Architecture & Standards

Security & Privacy Architecture

DocuNexa replaces risky cloud-upload workflows with client-side cryptography and in-browser WebAssembly execution.

In-Memory Volatile Processing

Files are read into local browser RAM (ArrayBuffer/Uint8Array) via standard Web APIs. No temporary files or caches are created on remote servers.

Standard AES-256 PDF Encryption

Password protection implements standard AES-256 and AES-128 algorithms via WebAssembly (@pdfsmaller/pdf-encrypt). Passwords are never sent over the network.

MIME & Header Byte Validation

All file drops undergo strict binary header inspection (e.g. verifying "%PDF-" magic bytes) to prevent corrupt files or disguised executables from loading.

Automatic Filename Sanitization

Output filenames automatically strip illegal filesystem characters (such as "/", "\", ":", "*", "?", "<", ">", "|") to protect against path traversal.

PDF Password Protection & Cryptography

When protecting a PDF document using our Protect PDF tool, DocuNexa applies the standard PDF encryption specification (ISO 32000) using client-side WebAssembly engines compiled from audited open-source libraries (@pdfsmaller/pdf-encrypt).

Both user passwords (required to open and view the document) and owner passwords (used to manage modification permissions) are processed in local memory. The password string and the resulting encrypted document never traverse the network.

Similarly, our Unlock PDF tool uses @pdfsmaller/pdf-decrypt to verify the provided password and remove standard encryption dictionaries directly inside the browser session.

Verifiable Technical Boundaries

We believe transparent engineering builds true trust. We explicitly document where client-side browser tools differ from enterprise desktop forensic software:

1. Sign PDF: Visual Appearance Stamping

The Sign PDF tool allows users to draw, type, or upload a visual signature image and stamp it onto document pages. It is an appearance-placement utility and does not provide cryptographic digital signatures backed by X.509 PKI certificate authorities or cryptographic tamper-evident hashes.

2. Redact PDF: Visual Blackout Rectangles

The Redact PDF tool draws opaque, permanent vector blackout rectangles directly into the PDF content stream over user-selected coordinates. While this completely obscures the text visually from screen rendering and printing, it does not strip low-level binary text streams embedded in raw font dictionaries. For classified government or medical redaction, specialized forensic sanitizers should be used.

3. PDF/A Metadata Preparation

The PDF to PDF/A tool injects ISO 19005-1 (PDF/A-1b) metadata markers and color profiles. It is marked as experimental because full pre-press compliance validation requires desktop preflight verification software.

4. Local Document Summarization

The Document Summarizer uses deterministic rule-based algorithms to analyze headings, paragraph boundaries, and key sentences directly in your browser. It does not send your document data to external cloud LLM providers.

5. OCR Language Data Retrieval vs. Document Processing

Optical character recognition (OCR PDF) analyzes document pages locally in your browser using a Tesseract.js WebAssembly worker. To execute recognition, the engine retrieves required public language model files (such as English or Amharic .traineddata.gz) on-demand from the public jsDelivr CDN repository into local browser cache/IndexedDB. Document pages, rendered canvas pixels, and extracted text are processed locally in browser memory without cloud server transmission.

External Network Resources & Boundaries

DocuNexa strictly separates user document processing from external resource downloads:

A. Document Processing (Local Only):

Document bytes remain entirely in browser memory (ArrayBuffer/Uint8Array). All operations—including parsing, unit detection, page extraction, password encryption, and optical character analysis—execute locally inside your browser or Web Worker. Your documents are never uploaded to remote servers for processing.

B. On-Demand OCR Language Data Retrieval:

When using OCR PDF, Tesseract.js may retrieve required public language trained data models (such as English or Amharic .traineddata.gz) from standard open-source CDN repositories (jsdelivr) on-demand when a language is first selected. The downloaded language file is cached locally in your browser so it is not re-downloaded repeatedly. This download is strictly a retrieval of public optical character recognition model weights; your PDF pages or images are NEVER transmitted externally.

C. Other External Resources:

  • Google Fonts: Typography stylesheets and font files (Inter and JetBrains Mono) loaded from Google's font CDN (fonts.googleapis.com and fonts.gstatic.com).
  • Static Website Delivery: Standard web hosting delivery (e.g. edge CDN) to serve the static HTML, JavaScript, and CSS bundle of DocuNexa.

Recommended Security Practices for Users

  • Always verify the SSL lock in your address bar: https://docunexa.pro.et.
  • Use strong, unique passwords when protecting sensitive documents.
  • Close your browser tab when finished to immediately free document buffers from device RAM.
Security & Privacy Architecture — DocuNexa | DocuNexa