Drop a file
A PDF, an Excel workbook (.xlsx), or CSV/TSV — up to ~60 MB. Conversion happens entirely in this tab — the file is never uploaded anywhere. Scanned pages and embedded screenshots in PDFs are OCR'd in-browser (English only); the Marker tool remains the higher-quality path for heavy scans and tables when its server is running.
Feeding a large table to an LLM? csv is the smallest — the column names are written once instead of on every row. json repeats a key per cell and is the largest; json (compact) keeps valid JSON but lists headers once with rows as arrays, roughly halving it. markdown sits in between and reads well.
Output
Files bigger than the cap are split into numbered parts — markdown at heading boundaries, JSON at row/block boundaries with every part standing alone as valid JSON — each carrying its source and place in the sequence, so parts can be fed to an LLM one at a time.
This is a dev tool (ADR-8): fully offline, no network requests at all — PDF parsing is Mozilla pdf.js (Apache-2.0) and OCR is Tesseract.js + English traineddata (Apache-2.0), vendored into this page and loaded only when needed. XLSX is unpacked with the browser's built-in decompressor — no spreadsheet library is vendored.