pgx-docling-parser: A parser plugin using docling to convert your favourite documents

[T-Eberle]

With the new feature coming in v3 of paperless - developing and integrating third party parsers - I am happy to share a docling flavored parser to convert your favorite documents! 🐥

Pypi: https://pypi.org/project/pgx-docling-parser-serve/

GitHub: https://github.com/T-Eberle/paperless-docling-parser

This is just a taste of what can come in future - means this is still quite experimental and WIP.

Here are a few things to point out:

• Currently, it only supports PDFs. Other document types will come in the near future.
• Currently, you can chose between easyOCR, tesseract and the granite_docling model. More probably to come.
• As content of the document, the parser sets the generated Markdown output of the document conversion.
• At the moment, only a "docling_serve" flavored plugin is available. I am working on a "local" version, where docling is installed as a separate dependency: I split them up, as the docling_serve plugin has a much lower footprint than a local version will have.

Things you can expect in the future:

• Support for more document types
• Adding metadata to the document (I would love to add the DoclingDocument JSON as metadata, if that is possible in the near future)
• I was thinking of a vector_database feature, where the parsed data can directly be chunked and uploaded to a vector database. Docling has the hybrid chunking feature, which I could imagine to use here.

To use the plugin, simply install it, e.g. by creating your own dockerfile:

 Use the official Paperless-ngx image as base
FROM ghcr.io/paperless-ngx/paperless-ngx:beta

# Switch to root to install packages
USER root

# Install minimal system dependencies (only what's needed for PyMuPDF for thumbnails)
RUN apt-get update && \
    apt-get install -y --no-install-recommends \
    libgl1 \
    libglib2.0-0 && \
    apt-get clean && \
    rm -rf /var/lib/apt/lists/*


# Install the docling-serve package (automatically installs core as dependency)
RUN pip install --no-cache-dir pgx-docling-parser-serve==0.1.15

# Switch back to paperless user
USER paperless

Feel free to contribute and share your ideas!🚀

[stumpylog]

One thing to note, for Docker, custom init scripts are probably easier for users: https://docs.paperless-ngx.com/advanced_usage/#custom-container-initialization