officepress

• Text documents: styles, lists, tables (merged and nested cells, repeated header rows), sections with their own page size and headers, footnotes and endnotes, columns, tab stops, fields (table of contents with page numbers, PAGE, REF, SEQ, DATE), equations, comments and tracked changes on request, hyperlinks and PDF bookmarks.
• Spreadsheets: number and date formats, conditional formatting (colour scales, data bars, icon sets), charts and pictures at their anchor, print areas, fit to page, repeated titles, page breaks, headers and footers. Formulas are not recalculated: the values saved in the file are printed.
• Presentations: masters and layouts, theme colours, gradients, pictures, tables, charts, SmartArt from the drawing stored in the file, speaker notes and hidden slides on request.
• Visio: inheritance from masters and themes, geometry (arcs, splines, NURBS), connectors with arrowheads, fill patterns and gradients, scaled drawings, several pages.
• Publisher: text frames and linked stories, master pages, pictures, shapes, tables, hyphenation in eight languages.
• Fonts: fonts installed on the server are matched by name. When Arial, Times New Roman, Courier New or Calibri are missing, bundled metric-compatible fonts keep the line breaks. Bundled Noto fonts cover Greek, Cyrillic, Hebrew, Arabic, Devanagari, Thai, Chinese, Japanese and Korean.

• PDF output: PDF 1.4; PDF/A-2b, 2u, 2a, 3b, 3u and 3a, checked with veraPDF 1.30.2; tagged PDF; PDF/UA-1 for Word files. The same input gives the same PDF, byte for byte.

• At run time the library uses only modules built into Node.js: zlib, fs, crypto, path, os, url and worker_threads. No npm package is installed with it.
• No network access, no child processes, no eval or new Function, no native add-ons.
• No external calls: the library contacts neither ToupieSoft nor anyone else.
• Macros are never run. Macro-enabled files are converted without them.
• Corrupt or unsupported input raises a typed error. A damaged legacy file falls back to its text.
• convertToPdfAsync runs in a worker thread and accepts a time limit (timeoutMs).

Measured on 14 October 2026 with officepress c98a0ac: each document converted three times in one Node.js process; the median of runs 2 and 3 is reported. Laptop with AMD Ryzen 9 5900HS (8 cores), 32 GB RAM, Windows 10, Node.js 24.15. A conversion uses one core.

Large files need time and memory: a 102 MB Publisher file takes 5.5 s and adds 670 MB to the process memory. For a server, run conversions in a pool of worker threads and limit the upload size.

Every version is compared page by page with the PDF printed by the application that made the document: Word, Excel, PowerPoint and Publisher 2021, Visio 2019, LibreOffice for OpenDocument. Each difference found is recorded with the file, the page and an image, and a correction counts only when the document itself prints correctly.

The examples come from this test bed. Download the source file and both PDFs, and compare.

• Formulas are not recalculated.
• The PDF output is not encrypted and not signed.
• WordArt warps and 3-D effects are rendered only in part.
• Scripts other than Latin, Greek, Cyrillic, Hebrew, Arabic, Devanagari, Thai and CJK are drawn without full shaping.
• When a font is missing on the server, line breaks can differ slightly from Office.

The complete list is in the documentation.

import { convertToPdf } from '@toupiesoft/officepress';

app.get('/documents/:id/preview.pdf', async (req, res) => {
  const file = await storage.read(req.params.id);       // your storage
  res.type('application/pdf').send(convertToPdf(file));  // in-process, offline
});

• In process: convertToPdf is synchronous; suitable for scripts, batch jobs and low traffic.
• Worker pool: for a preview service, a fixed pool of worker_threads behind a queue, sized at the number of cores minus one. See Document previews.