Local docs, clear limits.

ground.md converts PDF, Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, and CSV files to Markdown in the browser. Parsing runs in the page with WebAssembly: LiteParse for PDF, AnyDoc in a Web Worker for every other format. File bytes are never uploaded. There is no upload or conversion API. PDF reads the digital text layer only; scanned or image-only pages are flagged and are not OCRed.

Using the converter

Open the homepage, choose a local file, wait for conversion, and download the desired artifact. Spreadsheet results include the merged document and one Markdown file per visible, non-empty sheet. If an XLSX/XLSM workbook reaches the normal browser safety limit, ground.md offers an explicit sheet picker. Recovery streams selected sheets independently, keeps capped CSV for each sheet that fits, adds Markdown when safely sized, and does not create a merged workbook file. A sheet stops before 500,000 non-empty rows or 64 MB of CSV output, and recovered CSV and Markdown share a 64 MB total output budget; truly empty rows are omitted and counted. The header shows the unit that fits the format: pages for PDF, slides for pptx and odp, visible sheets for xlsx, sheets for ods, chapters for EPUB, and rows for CSV. Word docx, OpenDocument text, and RTF show pages only when the file's metadata stores a count from the authoring app; nothing is rendered to count them. Legacy doc, ppt, and xls show no count. For repeatable browser automation, use the selectors and instructions in agents.md. There are no accounts, no API keys, and no server-side document processing.

What is dropped

The converter drops comments, tracked deletions, headers and footers, and hidden sheets, rows, and columns. It keeps speaker notes, footnotes, and tracked insertions. Spreadsheets come through with stored cell values, not formulas. It rejects password-protected files. The cleanup transforms in the footer apply to PDF input only.

API documentation

The API is intentionally narrow: health, public processed-unit statistics, and a browser-emitted completion metric. Public stats report pages, slides, sheets, chapters, and rows by format. They do not report document or job counts. It is not a document upload or conversion API. The complete machine-readable contract is openapi.json; current routes use /api/v1, while unversioned routes are deprecated aliases.

Read privacy for data handling and the developer guide for discovery resources.