MarkItDown Document Pack

Problem

Engineers often need to turn scattered PDFs, Office documents, web pages, emails, images, audio/video files, and 2D/3D engineering files into material that can be read, searched, and audited. Manual copying of text, tables, screenshots, and metadata can lose source boundaries and present failed items as successful results.

How It Works

MarkItDown converts multiple inputs into a material pack whose primary output is Markdown. Ordinary documents extract text, headings, tables, and media references; folders, ZIP archives, and EML files are split into separate materials; images and scanned PDFs can be OCR-processed locally; audio/video tracks can be transcribed in segments; engineering files produce a fixed eight-section brief, an engineering-material-v1 JSON sidecar, and SVG/PNG previews. Batch runs precheck inputs, limit concurrency, commit receipts atomically, and support resumable execution.

Boundaries

It does not re-encode media files, nor replace B-Rep, assembly, or manufacturing approvals. OCR, transcription, and engineering parsing only reflect local capability results; failures, partial parsing, or metadata-only outputs are explicitly labeled and should not be treated as complete evidence.