DOCX Input
Drop .docx here or click to upload
Microsoft Word 2007+ — converted entirely in your browser
Markdown Output
DOCX files are ZIP archives packed with XML schemas, styles, and relationship graphs. Extracting text from DOCX for LLM context is noisy and inefficient. Markdown stores only content — typically cutting token count by 60–80% compared to naive DOCX text extraction.
Word headings (Heading 1, Heading 2) map directly to Markdown H1/H2. RAG chunking libraries split on these boundaries, creating semantically coherent chunks that improve retrieval precision and reduce hallucination from context fragmentation.
DOCX is binary — git diff shows only "Binary files differ". Markdown produces clean, line-by-line diffs that make code review, change tracking, and collaboration straightforward in any version control system.
A typical DOCX is 20–200KB — most of that is XML overhead. The same document as Markdown is 2–20KB of pure text. More documents fit in context windows per API call, and embedding costs drop proportionally.
Markdown renders natively in GitHub, GitLab, Notion, Obsidian, Confluence, VS Code, and virtually every AI developer tool. DOCX requires Microsoft Office or compatible software to open and render.
Markdown is plain text that will be readable in any text editor for decades. DOCX relies on the Office Open XML spec and Microsoft compatibility — documents from 2003 already require conversion steps. Markdown has no version lock-in.
The tool uses mammoth.js — an open-source JavaScript library that reads the Office Open XML structure inside .docx files directly in your browser. It interprets Word's semantic styles (Heading 1 → #, Heading 2 → ##, bold → **text**, italic → *text*) and converts them to their Markdown equivalents.
Unlike tools that attempt to replicate visual layout, mammoth focuses on semantic meaning — producing clean, portable Markdown that feeds directly into documentation systems and AI pipelines without post-processing.
No data is sent to any server. The entire conversion happens in memory inside your browser tab.
**bold** and *italic*| Feature | Markdown (.md) | DOCX (.docx) |
|---|---|---|
| LLM token cost | ✅ Very low — plain text only | ❌ High — XML overhead, binary data |
| RAG chunking quality | ✅ Semantic heading/paragraph structure | ⚠️ Requires XML parsing and style mapping |
| File size | ✅ 2–20 KB typical | ❌ 20–200 KB typical |
| Git / version control | ✅ Clean line-by-line diffs | ❌ Binary — no meaningful diff |
| Editable without software | ✅ Any text editor | ❌ Requires Word or LibreOffice |
| GitHub / Notion / Obsidian | ✅ Native rendering | ❌ Requires conversion or plugin |
| Print / visual fidelity | ⚠️ Requires rendering step | ✅ Exact layout preserved |
Why is Markdown more token-efficient than DOCX for LLMs?
A .docx file is a ZIP archive containing XML files for content, styles, relationships, and media. When you extract text from DOCX for an LLM, you're either parsing through noisy XML or using a library that loses structural information. Markdown is clean, linear plain text — an LLM reads it directly with no parsing overhead. A 10-page DOCX document that produces 6,000 tokens of raw text may produce only 1,200–2,000 tokens as structured Markdown, because heading markers, list bullets, and paragraph breaks consume almost no tokens while preserving full semantic meaning.
How do I convert a Word document to Markdown?
Click Upload DOCX or drag your .docx file onto the left panel. The conversion is instant. Then click Copy to copy to clipboard or Download .md to save the file.
Is my Word document uploaded to a server?
No. The conversion runs entirely in your browser using mammoth.js. Your document never leaves your device.
Does it support .doc files (older Word format)?
No — only .docx (Office Open XML, Word 2007+). To convert a .doc file, open it in Microsoft Word or LibreOffice and save as .docx, then upload here.
What formatting is preserved?
Headings (H1–H4), bold, italic, ordered and unordered lists (with nesting), and hyperlinks are preserved. Images, text boxes, footnotes, endnotes, and complex table layouts are not included in the Markdown output.
Can I use the output directly in a RAG pipeline?
Yes. Download the .md file and pass it to LangChain's MarkdownHeaderTextSplitter or LlamaIndex's MarkdownNodeParser. These chunk by heading, creating semantically coherent pieces that improve retrieval relevance significantly over arbitrary character-count chunks from raw DOCX text.
Can I import the output into Obsidian or Notion?
Yes. For Obsidian, copy the .md file into your vault folder. For Notion, use Import → Markdown & CSV. Both render headings, bold, italic, and lists natively.
What is mammoth.js?
mammoth.js is an open-source JavaScript library by Michael Williamson that converts .docx files to HTML or Markdown in the browser. It maps Word's semantic styles (Heading 1, Heading 2, Strong, Emphasis) to their HTML and Markdown equivalents, rather than trying to replicate the visual layout — making conversions clean and portable.
Should I store my knowledge base as DOCX or Markdown?
For AI-first and developer-first knowledge bases, Markdown is the better choice: smaller files, Git-diffable, renders everywhere, and feeds LLMs with minimal token overhead. Use DOCX when you need precise print layout, track changes for legal review, or share with non-technical stakeholders who use Microsoft Office.