DoMark

Convert PDF and office documents to clean Markdown, powered by Docling.

DoMark is a small command-line tool built for complex documents such as textbooks with tables, formulas and diagrams, where simpler converters often produce broken tables and lose figures.

Features

• Convert a single file or a whole folder (optionally recursive)
• Interactive mode (just answer two questions) and full CLI mode
• OCR for scanned pages, with an option to force OCR on every page
• Table structure detection (accurate or fast mode)
• Figures exported as image files, embedded as base64, or skipped
• Page range selection for PDFs (5, 5-20, 5-)
• Skip files that were already converted
• Progress logging and a summary at the end
• Unit tests that run without installing Docling

Supported formats

PDF, DOCX, PPTX, XLSX, HTML, Markdown, CSV, and images (PNG, JPG, TIFF, BMP).

Installation

Requires Python 3.10 or newer.

git clone https://github.com/<your-username>/DoMark.git
cd DoMark
python -m venv .venv
# Windows: .venv\Scripts\activate    macOS/Linux: source .venv/bin/activate
pip install -e .

The first run downloads Docling's AI models (a few hundred MB), so an internet connection is required once.

Usage

Interactive mode

domark

Folder path: C:\Users\you\Documents
File name (with extension, optional): book.pdf

Leave the file name empty to convert every supported file in the folder.

Command-line mode

# One file (Markdown is written next to the source)
domark book.pdf

# A whole folder, including subfolders, into a separate output folder
domark ./documents -r -o ./markdown

# Only pages 10 to 40
domark book.pdf -p 10-40

# Faster run: no OCR, fast table model, no figures
domark book.pdf --no-ocr --fast-tables --images none

# Scanned book: force OCR on every page
domark scan.pdf --force-ocr

# Skip files that already have a .md output
domark ./documents --no-overwrite

You can also run it as a module: python -m domark book.pdf.

Options

Option	Description
input	File or folder to convert. Omit for interactive mode.
-o, --output-dir	Folder for output Markdown (default: next to each source).
-r, --recursive	Search subfolders when input is a folder.
-p, --pages	Page range for PDFs: 5, 5-20 or 5- (1-based, inclusive).
--no-ocr	Disable OCR (faster, scanned pages will be empty).
--force-ocr	Run OCR on every page.
--no-tables	Disable table structure detection.
--fast-tables	Use the faster, less accurate table model.
--images {referenced,embedded,none}	Save figures as files, embed as base64, or skip. Default: referenced.
--images-scale	Figure resolution multiplier. Default: 2.0.
--no-overwrite	Skip files whose .md already exists.
-v, --verbose	Show detailed logs.
--version	Show the version.

Exit codes: 0 success, 1 no files found or at least one file failed, 2 invalid arguments.

Performance notes

• Large books (hundreds of pages) can take a long time on a CPU-only machine. Test with -p 1-20 first.
• --no-ocr and --fast-tables make conversion noticeably faster when your PDF already contains selectable text.
• With --images referenced (default), figures are saved in a folder next to the Markdown file and image links are written as relative paths (e.g. book_artifacts/image_000000_abc.png), so the .md and its _artifacts folder can be moved together.

Development

pip install -e ".[dev]"
pytest -q

Project layout:

DoMark/
  src/domark/
    cli.py         # argument parsing, interactive mode, batch loop
    converter.py   # Docling integration (lazy import)
    utils.py       # pure helpers: page ranges, file discovery, output paths
  tests/           # pytest suite (does not require Docling)
  .github/workflows/tests.yml

See CONTRIBUTING.md for guidelines.

Acknowledgements

DoMark is a thin wrapper around Docling. All document understanding is done by Docling.

License

MIT