Docling — The Complete Guide: Converting PDF and DOCX to Markdown and JSON with Tables
Docling started at IBM Research and is now maintained as an independent open-source project. It converts documents into structured output suitable for searching, comparing and feeding to AI tools. The code is under the MIT licence, and the models the tool downloads are subject to their own licences.
Docling
Docling Project (originally IBM Research)
Open-source tool that converts PDF, DOCX and scanned images to Markdown or JSON, with table structure recognition and OCR, running locally.
Get it from the vendorWhat you get
- Varied input — mainly PDF, plus DOCX and images.
- Structure preserved — headings, reading order and tables.
- OCR of your choice — Tesseract, EasyOCR or RapidOCR for scanned files.
- Markdown or JSON output — clean text or a full structure for processing in code.
- Command line and Python — a single file or part of a script.
A practical tip and the limits
If Hebrew documents matter to you, pick an OCR engine that supports Hebrew — Docling itself does not decide the language, and on scans the result depends on the engine you chose. The installation is heavy and needs Python 3.10 or later. The big advantage: everything runs locally, so it also suits specifications and client drawings.
Related tools
For an OCR engine with many language packs — Tesseract. For advanced cell-level table analysis — PaddleOCR. For extracting tables only from PDFs with text — Camelot.
Tesseract OCR
Tesseract OCR project
The open OCR engine behind many tools — turns scans into text in more than 100 languages.
Get it from the vendor
PaddleOCR
PaddlePaddle (Baidu)
Open-source Python OCR and document parsing toolkit: text recognition in dozens of languages, page layout, tables and PDF to Markdown or JSON.
Get it from the vendor
Camelot
Camelot developers (camelot-dev)
Python library for extracting tables from PDF files into data frames, CSV, Excel or JSON, with several parsing methods for ruled and unruled tables.
Get it from the vendor
Surya
Datalab
Python OCR toolkit for 90+ languages with layout and reading-order analysis, tables and equations as LaTeX; models free for personal use and research.
Get it from the vendorThe bottom line
A free, local way to turn a pile of standards and reports into structured text you can work with — provided you are comfortable with a command line.