docling-project/docling-core

Docling Core is a library that defines core data types and transformations for document processing in the Docling ecosystem.

The library addresses the need for a standardized, extensible foundation when building document processing and generative AI applications. It centers on the DoclingDocument, a Pydantic model that serves as the primary data structure for representing processed documents. This model-first approach enables strong type safety, schema validation, and interoperability while providing an API for constructing documents and exporting them to formats like Markdown and HTML. Beyond the core data model, the tool offers pluggable APIs for serialization and chunking, allowing developers to customize how documents are converted to different formats and split into segments for downstream tasks. A profiling API extracts structural and statistical metrics from documents, supporting both individual analysis and collection-level insights with visualization capabilities.

Docling Core is suited for teams building document processing pipelines or retrieval-augmented generation systems that need a reliable, schema-driven representation of document content. It works best as a foundation library within the broader Docling ecosystem rather than as a standalone tool. Developers should adopt it if they require fine-grained control over document structure, need to serialize documents in multiple formats, or want to implement custom chunking strategies without reinventing the data model layer.

The project maintains active development with regular updates across its core components. The codebase supports a wide range of Python versions, indicating attention to compatibility across environments. Development is structured around clear separation of concerns, with distinct APIs for serialization, chunking, and profiling that can be extended independently. The project includes comprehensive documentation with architecture guides, API references, and practical examples covering both basic usage and advanced scenarios.