docling-project/docling

Get your documents ready for gen AI

Docling – Quick Overview

What it does – Docling is a Python library and CLI that turns all kinds of documents (PDF, Word, slides, spreadsheets, HTML, e‑books, audio, video, emails, etc.) into a structured, machine‑readable representation. It extracts layout, tables, code, formulas, images, and even speech‑to‑text from videos, then lets you export the result as Markdown, HTML, JSON, or specialised formats like DocLang and DocTags.

Why it matters for AI – The output is designed to be fed directly into generative‑AI pipelines (LangChain, LlamaIndex, Crew AI, Haystack, etc.). By handling OCR, visual‑language‑model (VLM) analysis, and audio transcription locally, Docling lets developers build retrieval‑augmented generation (RAG) or agentic applications without sending raw files to the cloud.

Key features

  • Broad format support – PDFs, DOCX, PPTX, XLSX, HTML, EPUB, Apple Pages/Numbers/Keynote, ODF, XBRL, email (EML/MSG), images, audio (WAV/MP3), video (MP4/AVI/MOV/…​) and more.
  • Deep PDF understanding – layout, reading order, tables, code blocks, formulas, image classification.
  • Unified DoclingDocument model – a single Python object that holds the extracted structure and metadata.
  • Export options – Markdown, HTML, lossless JSON, DocLang, DocTags, etc.
  • Local‑first – runs on macOS, Linux, Windows, x86_64/arm64; no need for external services, good for sensitive data.
  • AI integrations – ready‑made adapters for LangChain, LlamaIndex, Crew AI, Haystack; can plug in visual language models like IBM’s Granite‑Docling.
  • OCR & VLM – built‑in OCR for scanned PDFs/images and optional VLM pipelines for visual content.
  • Audio/Video – automatic speech‑recognition transcripts and key‑frame extraction.
  • Server modes – an API server (docling-serve) and an MCP (multi‑client‑protocol) server for connecting to agents.

Typical workflow

  1. Install with pip install docling (Python ≥ 3.10).
  2. Use the CLI: docling <url-or-path> → produces a Markdown file.
  3. In code, create a DocumentConverter, call convert(source), then export via document.export_to_markdown() or other formats.
  4. Pass the structured document to an LLM‑based RAG chain or an autonomous agent.

Getting started resources

  • Live docs: https://docling-project.github.io/docling/
  • Quick‑start CLI example and Python snippet in the README.
  • Example notebooks and use‑case recipes under the examples section of the docs.
  • Community: Discord, GitHub Discussions, and an Apache‑licensed codebase (MIT).

Who’s behind it – Initiated by IBM Research Zurich’s AI‑for‑Knowledge team and now a Linux Foundation AI & Data project.

TL;DR

Docling is a versatile, open‑source document‑to‑structured‑data engine that bridges raw files and generative‑AI applications, supporting a huge array of formats, local execution, and ready integrations with popular LLM toolkits.

Related

  • Project
    docling-project/docling-parse

    Docling Parse is a fast, configurable PDF‑parsing library (Python API with a C++ core) that extracts text, words, lines, vector shapes and bitmap images together with their coordinates. It supports both sequential and multi‑threaded parsing, lets you choose what to compute or materialize per page, and includes rendering utilities. Ideal for AI workflows that need layout‑aware document data.

  • Project
    magicrew/doc7

    A document conversion tool that turns PDFs, Office files, and images into AI-ready Markdown using multimodal vision models, supporting local inference for privacy and cost-efficiency.

  • Project
    docling-project/docling-graph

    A toolkit that transforms unstructured documents into validated Pydantic objects and directed knowledge graphs for high-precision entity relationship extraction.

  • Project
    MarkPDFdown/markpdfdown

    A tool that uses multimodal LLMs to transcribe PDF files and images into well-formatted Markdown, preserving complex structures like tables and formulas.

  • Project
    docling-project/docling-mcp

    An MCP server that enables AI agents to convert PDFs into structured formats, generate documents, and integrate with RAG pipelines using the Docling library.