⚒️ Markforge

A modern TypeScript library for crafting HTML into clean, readable Markdown. Built with LLMs and AI agents in mind. This is a modern TypeScript fork of the Turndown library, completely rewritten for modern use cases.

Features

• 🤖 LLM-Optimized Output: Produces clean, consistent Markdown that's ideal for LLMs and AI agents
• 🚀 Modern TypeScript: Full type safety and modern ES module support
• 🔧 DOM-Independent: Works in any Node.js environment without requiring a DOM
• ✨ Built-in GFM: GitHub Flavored Markdown support included by default
• 🎯 Zero Runtime DOM: Uses virtual DOM for parsing, keeping your bundle size small
• 🌳 Tree-Shakeable: Import only what you need
• 🔄 Functional Core: Built with functional programming patterns

Installation

npm install markforge
# or
yarn add markforge
# or
pnpm add markforge

Usage

Basic Usage

import markforge from 'markforge'

const html = '<h1>Hello, World!</h1>'
const markdown = markforge.toMarkdown(html)
console.log(markdown) // # Hello, World!

LLM/AI Agent Usage

Perfect for cleaning up HTML content before sending to LLMs:

import Markforge from 'markforge'

// Example: Processing HTML content for an LLM
async function processForLLM(htmlContent: string) {
  const markforge = new Markforge()
  const markdown = markforge.toMarkdown(htmlContent)

  // The output is clean, consistent Markdown that LLMs can easily understand
  const llmResponse = await llm.complete({
    prompt: markdown,
    // ... other options
  })

  return llmResponse
}

Custom Instance

import Markforge from 'markforge'

const service = new Markforge({
  headingStyle: 'atx',
  codeBlockStyle: 'fenced',
  emDelimiter: '*',
  strongDelimiter: '**'
})

const markdown = service.toMarkdown('<h1>Custom Options</h1>')

GFM Features

Built-in support for GitHub Flavored Markdown:

// Tables
const table = `
<table>
  <thead>
    <tr>
      <th>Feature</th>
      <th>Support</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Tables</td>
      <td>✅</td>
    </tr>
  </tbody>
</table>
`
markforge.toMarkdown(table)
// | Feature | Support |
// | ------- | ------- |
// | Tables  | ✅      |

// Task Lists
markforge.toMarkdown('<li><input type="checkbox" checked> Task</li>') // - [x] Task

// Strikethrough
markforge.toMarkdown('<del>removed</del>') // ~~removed~~

Custom Rules

import type { Rule } from 'markforge'

const customRule: Rule = {
  filter: 'span',
  replacement: (content) => `{${content}}`
}

markforge.addRule('customSpan', customRule)

Why Markforge?

For LLM/AI Applications

• Clean Output: Generates consistent, well-formatted Markdown that LLMs can easily process
• Reliable Parsing: Handles messy HTML input gracefully
• Semantic Preservation: Maintains document structure and meaning
• Lightweight: No heavy DOM dependencies that could bloat your AI application

For Modern Applications

• TypeScript First: Built from the ground up with TypeScript
• Modern Bundle: ES modules with tree-shaking support
• Minimal Dependencies: Only uses domino for HTML parsing
• Framework Agnostic: Works anywhere JavaScript runs

API

MarkforgeOptions

interface MarkforgeOptions {
  headingStyle?: 'setext' | 'atx'
  hr?: string
  bulletListMarker?: '*' | '+' | '-'
  codeBlockStyle?: 'indented' | 'fenced'
  fence?: '```' | '~~~'
  emDelimiter?: '_' | '*'
  strongDelimiter?: '__' | '**'
  linkStyle?: 'inlined' | 'referenced'
  linkReferenceStyle?: 'full' | 'collapsed' | 'shortcut'
  br?: string
  preformattedCode?: boolean
}

Methods

• toMarkdown(input: string | Node): string - Convert HTML to Markdown
• use(plugin: Plugin | Plugin[]): MarkforgeService - Use plugins
• addRule(key: string, rule: Rule): MarkforgeService - Add custom rules
• keep(filter: string | string[] | Function): MarkforgeService - Keep elements as HTML
• remove(filter: string | string[] | Function): MarkforgeService - Remove elements

Contributing

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

Credits

This project is a modern TypeScript rewrite of Turndown by Dom Christie, with additional features and optimizations for modern use cases, particularly LLM and AI applications.

License

MIT © Max Schedin