Markdown, Word, HTML, TXT, JSON and CSV convert into one another in any direction — 36 paths in all, and each one has its own page below.
Full GitHub-flavoured Markdown: headings, nested lists, tables, fenced code with language tags, quotes, links and images. Formulas written between dollar signs are recognised as maths rather than left as literal text.
The .docx is unzipped in the page and read as WordprocessingML: heading styles become heading levels, real numbering definitions become ordered and bulleted lists, bordered paragraphs become quotes, and fully monospaced runs become code blocks. Embedded pictures are not carried over, and the converter tells you how many it left behind.
The page is parsed as a real document, not scraped with regular expressions. When the input is a whole web page, the <article> or <main> body is preferred so navigation, sidebars and footers stay out; scripts, styles and inline SVG are dropped, and headings, lists, tables, code blocks with their language class, quotes, links and images all survive.
Plain text is taken literally. A line beginning with # stays a line beginning with #, because a .txt file has no markup to interpret — nothing is silently promoted into a heading, a list or a link, and blank lines are all that separate one paragraph from the next.
JSON written in our own saveai.doc.v1 shape — a title plus an array of typed blocks — is rebuilt into a real document, with headings, paragraphs, lists, quotes, code and tables all restored. Any other JSON is kept intact inside a fenced code block instead of being guessed at, so nothing is invented and nothing is lost.
A CSV, TSV or semicolon-separated export is read to RFC 4180: quoted fields, doubled quotes, and commas or line breaks inside a cell all survive, and the delimiter is detected rather than assumed. The first row becomes the table header, and a cell holding a line break is flattened onto one line — a Markdown table row cannot span lines.
PDF appears as a target only. We render documents into PDF, but we do not parse a PDF back into text — so there is no PDF-to-anything page here. Doing that well needs layout reconstruction and OCR, and a page that promised it would only disappoint.
SaveAI is a browser extension that saves whole ChatGPT, Claude, Gemini and DeepSeek conversations as PDF, Word, Markdown and more — images, code and formulas included.
