Glossary

Relevant source files

• .github/workflows/checks.yml
• CHANGELOG.md
• README.md
• docling/backend/abstract_backend.py
• docling/backend/docling_parse_backend.py
• docling/backend/docling_parse_v2_backend.py
• docling/backend/docling_parse_v4_backend.py
• docling/backend/html_backend.py
• docling/backend/md_backend.py
• docling/backend/msword_backend.py
• docling/backend/pypdfium2_backend.py
• docling/cli/main.py
• docling/datamodel/backend_options.py
• docling/datamodel/base_models.py
• docling/datamodel/document.py
• docling/datamodel/pipeline_options.py
• docling/datamodel/pipeline_options_vlm_model.py
• docling/datamodel/vlm_model_specs.py
• docling/document_converter.py
• docling/pipeline/vlm_pipeline.py
• docling/utils/code_language.py
• docling/utils/layout_postprocessor.py
• docling/utils/locks.py
• docs/index.md
• docs/usage/supported_formats.md
• mkdocs.yml
• packages/docling/pyproject.toml
• pyproject.toml
• tests/data/html/groundtruth/html_code_snippets.html.json
• tests/test_backend_html.py
• tests/test_backend_markdown.py
• tests/test_backend_msword.py
• tests/test_code_language_detection.py
• tests/test_e2e_ocr_conversion.py
• tests/test_input_doc.py
• tests/test_invalid_input.py
• tests/test_verify_utils.py
• tests/test_vlm_pipeline_force_backend_text.py
• tests/verify_utils.py
• uv.lock

This glossary defines the core technical terms, domain concepts, and architectural components of the Docling codebase. It is intended for onboarding engineers to bridge the gap between high-level document processing concepts and their specific implementations in the code.

Core System Entities

DocumentConverter

The primary orchestrator of the Docling SDK. It manages the conversion lifecycle, including format routing, backend selection, and pipeline execution.

• Implementation: DocumentConverter in docling/document_converter.py133
• Key Method: convert() handles single or batch sources, returning a ConversionResult docling/document_converter.py343-348
• Format Options: Users can customize conversion via FormatOption subclasses like PdfFormatOption, WordFormatOption, or ExcelFormatOption docling/document_converter.py95-248

DoclingDocument

The unified, hierarchical data model representing a document. It is format-agnostic and stores structural elements like headers, tables, and paragraphs.

• Implementation: Defined in docling-core and re-exposed in docling/datamodel/document.py27-35
• Data Model: Includes structural items such as TextItem, TableItem, SectionHeaderItem, and PictureItem docling/datamodel/document.py28-35

ConversionResult

The output object of a conversion process. It encapsulates the resulting DoclingDocument, execution metadata, and any errors encountered.

• Implementation: ConversionResult in docling/datamodel/document.py280
• Status: Tracks the outcome using ConversionStatus (e.g., SUCCESS, PARTIAL_SUCCESS, FAILURE) docling/datamodel/base_models.py85-91

---

Architectural Patterns

Document Backend

An abstraction layer for interacting with raw document files. Backends provide methods to extract text, images, and metadata from specific formats.

• Abstract Base: AbstractDocumentBackend docling/backend/abstract_backend.py39
• Paginated Backend: For page-based formats like PDF, implementing PaginatedDocumentBackend docling/datamodel/document.py52
• Declarative Backend: For structured formats like DOCX or HTML, implementing DeclarativeDocumentBackend docling/backend/abstract_backend.py175

Processing Pipeline

A sequence of operations that transforms backend data into a DoclingDocument.

• StandardPdfPipeline: A multi-stage pipeline for PDFs involving layout analysis and OCR docling/pipeline/standard_pdf_pipeline.py46
• SimplePipeline: A direct mapping pipeline for structured formats docling/pipeline/simple_pipeline.py18
• VlmPipeline: A vision-language model-based pipeline for document understanding docling/pipeline/vlm_pipeline.py34
• AsrPipeline: A pipeline for audio transcription using Whisper models docling/pipeline/asr_pipeline.py84
• VideoPipeline: A pipeline for video files providing transcripts and keyframes docling/pipeline/video_pipeline.py43

Data Flow Overview

The following diagram bridges the gap between natural language concepts and the specific code entities involved in a conversion.

Natural Language to Code Entity Mapping

Sources: docling/document_converter.py133-348 docling/datamodel/document.py125-280 docling/pipeline/base_pipeline.py34

---

Domain Specific Terms

Term	Definition	Code Pointer
OCR Options	Configuration for OCR engines like Tesseract, EasyOCR, or RapidOCR.	docling/datamodel/pipeline_options.py164-198
TableFormer	ML model for table structure recognition and cell extraction.	docling/datamodel/pipeline_options.py122-157
Furniture	Non-body document elements like headers, footers, and page numbers.	docling/datamodel/document.py84-103
InputFormat	Enum defining supported input types (PDF, DOCX, AUDIO, etc.).	docling/datamodel/base_models.py94-126
Accelerator	Hardware acceleration settings (CPU, CUDA, MPS, XPU).	docling/datamodel/accelerator_options.py52

---

Technical Jargon & Abbreviations

• ASR: Automatic Speech Recognition. Handled by AsrPipeline docling/pipeline/asr_pipeline.py84
• VLM: Vision Language Model. Used for document understanding. Options include InlineVlmOptions and ApiVlmOptions docling/datamodel/pipeline_options.py47-52
• JATS: Journal Article Tag Suite. Supported via JatsDocumentBackend docling/backend/xml/jats_backend.py33
• XBRL: eXtensible Business Reporting Language. Supported via XBRLDocumentBackend docling/backend/xml/xbrl_backend.py40
• USPTO: U.S. Patent and Trademark Office XML format. Handled by PatentUsptoDocumentBackend docling/backend/xml/uspto_backend.py39
• OTSL: Optimised Table Structure Language. Used by TableFormer models for structure prediction CHANGELOG.md330

---

Internal Components & Registry

Docling uses a factory and plugin architecture to manage model selection and extensibility.

Model Factory and Plugin Architecture

Sources: docling/cli/main.py147-152 docling/document_converter.py133 docling/models/factories/base_factory.py1-20

• Plugin System: Uses pluggy to discover and load OCR and layout engines pyproject.toml56
• MIME Type Routing: Maps file extensions and MIME types to InputFormat docling/datamodel/base_models.py142-228
• KServe v2: Integration for remote model serving via gRPC or HTTP docling/datamodel/pipeline_options.py30
• Docling-Slim: A modular version of the package with minimal base dependencies (~50MB) pyproject.toml6-58
• Strict OOXML: A cheap detection mechanism for Word documents to ensure compatibility with python-docx docling/backend/msword_backend.py167-180

---

Sources:

• docling/document_converter.py1-248
• docling/datamodel/document.py1-280
• docling/datamodel/base_models.py1-228
• docling/datamodel/pipeline_options.py1-200
• pyproject.toml1-240
• docling/cli/main.py1-152
• docling/backend/msword_backend.py1-180
• CHANGELOG.md1-100