Things people ask first.

Yes. Use the AI's built-in copy button (so markdown formatting is preserved), paste into the left panel, and click DOCX. Headings, lists, code blocks, tables, and LaTeX equations all carry over. Same flow works for Claude, Gemini, Perplexity, and any tool that emits markdown.

Yes. Inline (`$E = mc^2$`) and block (`$$ ... $$`) both render via KaTeX. DOCX exports translate equations to native Office Math (OMML), so they stay fully editable in Word. PDF and HTML render the equations as-is.

That was a real bug — html2canvas mis-positioned KaTeX's fraction bars. It's now fixed: the PDF exporter renders math via the browser's native SVG <foreignObject> path, which lays fractions out correctly.

Free, no account, no quota. PDF/HTML/MD generation runs in your browser; DOCX takes a brief Pandoc detour on our side. Either way there's no per-document cost — nothing to charge for.

Mostly local, with one honest exception: PDF, HTML, Markdown export, and the live preview all run entirely in your browser. DOCX export POSTs the markdown text to our /api/convert-to-docx route so Pandoc can produce a higher-fidelity Word file — the markdown lives in /tmp on the server for the duration of the conversion (a second or two), then the temp files are deleted and the binary is streamed back. We don't log document content, store it, or send it anywhere else. If you'd rather skip the server roundtrip entirely, the converter falls back to a fully-local DOCX path on any failure.

GitHub-Flavored Markdown: headings, bold/italic, lists (nested, with custom start numbers), tables (with alignment), code blocks (with syntax highlighting via highlight.js), inline code, blockquotes, links, images, horizontal rules, strikethrough, task lists, and LaTeX math.

Yes. Code blocks export with monospace fonts and highlighting preserved — 100+ languages auto-detected by highlight.js. Specifying the language (```python) gives the most reliable output.

Markdown images keep their src URLs in HTML/Markdown exports. For DOCX and PDF, cross-origin images are fetched and embedded inline — they render in the document only if the source serves with permissive CORS (most CDNs do; some private hosts don't).

Yes — switch to HTML Input at the top, paste, and the right panel shows clean Markdown you can copy or save as .md. Code blocks, tables, links, and lists round-trip faithfully.

The /batch route accepts up to 20 markdown files via drag-and-drop. Three convert in parallel; the rest queue. Results bundle into a single ZIP archive.

DOCX, PDF, HTML, and Markdown ship today. Notion, Google Docs, EPUB, PPTX, LinkedIn, X, and HTML email are on the roadmap (visible in the More Formats picker).

Yes. The current markdown is saved to localStorage (debounced) and restored on next visit. Recent conversions also appear in the History panel. Clearing browser storage wipes both.

Yes — toggle from the header. The PDF export always uses the light theme regardless of the UI mode, so exports are print-ready.

There's no hard cap, but DOCX and PDF generation is CPU-bound on the client. Documents under ~50,000 words export in a few seconds; larger ones may take longer or stress lower-end devices. The browser tab handles all the work.

There are two DOCX paths. The default sends your markdown to /api/convert-to-docx (server-side Pandoc) for highest fidelity — better tables, cleaner equations, nicer typography. If that route is unreachable or fails, the converter falls back to a fully-local DOCX path using the `docx` npm package. Both are good; Pandoc just wins on edge cases.

Modern Chrome, Edge, Firefox, and Safari. The exporter uses html2canvas, jsPDF, the `docx` package, and KaTeX — all of which require ES2020+ and standard Canvas/SVG support.

ChatGPT occasionally outputs inline math as `( x = 1 )` and block math as `[ x = 1 ]` instead of `$x = 1$` / `$$x = 1$$`. Our markdown preprocessor (lib/utils/markdown-preprocessor.ts) detects the pattern (parens/brackets containing backslashes, carets, or underscores — telltale LaTeX) and rewrites them automatically. If something still won't render, search-and-replace those delimiters by hand.

Highlight.js auto-detects 100+ languages, with explicit hints (```python, ```typescript, ```rust) producing the most reliable output. The same GitHub-Light theme renders in the live preview, HTML export, and PDF export. DOCX exports preserve the language hint as a label and use a monospace font with a gray background.

Yes — the History panel keeps the last 10 conversions in localStorage with a short content preview, timestamp, input/output formats, and success status. Click any entry to re-export. Stored entirely on your device; nothing about the history syncs anywhere.

Three things, all aimed at making LLM-emitted text render correctly: (1) normalizes ChatGPT's parenthesis-wrapped LaTeX to standard `$...$` / `$$...$$`, (2) preserves dollar-sign delimiters that are already correct, and (3) leaves non-LaTeX content untouched. It's intentionally conservative — if a parenthesized expression doesn't look like LaTeX (no backslash, caret, or underscore), it's treated as ordinary prose.

No rate limits on PDF/HTML/MD (those run entirely in your browser; the limit is your CPU). The /api/convert-to-docx Pandoc route is shared infrastructure — under normal use it's instant, but a sustained high-volume batch could hit transient limits. The fully-local DOCX fallback always works regardless.