High-performance HTML to Markdown converter with full GitHub Flavored Markdown support. Written in Rust, available for Node.js and as a native Rust crate.

• Fast - Written in Rust with O(n) algorithms, significantly faster than JavaScript alternatives
• Full GFM Support - Tables with alignment, strikethrough, autolinks, fenced code blocks
• Accurate - Handles malformed HTML gracefully via html5ever
• Configurable - Multiple heading styles, link styles, custom selectors
• Zero Dependencies - Single native binary, no JavaScript runtime overhead
• Cross-Platform - Pre-built binaries for Windows, macOS, and Linux (x64 & ARM64)
• TypeScript Ready - Full type definitions included
• Async Support - Non-blocking conversion for large documents

npm install @vakra-dev/supermarkdown

cargo add supermarkdown

Install the CLI binary via cargo:

cargo install supermarkdown-cli

The CLI allows you to convert HTML files from the command line or via stdin:

# Convert a file
supermarkdown page.html > page.md

# Pipe HTML from curl
curl -s https://example.com | supermarkdown

# Exclude navigation and ads
supermarkdown --exclude "nav,.ad,#sidebar" page.html

# Use setext-style headings and referenced links
supermarkdown --heading-style setext --link-style referenced page.html

import { convert } from "@vakra-dev/supermarkdown";

const html = `
  <h1>Hello World</h1>
  <p>This is a <strong>test</strong> with a <a href="https://example.com">link</a>.</p>
`;

const markdown = convert(html);
console.log(markdown);
// # Hello World
//
// This is a **test** with a [link](https://example.com).

When scraping websites, HTML often contains navigation, ads, and other non-content elements. Use selectors to extract only what you need:

import { convert } from "@vakra-dev/supermarkdown";

// Raw HTML from a web scrape
const scrapedHtml = await fetchPage("https://example.com/article");

// Clean conversion - remove nav, ads, sidebars
const markdown = convert(scrapedHtml, {
  excludeSelectors: [
    "nav",
    "header",
    "footer",
    ".sidebar",
    ".advertisement",
    ".cookie-banner",
    ".social-share",
    ".comments",
    "script",
    "style",
  ],
});

When feeding web content to LLMs, you want clean, focused text without HTML artifacts:

import { convert } from "@vakra-dev/supermarkdown";

// Extract just the article content for RAG pipelines
const markdown = convert(html, {
  excludeSelectors: [
    "nav",
    "header",
    "footer",
    "aside",
    ".related-posts",
    ".author-bio",
  ],
  includeSelectors: ["article", ".post-content", "main"],
});

// Now feed to your LLM
const response = await llm.chat({
  messages: [
    {
      role: "user",
      content: `Summarize this article:\n\n${markdown}`,
    },
  ],
});

Convert blog HTML while preserving code blocks and formatting:

import { convert } from "@vakra-dev/supermarkdown";

const blogHtml = `
<article>
  <h1>Getting Started with Rust</h1>
  <p>Rust is a systems programming language focused on safety.</p>
  <pre><code class="language-rust">fn main() {
    println!("Hello, world!");
}</code></pre>
  <p>The <code>println!</code> macro prints to stdout.</p>
</article>
`;

const markdown = convert(blogHtml);
// Output:
// # Getting Started with Rust
//
// Rust is a systems programming language focused on safety.
//
// ```rust
// fn main() {
//     println!("Hello, world!");
// }
// ```
//
// The `println!` macro prints to stdout.

Handle tables, definition lists, and nested structures common in docs:

import { convert } from "@vakra-dev/supermarkdown";

const docsHtml = `
<h2>API Reference</h2>
<table>
  <tr><th>Method</th><th>Description</th></tr>
  <tr><td><code>convert()</code></td><td>Sync conversion</td></tr>
  <tr><td><code>convertAsync()</code></td><td>Async conversion</td></tr>
</table>
<dl>
  <dt>headingStyle</dt>
  <dd>ATX (#) or Setext (underlines)</dd>
</dl>
`;

const markdown = convert(docsHtml);
// Output:
// ## API Reference
//
// | Method | Description |
// | --- | --- |
// | `convert()` | Sync conversion |
// | `convertAsync()` | Async conversion |
//
// headingStyle
// :   ATX (#) or Setext (underlines)

Process multiple documents efficiently with async conversion:

import { convertAsync } from "@vakra-dev/supermarkdown";

const urls = [
  "https://example.com/page1",
  "https://example.com/page2",
  "https://example.com/page3",
];

// Fetch and convert in parallel
const markdownDocs = await Promise.all(
  urls.map(async (url) => {
    const html = await fetch(url).then((r) => r.text());
    return convertAsync(html, {
      excludeSelectors: ["nav", "footer"],
    });
  })
);

import { convert } from "@vakra-dev/supermarkdown";

const markdown = convert("<h1>Title</h1><p>Paragraph</p>");

import { convert } from "@vakra-dev/supermarkdown";

const markdown = convert(html, {
  headingStyle: "setext", // 'atx' (default) or 'setext'
  linkStyle: "referenced", // 'inline' (default) or 'referenced'
  excludeSelectors: ["nav", ".sidebar", "#ads"],
  includeSelectors: [".important"], // Override excludes for specific elements
});

For large documents, use convertAsync to avoid blocking the main thread:

import { convertAsync } from "@vakra-dev/supermarkdown";

const markdown = await convertAsync(largeHtml);

// Process multiple documents in parallel
const results = await Promise.all([
  convertAsync(html1),
  convertAsync(html2),
  convertAsync(html3),
]);

Converts HTML to Markdown synchronously.

Parameters:

• html (string) - The HTML string to convert
• options (object, optional) - Conversion options

Returns: string - The converted Markdown

Converts HTML to Markdown asynchronously.

Parameters:

• html (string) - The HTML string to convert
• options (object, optional) - Conversion options

Returns: Promise - The converted Markdown

Elements without Markdown equivalents are preserved as HTML:

• <kbd> - Keyboard input
• <mark> - Highlighted text
• <abbr> - Abbreviations (preserves title attribute)
• <samp> - Sample output
• <var> - Variables

Extracts alignment from align attribute or text-align style:

<table>
  <tr>
    <th align="left">Left</th>
    <th align="center">Center</th>
    <th align="right">Right</th>
  </tr>
</table>

Output:

| Left | Center | Right |
| :--- | :----: | ----: |

Respects the start attribute on ordered lists:

<ol start="5">
  <li>Fifth item</li>
  <li>Sixth item</li>
</ol>

Output:

5. Fifth item
6. Sixth item

When a link's text matches its URL or email, autolink syntax is used:

<a href="https://example.com">https://example.com</a>
<a href="mailto:test@example.com">test@example.com</a>

Output:

<https://example.com>
<test@example.com>

Automatically detects language from class names:

• language-* (e.g., language-rust)
• lang-* (e.g., lang-python)
• highlight-* (e.g., highlight-go)
• hljs-* (highlight.js classes, excluding token classes like hljs-keyword)
• Bare language names (e.g., javascript, python) as fallback

<pre><code class="language-rust">fn main() {}</code></pre>

Output:

```rust
fn main() {}
```

Code blocks containing backticks automatically use more backticks as delimiters.

Line number gutters are automatically stripped from code blocks. Elements with these class patterns are skipped:

• gutter
• line-number
• line-numbers
• lineno
• linenumber

Spaces and parentheses in URLs are automatically percent-encoded:

// <a href="https://example.com/path (1)">link</a>
// → [link](https://example.com/path%20%281%29)

Remove unwanted elements like navigation, ads, or sidebars:

const markdown = convert(html, {
  excludeSelectors: [
    "nav",
    "header",
    "footer",
    ".sidebar",
    ".advertisement",
    "#cookie-banner",
  ],
  includeSelectors: [".main-content"],
});

Some HTML features cannot be fully represented in Markdown:

supermarkdown handles many edge cases gracefully:

Invalid or malformed HTML is parsed via html5ever, which applies browser-like error recovery:

// Missing closing tags, nested issues - all handled
const html = "<p>Unclosed paragraph<div>Mixed<p>nesting</div>";
const markdown = convert(html); // Produces sensible output

Nested lists maintain proper indentation:

const html = `
<ul>
  <li>Level 1
    <ul>
      <li>Level 2
        <ul>
          <li>Level 3</li>
        </ul>
      </li>
    </ul>
  </li>
</ul>`;
// Output:
// - Level 1
//   - Level 2
//     - Level 3

When code contains backticks, the fence automatically uses more backticks:

const html = "<pre><code>Use `backticks` for code</code></pre>";
// Output uses 4 backticks as fence:
// ````
// Use `backticks` for code
// ````

Empty paragraphs, divs, and spans are stripped to avoid blank lines:

const html = "<p></p><p>Real content</p><p>   </p>";
const markdown = convert(html);
// Output: "Real content" (empty paragraphs removed)

Spaces, parentheses, and other special characters in URLs are percent-encoded:

const html = '<a href="https://example.com/file (1).pdf">Download</a>';
// Output: [Download](https://example.com/file%20%281%29.pdf)

Tables missing <thead> use the first row as header:

const html = `
<table>
  <tr><td>A</td><td>B</td></tr>
  <tr><td>1</td><td>2</td></tr>
</table>`;
// Output:
// | A | B |
// | --- | --- |
// | 1 | 2 |

List items with mixed block/inline content are handled:

const html = `
<ul>
  <li>Simple item</li>
  <li>
    <p>Paragraph in list</p>
    <pre><code>code block</code></pre>
  </li>
</ul>`;
// Outputs proper markdown with preserved formatting

Problem: convert() returns empty string or very little content.

Causes & Solutions:

1. Content is in excluded elements - Check if your content is inside nav, header, etc. that might match default patterns

   // Try without selectors first
   const markdown = convert(html);

2. JavaScript-rendered content - supermarkdown converts static HTML only. If the page uses client-side rendering, you need to render it first (e.g., with Puppeteer or Playwright)

3. Content in iframes - iframe content is not extracted. Fetch iframe src separately if needed

Problem: Code blocks don't have language annotation.

Solution: supermarkdown looks for language-*, lang-*, or highlight-* class patterns. Ensure your HTML uses standard class naming:

<!-- Detected -->
<pre><code class="language-python">...</code></pre>
<pre><code class="lang-js">...</code></pre>

<!-- Not detected -->
<pre><code class="python-code">...</code></pre>

Problem: Tables appear as plain text or are malformed.

Causes & Solutions:

1. Missing table structure - Ensure proper <table>, <tr>, <td> structure
2. Nested tables - GFM doesn't support nested tables; inner tables are flattened
3. colspan/rowspan - These are not supported in GFM; content goes in first cell

Problem: Links don't appear or have wrong URLs.

Solutions:

1. Relative URLs - Use baseUrl option to resolve relative links:

   convert(html, { baseUrl: "https://example.com" });

2. Links in excluded elements - Navigation links are often in <nav> which may be excluded

Problem: Conversion is slow for very large HTML files.

Solutions:

1. Use async - convertAsync() won't block the event loop
2. Pre-filter HTML - Remove obvious non-content before conversion
3. Stream processing - For very large docs, consider splitting into sections

Problem: Characters like <, >, & appear as entities.

Solution: This is usually correct behavior - these characters need escaping in markdown. If you're seeing & where you expect &, the source HTML may have double-encoded entities.

Add to your Cargo.toml:

[dependencies]
supermarkdown = "0.0.6"

use supermarkdown::{convert, convert_with_options, Options, HeadingStyle};

// Basic conversion
let markdown = convert("<h1>Hello</h1>");

// With options
let options = Options::new()
    .heading_style(HeadingStyle::Setext)
    .exclude_selectors(vec!["nav".to_string()]);

let markdown = convert_with_options("<h1>Hello</h1>", &options);

supermarkdown is designed for high performance:

• Single-pass parsing - O(n) HTML traversal
• Pre-computed metadata - List indices and CSS selectors computed in one pass
• Zero-copy where possible - Minimal string allocations
• Native code - No JavaScript runtime overhead

Contributions are welcome! Please feel free to submit a Pull Request.

# Clone the repository
git clone https://github.com/vakra-dev/supermarkdown.git
cd supermarkdown

# Run tests
cargo test

# Build Node.js bindings
cd crates/supermarkdown-napi
npm install
npm run build

MIT License - see LICENSE for details.