible vs cloud-processed: an important distinction
Most "cloud-based" redaction tools require you to upload your documents to their servers for processing. Your files travel across the internet, are stored on their infrastructure, and are processed by their systems. For sensitive documents — DSAR responses, medical records, legal files, financial data — this creates a data processing chain with significant compliance implications.
SafeRedact takes a different approach. The application loads in your browser from the web (cloud-accessible), but all document processing happens locally on your device. The PDF is rendered in your browser. Text is extracted locally. Only the extracted text coordinates are sent to the AI API for PII classification, and Anthropic's zero-retention policy means even that data is not stored.
The result: you get the convenience of a cloud tool (no install, accessible from any computer, always up to date) with the privacy guarantees of local processing.
Why this matters for compliance
When you upload a document to a cloud redaction service, you're creating a new data processing relationship. Under GDPR, this typically requires a Data Processing Agreement with the service provider, documentation in your Record of Processing Activities, potentially a Data Protection Impact Assessment, and ensuring the data transfer has a lawful basis.
For DSAR documents specifically, this is particularly problematic. The documents you're redacting contain other people's personal data — the entire reason you're redacting is to protect those third parties. Uploading their data to yet another service to achieve that protection is a contradiction that some Data Protection Officers flag as a compliance concern.
SafeRedact's browser-based architecture eliminates this entire category of risk. No upload means no additional data processing relationship, no DPA required, no ROPA entry, no DPIA consideration for the redaction tool itself.
How browser-based redaction works
1. Load the app: Navigate to saferedact.app. The application JavaScript loads in your browser. No plugins, extensions, or desktop software required.
2. Open your document: Select a PDF from your device. The file is read into browser memory using the File API — it stays on your machine.
3. AI detection: Text is extracted from the PDF locally using PDF.js (or Tesseract.js OCR for scans). The text with position coordinates is sent to Anthropic's Claude API for PII classification. The API returns detection results; the original file was never transmitted.
4. Review and redact: Detections appear as highlighted overlays on your document. Approve, dismiss, or add manual redactions. Click Apply to generate a permanently redacted output using pixel-burn rendering.
5. Download: The clean output file downloads to your device. No copy exists on any server.
Compared to traditional cloud redaction
Redactable: Cloud-processed. Documents are uploaded to and stored on their servers. Accounts required. Team collaboration features depend on cloud storage. Data processing relationship created with each document.
iDox.ai: Cloud or on-premise deployment options. Cloud version requires document upload. On-premise version keeps data local but requires infrastructure.
Adobe Acrobat (online): Adobe's online tools process files on Adobe's servers. The desktop app processes locally but requires installation and lacks AI detection.
SafeRedact: Browser-accessed, locally processed. No upload, no storage, no data processing relationship. AI-powered detection via zero-retention API.
Frequently asked questions
Do I need to install anything?
No. SafeRedact runs entirely in your web browser. Works on Chrome, Firefox, Edge, and Safari on any operating system.
Where are my documents processed?
On your device, in your browser. The original PDF file never leaves your machine. Only extracted text is sent to the AI API for PII classification, under zero data retention terms.
Can I use SafeRedact on a phone or tablet?
SafeRedact works on tablet browsers. For the best experience with large documents, a desktop or laptop browser is recommended.
Is there an offline mode?
The AI detection step requires an internet connection to reach the classification API. Manual redaction (drawing boxes) can work offline once the application is loaded, but AI-powered PII detection needs connectivity.