Local PDF to Markdown Converter

A blazing-fast, privacy-first static web application that converts PDF documents to Markdown entirely within the browser.

No file ever leaves your machine. There is no server-side processing, no telemetry, and zero data exfiltration.

Key Features

• 100% Client-Side Processing: Your PDFs and data stay on your device.
• Lazy Local OCR: Automatically detects scanned/image-only PDFs and runs OCR via ONNX Runtime Web only when needed.
• Pages-Compatible Assets: Keeps the default OCR path small and splits the optional accurate detector into deployable static chunks.
• Intelligent Heuristics: Analyzes font sizes and text positions to reconstruct appropriate Markdown headers, paragraphs, and lists.
• Performance Focused: Built with Vanilla JavaScript, pure CSS, and Vite. All heavy lifting happens in Web Workers to keep the UI perfectly responsive.

Tech Stack

• Build Tool: Vite
• PDF Parsing & Rasterization: pdfjs-dist
• Machine Learning / OCR: onnxruntime-web
• Deployment: Cloudflare Pages (Static Site)

How It Works

The application intelligently selects a processing pipeline based on the PDF content:

1. Text Extraction Pipeline: For PDFs with a native text layer. Uses pdfjs-dist to extract text layout and applies heuristics to generate structured Markdown.
2. OCR Pipeline: For scanned or image-only PDFs. Rasterizes pages to an OffscreenCanvas in a Web Worker, then lazily loads local PaddleOCR ONNX assets to detect English text boxes and recognize text.

All heavy tasks (PDF rasterization, extraction loops, and ONNX inference) run exclusively in Web Workers.

Security & Privacy

This application is designed with strict security boundaries:

• Data Privacy: No data is uploaded anywhere. No external APIs are called for processing.
• Hardened Headers: Implements Cross-Origin-Opener-Policy: same-origin and Cross-Origin-Embedder-Policy: require-corp to enable secure SharedArrayBuffer usage for multi-threaded WASM execution.
• Strict CSP: Enforces a strict Content Security Policy allowing only local scripts, workers, and secure WebAssembly execution.

Getting Started

Prerequisites

• Node.js (v18+)
• npm

Installation

1. Clone the repository and navigate to the project directory:

   git clone <repository-url>
   cd converter

2. Install dependencies:

   npm install

Development

Start the Vite development server:

npm run dev

Run the pure logic tests:

npm test

Build & Deployment

For Cloudflare Pages Git deployments, use these build settings:

• Build command: npm run build
• Build output directory: dist
• Deploy command: leave blank if Pages is auto-publishing the build output, or use npm run deploy if your Cloudflare build settings require an explicit deploy command.

Do not use npx wrangler deploy for this project. That command deploys a Worker and will fail with "Missing entry-point to Worker script or to assets directory" because this repository is a static Cloudflare Pages app.

Build the production static assets:

npm run build

The optimized bundle will be placed in the dist/ directory, ready to be deployed to Cloudflare Pages:

npm run deploy

The build removes the unchunked server detector from dist/ and serves the optional accurate detector as sub-25 MiB chunks for Cloudflare Pages.

Out of Scope

• Server-side processing or backend APIs.
• LocalStorage caching of uploaded documents.
• Table-to-Markdown conversion (treat as a future enhancement).