Docling parses documents in many formats into a single DoclingDocument representation and exports them for generative-AI pipelines. Advanced PDF understanding covers page layout, reading order, table structure, code, formulas, and image classification, and the whole pipeline can run locally for sensitive data and air-gapped environments.

Input formats include PDF, DOCX, PPTX, XLSX, HTML, EPUB, Apple Pages, LaTeX, DocLang, plain text, images, audio, and email files, alongside application-specific XML schemas for DocLang, USPTO patents, JATS articles, and XBRL financial reports. Documents export to Markdown, HTML, WebVTT, DocLang, DocTags, and lossless JSON. Scanned material is handled by OCR, and visual language models such as GraniteDocling can drive the parse instead of the default pipeline.

Recent additions parse video files with an automatic speech recognition transcript and representative keyframes, OpenDocument text documents, spreadsheets, and presentations, XBRL reports, email files, EPUB e-books, Apple Pages documents from both container generations, and Markdown supersets. Chart understanding converts bar charts, pie charts, and line plots into tables or code with descriptions. Metadata extraction and molecular structure understanding are listed as planned.

Access is through a command-line tool, a Python API centred on DocumentConverter, an MCP server for agents, and an API server (docling-serve) for running Docling as a service. Native integrations connect it to LangChain, LlamaIndex, CrewAI, and Haystack. It runs on macOS, Linux, and Windows on x86_64 and arm64, and needs Python 3.10 or later.

Features

Input formats: PDF, DOCX, PPTX, XLSX, HTML, EPUB, Apple Pages, LaTeX, images, audio, video, email, and plain text
Export formats: Markdown, HTML, WebVTT, DocLang, DocTags, and lossless JSON
PDF understanding: page layout, reading order, table structure, code, formulas, and image classification
Unified representation: one DoclingDocument model shared by every input and export path
XML schemas: DocLang, USPTO patents, JATS articles, and XBRL financial reports
OCR and VLM pipelines: OCR for scanned documents and visual language models such as GraniteDocling
Audio and video: automatic speech recognition transcripts, with keyframes for video files
Chart understanding: bar charts, pie charts, and line plots converted to tables or code with descriptions
Local execution: the full pipeline runs offline for sensitive or air-gapped environments
Framework integrations: LangChain, LlamaIndex, CrewAI, and Haystack
Agent and service access: an MCP server for agents and an API server for running it as a service