How does the HTML to Markdown converter handle HTML tables and code blocks?

Every is parsed row-by-row into a valid GitHub-Flavored Markdown pipe table, and blocks are converted into fenced code blocks with language identifiers preserved.

Are my HTML snippets uploaded to any server?

No. HTML-to-Markdown compilation runs 100% locally inside your browser tab with zero server requests.

HTML to Markdown Converter Online — Clean GFM Table & Code Extractor

Dual-Engine Compiler Ready — Local Browser AST (0ms) + Free Workers AI Edge

Source HTML, XML, JSON, or text payload Auto-counts BPE tokens

<article>
  <header><nav>Home > Docs > AI Engineering</nav></header>
  <h1>Cloudflare Workers AI Document Pipeline</h1>
  <p>Converting noisy HTML and PDF documents into <strong>structured Markdown</strong> reduces LLM context window token consumption by up to <em>74%</em>.</p>
  <h2>Supported Conversion Modes</h2>
  <ul>
    <li><strong>Local AST Compiler:</strong> Instant 0ms conversion in your browser for HTML, CSV, TSV, and JSON.</li>
    <li><strong>Workers AI Edge Binding:</strong> Server-side extraction via <code>env.AI.toMarkdown()</code> for PDF, DOCX, and XLSX.</li>
  </ul>
  <table>
    <thead><tr><th>Format</th><th>Raw Tokens</th><th>Markdown Tokens</th><th>Savings</th></tr></thead>
    <tbody>
      <tr><td>HTML Web Page</td><td>4,280</td><td>1,120</td><td>73.8%</td></tr>
      <tr><td>PDF Technical Spec</td><td>6,900</td><td>2,450</td><td>64.5%</td></tr>
    </tbody>
  </table>
  <pre><code class="language-typescript">const [doc] = await env.AI.toMarkdown([{ name: "spec.pdf", blob }]);</code></pre>
  <footer>Copyright 2026 Boilerplate Footer</footer>
</article>

Strip <nav>/<footer>/<script> Convert HTML <table> to GFM Preserve [links](url) Prepend YAML frontmatter

Document title metadata

File upload execution mode

Source BPE tokens

182

Markdown BPE tokens

114

Context reduction

-37.4% saved

Active engine

Local AST (0ms)

Compiled GitHub-Flavored Markdown (GFM) output Ready for LLM / RAG ingestion

Chunk for RAG vector DB

Why Clean HTML-to-Markdown Conversion Matters for RAG & Web Scraping

Modern web pages wrap article prose in deeply nested <div> trees, SVG icons, tracking scripts, and navigation menus. When building Retrieval-Augmented Generation (RAG) crawlers or feeding documentation into coding agents:

Boilerplate Elimination: Removing <nav>, <footer>, <aside>, and <script> tags prevents irrelevant navigation links from polluting vector embeddings.
Deterministic Table Serialization: Converting <table> elements into GFM pipe tables (| Header | Value |) preserves row-column relationships that plain innerText extraction destroys.
YAML Frontmatter Metadata: Prepending structured YAML frontmatter (title, estimated_bpe_tokens) enables downstream chunkers like our RAG Semantic Markdown Chunker to attach document provenance to every vector record.

HTML & Web DOM to GitHub-Flavored Markdown Converter

Why Clean HTML-to-Markdown Conversion Matters for RAG & Web Scraping