Why we built ground.md.
We made a free tool at CURTIS Digital that turns a PDF, Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, or CSV file into Markdown in your browser. No upload, no account, and no signup. The parsers run locally through WebAssembly, so the file never gets sent to our servers or anyone else’s. Drop in a contract, a slide deck, a workbook, or a 400-page book. You get Markdown, plain text, and JSON along with the images, and you can preview all of it before you download anything.
Why local processing
That “never leaves your machine” part is the whole point. While we were building this, we found a client using the free tier of an online converter, and we noticed it was uploading every file to the converter’s servers. The client had no idea. If the file is a contract or an internal report, that upload is a disclosure. So we cut this piece out of our client tooling and made it free for everyone. The processing happens on your device, and you don’t need to take our word for it. Open your browser’s network tab and watch. No document gets sent.
Why Markdown
A PDF only records where each character sits on the page. It cannot tell a heading from a table, and it has no idea what reading order is. Paste PDF text into an LLM, or index it for retrieval, and headings sink into the body text, columns come out in the wrong order, and tables collapse. Markdown keeps the structure. Headings stay headings and tables stay tables. For retrieval, that means chunks split on real section boundaries instead of on a character count. It is also compact, so more of the document fits in a context window.
Word, PowerPoint, Excel, and EPUB files have the opposite problem. The structure is all there, buried in XML nobody wants to paste into a prompt. ground.md pulls out the headings, lists, tables, footnotes, and speaker notes and writes them as Markdown.
Spreadsheets also come out as one Markdown file per visible, non-empty sheet. If an XLSX/XLSM workbook is too big to convert in one pass, you pick the sheets you need and it converts those on their own as capped CSV, with Markdown added when the sheet is small enough to be safe.
We hit this problem constantly, most often when we help clients get their content ready for AI tooling. Pulling documents into a shape a language model understands is the unglamorous half of the job, and this tool covers the everyday formats without leaving the browser.
Who built this
My name is John Curtis, and I run CURTIS Digital. The work is broader than a services menu. We follow the problem wherever it goes, across product, cloud, data, AI, operations, and the small tools in between. Some of that work is operational platforms that reconcile messy inputs and make the real state visible. Some of it is bounded AI systems, where analysis, writing, and agent workflows sit inside schemas, cost records, review states, and human judgment. Some of it is data and content machinery built for volume and traceability.
And some of it is a single-purpose tool, built for the moment when a platform would be a waste. ground.md is one of those. It does one thing, it runs where your data already is, and everything it does can be verified.
Why it’s free
This tool solves one problem completely. It gets the text and structure of an everyday document into a shape language models read well. I give that away because it is the first step in nearly every AI project we take on, and most teams shouldn’t have to pay anyone for it.
The steps after it are where the real work begins. Scanned pages. Image-only files. Tables that span pages. Thousands of files with provenance and audit requirements. A retrieval system that has to answer correctly under load. We build those systems. If ground.md gets you to the point where you can see that problem clearly, it has done its job.
If you are there, email me at john@curtisdigital.com or find me on X at @jcurtis. Tell me what the documents are and what decision they need to support, and I will tell you plainly whether it is a project or a Tuesday afternoon.
Limits
PDFs are read from the digital text layer only. There is no visual processing and no OCR, and it warns you on drop if the file looks scanned. Office and OpenDocument conversions drop comments, tracked deletions, headers and footers, and hidden sheets. Spreadsheets come through with stored values, not formulas. Password-protected files are rejected. There is no conversion API.
It is free at ground.md. See also privacy, the agent guide, and CURTIS Digital.