docx-fast-node

High-performance Word (.docx) parser with a native Rust core.

Features

• Complete: Extracts paragraphs, tables, headers, footers, footnotes, endnotes, comments
• Styled: Paragraph and character styles with inheritance
• Lists: Numbering support (bullets and numbered lists)
• Track Changes: Revision tracking support
• Comments: Threaded comments with author information
• TOC: Automatic table of contents generation from headings
• Images: Extract embedded images
• Arrow IPC: Export to Apache Arrow format for data processing
• Fast: Memory-mapped files and streaming decompression
• Safe: Rust's memory safety guarantees

Installation

npm install docx-fast-node

Quick Start

const { parseDocx } = require('docx-fast-node');

const doc = parseDocx('/path/to/document.docx');

// Metadata
console.log('Title:', doc.metadata.core.title);
console.log('Author:', doc.metadata.core.creator);
console.log('Pages:', doc.page_count);

// Table of contents
for (const entry of doc.toc) {
  console.log('  '.repeat(entry.level) + entry.text);
}

// Document content
for (const block of doc.blocks) {
  if (block.type === 'paragraph') {
    console.log(`[${block.style_name || 'Normal'}] ${block.text}`);
  } else if (block.type === 'table') {
    for (const row of block.rows) {
      const cells = row.cells.map(cell => {
        // Extract text from cell blocks
        return cell.blocks
          .filter(b => b.type === 'paragraph')
          .map(p => p.text)
          .join(' ');
      });
      console.log(cells.join(' | '));
    }
  }
}

Options

const doc = parseDocx('/path/to/document.docx', {
  maxInflate: 128 * 1024 * 1024,  // Max decompressed size in bytes (default: 128 MiB)
});

NDJSON Streaming

const { parseDocxNdjson } = require('docx-fast-node');

parseDocxNdjson('/path/to/document.docx', '/tmp/out.ndjson');

Output format (one JSON object per line):

{"kind":"metadata","metadata":{"core":{"title":"My Doc"}},"settings":{},"toc":[]}
{"kind":"style","style":{"id":"Heading1","name":"Heading 1","is_default":false}}
{"kind":"block","block":{"type":"paragraph","text":"Hello World"}}

Arrow IPC Export

Export document tables to Apache Arrow format:

const { parseDocxArrowDataset } = require('docx-fast-node');

// Export all tables to Arrow IPC files
const files = parseDocxArrowDataset('/path/to/doc.docx', '/tmp/arrow-output');
console.log('Created:', files);

Image Extraction

const { extractDocxImages } = require('docx-fast-node');

const imagePaths = extractDocxImages('/path/to/document.docx', './extracted-images');
console.log('Extracted images:', imagePaths);

TypeScript Support

TypeScript definitions are included:

import { parseDocx, DocxDocument, Paragraph, Table } from 'docx-fast-node';

const doc: DocxDocument = parseDocx('document.docx');

Document Structure

DocxDocument
├── metadata (core, app, custom properties)
├── settings (document settings)
├── page_count
├── toc (auto-generated table of contents)
├── styles (paragraph, character, table styles)
├── numbering (list definitions)
├── blocks (paragraphs and tables)
│   ├── Paragraph { runs: [Run { text, bold, italic, ... }] }
│   └── Table { rows: [Row { cells: [Cell { blocks }] }] }
├── headers / footers
├── footnotes / endnotes
├── comments (with thread support)
├── revisions (track changes)
├── fields (TOC, page numbers, etc.)
└── images (embedded references)

Performance

• 10x faster than python-docx
• Low memory footprint: Streaming decompression for large files
• Parallel-ready: Stateless parsing allows concurrent processing

Platform Support

Prebuilt binaries available for:

• macOS (Intel & Apple Silicon)
• Linux (x64, ARM64)
• Windows (x64)

License

MIT

Related Packages

• xlsx-fast-node - Excel (.xlsx) parser
• pptx-fast-node - PowerPoint (.pptx) parser