Any document in. Clean Markdown out.

Word, Excel, PowerPoint, OpenDocument, EPUB, RTF, HTML, CSV and PDFs with a text layer convert right here in your browser — your file is never uploaded. Real GFM tables, heading levels from styles, nested lists, footnotes, YAML front matter — and out again as Markdown, a JSON AST, standalone HTML or plain text. An optional AI pass repairs the structure a parser cannot decide.

⇩

Drop a document here or

Type is detected from the bytes, not the file name. Up to 40 MB. Nothing leaves your machine in the free lane.

Or paste a link to a document

SkillSafe fetches the link on its servers and returns the text — PDF, .docx, HTML or plain text, up to 4 MB. It is free, but it needs a sign-in and it is the one input that does not stay in your browser.

No file handy? Try a sample:

Conversion options

YAML front matter from document metadata Page / slide / sheet markers as comments Structural notes as HTML comments Promote bold PDF lines to headings (inference) Max table rows per sheet

Options apply when a file is converted. Change one and re-convert without dropping the file again — it is free and stays in your browser. Promote bold PDF lines to headings is the free answer to the same missing headings the metered repair below charges to fix; try it first.

Convert from your SkillSafe drive

Pick documents you already have stored, convert them in a batch, and write the Markdown back to the drive next to the source. Share a whole folder with doc-to-md and every document you drop into it later is covered too.

Other formats — all free, all in your browser

The same document, serialized differently. None of these costs credits or needs an account; they come out of the model the Markdown was written from.

The .zip above bundles every one of these together, with any extracted images under assets/.

AI structure repair metered

The free conversion above measured what it produced and named what it could not decide. The repair pass fixes exactly those gaps — heading hierarchy, tables a PDF never declared, merged headers, reading order, OCR noise — and its output is checked against the measured facts afterwards. It never sees your original file, only the Markdown above.

Model for this run

Anything the repair should know (optional)

Repaired document

Past conversions

Stored with your account, so they follow you across devices. The converted Markdown is kept; your original file never is.

None yet. Convert something above.

What converts, and how

Format		Lane	What is read

Legacy binary Office files (.doc, .xls, .ppt) are a different format entirely — save them as .docx/.xlsx/.pptx first. Scanned PDFs have no text layer: the free lane names the scanned pages, then SkillSafe's server-side extractor gets a free attempt at them, and only if that finds nothing usable does the metered page-image read become the answer — and it needs the scan to be JPEG (most are).

Questions people actually ask

Is my document uploaded anywhere?

Not in the free lane. Detection, parsing and Markdown writing all run in this page, so a .docx or a PDF you drop in never leaves the machine. Three things do leave, and each is a button you press deliberately: pasting a link (SkillSafe fetches it server-side), the free server-side extractor for scanned PDFs, and the metered AI repair — which is sent the converted Markdown and the measured counts, never the original file.

Which formats convert?

Word (.docx), Excel (.xlsx), PowerPoint (.pptx), OpenDocument (.odt, .ods, .odp), EPUB, RTF, HTML, CSV, TSV, JSON, plain text and Markdown, plus PDFs that carry a text layer. The type is detected from the file's bytes, not its extension. Legacy binary .doc, .xls and .ppt are a different format and must be saved as the modern equivalent first.

Why did my PDF's tables come out as blocks of spaces?

Because a PDF usually does not declare tables at all — it declares glyphs at coordinates. The converter reports this honestly as a pdf-tables gap rather than guessing, and the metered AI repair rebuilds those regions as real GitHub-Flavored Markdown tables. Its output is then reconciled against the counts and the numbers the parser measured, so a repair cannot quietly drop or alter a figure.

Does it work on scanned PDFs?

A scanned PDF has no text layer, so the free in-browser lane can only tell you that. From there SkillSafe's server-side extractor gets a free attempt, and only if that finds nothing usable does reading the page images with the model — which is metered, up to four pages a run, and needs the pages stored as JPEG — become the answer.

What costs credits?

Conversion, every export format, the diff, the .zip and your conversion history are free. Only the AI structure repair, the scanned-page read and each refinement turn are metered. The reservation is shown before you press the button, the run is settled at actual usage afterwards, and the button stays disabled if your balance cannot cover the minimum.

How do I see what the AI repair actually changed?

The What changed tab on the repaired document is a line-level diff against the free conversion, or against the previous refinement turn. It is computed in your browser from documents already on the page, so looking costs nothing.

Can I get something other than Markdown?

Yes — the same parsed document also serializes to a JSON AST, standalone HTML, plain text and YAML front matter, all free and all in the browser. The .zip bundles every one of them together with any images extracted under assets/.

Can I drive it from a script?

Yes. The API guide documents the exact input fields and the output contract in cURL, Python, JavaScript, Go, Java, Ruby, PHP and C#, and API tokens gets you a token without opening a developer console.