document-svg

See every page of a PDF, Word, Excel, PowerPoint, diagram or CAD file as an SVG image — on your own machine, without the app that made it.

日本語 · 简体中文 · Project site · Samples · Releases

document-svg turns documents, diagrams and technical drawings into ordinary SVG images, one per page. Any browser can show them, they stay sharp when you zoom in, and they drop into a web page like any other picture. Alongside the pages you get a short report that says plainly which parts could not be reproduced exactly.

It is a converter you build into something else — a website, an internal tool, a review process or an AI assistant — and it comes as a Node.js package, a Python package, a Rust library and a command-line tool (docsvg).

Every image above is an unretouched conversion of a sample file in this repository.

Why use it

• Your files stay with you. Conversion runs on your computer or your own server. Nothing is uploaded, there is no account, and nothing is tracked, so contracts and design drawings that must not leave the company can still get a preview.
• Anyone can look. The people reviewing the result don't need Office, AutoCAD or a viewer licence — a browser is enough.
• It tells you what it missed. No converter reproduces everything. Instead of quietly dropping a chart or a font, document-svg lists what it skipped or approximated, page by page, so you can decide whether the result is good enough to share.
• Opening a file never runs it. Macros, scripts, form actions and links to outside resources are shown or skipped, never executed. Size and page limits stop a huge or broken file from exhausting memory.
• One tool instead of many. Office documents, PDFs, diagrams, CAD and 3D models, and scientific and business data all go through the same call and come out the same way.
• And the way back. SVG pages can be packaged back into PowerPoint, Word, Excel, draw.io or CAD files.

What people use it for

• Seeing document changes in code review. A GitHub Action converts the slides, spreadsheets and PDFs changed in a pull request, instead of leaving reviewers with "binary file changed".
• Previewing uploads in their own app, without installing Office on the server or paying for a viewer SDK.
• Keeping diagrams in step with the code by rendering draw.io, Mermaid, PlantUML, D2 or Graphviz files every time the documentation is built.
• Checking drawings and 3D models without CAD — floor plans, circuit boards, machining paths and models — from a web page.
• Letting an AI assistant read documents through one dependable step instead of screenshots.

Each example is explained in more detail on the project site.

Install

Pick the one that matches where you will call it from. You only need one.

No Rust is needed for the npm and pip packages: they come ready-made for Windows, macOS and Linux. If you want the command without installing Rust, download the file for your system from Releases and put docsvg on your PATH.

Where to get it and where it goes

To check what is installed and where: npm ls document-svg, python -m pip show document-svg, or docsvg --version. The Rust API reference is on docs.rs. An authenticated npm mirror on GitHub Packages is described in docs/PUBLISHING.md.

Try it

docsvg slides.pptx --output out/slides

The output folder must be new or empty. You get one SVG per page and one report of the conversion:

out/slides/
├── page-0001.svg
├── page-0002.svg
└── conversion.json

Open a page in any browser. The report lists warnings — things that were approximated or left out, such as a substituted font or an unsupported fill. Finishing without an error is not the same as perfect: if there are warnings, have a person look at the pages before you share them.

A "page" means what you would expect for each kind of file: a PDF page, a slide, a printed page of a Word document, a printed page of a worksheet, or one page of a draw.io diagram.

From Node.js / TypeScript

const { convert, preview } = require('document-svg')

async function main() {
  // Write SVG files to a folder
  const report = await convert('slides.pptx', 'out/slides')
  console.log(report.pageCount, report.warnings)

  // Or keep the pages in memory, for example to send them to a browser
  const result = await preview('slides.pptx')
  const firstPageSvg = result.pages[0].svg
  console.log('needs review:', result.needsReview)
}

main().catch(console.error)

Conversion runs in Node.js (a server or Electron's main process), off the main thread. In the browser, show each page as an image with the small document-svg/preview-ui helper, which does not load any native code. See the Node.js guide.

From Python

from document_svg import convert, preview

report = convert("report.docx", "out/report")
print(report["page_count"], report["warnings"])

result = preview("report.docx")          # pages in memory, no files written
first_page_svg = result["pages"][0]["svg"]

See the Python guide.

From Rust

use document_svg::{convert_path, ConvertOptions};

fn main() -> Result<(), document_svg::Error> {
    let report = convert_path("report.pdf", "out/report", &ConvertOptions::default())?;
    println!("{} pages, {} warnings", report.page_count, report.warnings.len());
    Ok(())
}

API documentation is on docs.rs.

Only in the browser

A separate WebAssembly build converts PDF, Word, Excel and PowerPoint files entirely inside the browser, without a server and without uploading the file. It is not published as a package; you build it yourself from bindings/wasm.

Show the pages in your app

Display each page as an image (for example an <img> element). Don't paste the SVG markup straight into your page's HTML: the included display helpers do basic checks, but they are not a filter that makes any SVG harmless. Text in an image cannot be selected or searched.

Text uses the fonts available where the SVG is viewed, so it can look slightly different on another computer. For PDFs, you can turn embedded text into outlines when exact appearance matters more than selectable text.

Turn SVG back into Office or CAD files

docsvg reverse out/slides --output slides.pptx
docsvg reverse out/report --output report.docx
docsvg reverse out/drawing --output drawing.dxf

The file extension you choose decides the format: PowerPoint, Word, Excel, PDF, draw.io, DXF, G-code, STL and more. What comes back is how the pages look — paragraphs, cells, formulas and charts are not rebuilt.

draw.io diagrams are the exception: convert with --embed-drawio-source, and reverse restores the original, editable diagram.

Supported formats

Some formats are drawn in full; for others, such as DWG or Access databases, you get a clear summary of what is inside rather than the drawing. Each format is rated A (drawn from the file's own contents), B (traced from pixels) or C (structure such as tables inferred from the drawing).

• Searchable format list — look up your file type
• Technical format matrix — every extension with its fidelity notes
• Supported features and limitations

Use it from AI assistants and GitHub

The installers use a checksum-verified release build when Rust is not installed. To preview the documents a pull request changes, add this workflow to your repository:

name: Document preview
on:
  pull_request:
    paths: ['**.pdf', '**.pptx', '**.xlsx', '**.docx']
jobs:
  preview:
    uses: ryusui-hiro/document-svg/.github/workflows/document-preview.yml@main
    permissions:
      contents: read
      pull-requests: write

Each changed document is converted into pages you can download from the run, and a comment lists how many pages and warnings each one has. More in the GitHub and agent integration guide and the guide for AI agents.

Safety and limits

• Files are never uploaded, and macros, scripts and external links are never run or fetched.
• A PDF that needs a password to open is refused, not cracked. A PDF whose only restriction is an owner password (such as "no printing") opens as it would in any viewer; those flags are not enforced.
• Input size, archive contents and page count have limits. Don't raise them just to push a difficult file through.
• Existing files are never overwritten.
• If strangers can upload files to your service, run conversion in a separate, restricted process with memory and time limits.
• Warnings are a signal to review, not a certificate of accuracy or safety. Medical images and reports are not anonymized.

Read Safety and limits and the security policy before using it on a public service.

Platform support and troubleshooting

• If pip starts building from source, there is no ready-made package for your Python and system; update pip first. Building needs Rust and a C linker.
• If the npm package installs but won't load, check that optional dependencies were installed and that Node.js matches your machine's CPU. Copying node_modules to another system does not work.
• The SVG pages display in current browsers, resvg, and librsvg 2.46 or newer.

Drawing cloud architecture diagrams

Separately from conversion, the repository includes a tool that builds architecture diagrams with the official Azure, AWS and Google Cloud icons. See the cloud architecture guide and the business diagram templates.

Contributing

cargo test --workspace --locked

See CONTRIBUTING.md for the full checks, and docs/ARCHITECTURE.md for how the converter is put together. Release steps are in docs/PUBLISHING.md and changes in the changelog.

License

MIT OR Apache-2.0, at your option. When you redistribute the software, keep the copyright and license notices required by the license you choose. See LICENSE, LICENSE-MIT, LICENSE-APACHE and the bundled third-party notices.