DocLang: optimizing AI workflows

What is DocLang?

DocLang is an open specification for representing documents in a machine-readable format optimized for AI consumption. Co-developed by ABBYY, IBM, NVIDIA, Red Hat, Human Signal, and the LF AI and Data Foundation, DocLang defines a standardized, AI-native way to encode:

• Document structure and semantic meaning
• Geometric layout information
• Governance metadata

This ensures that AI models, RAG systems, and agentic workflows receive documents in a consistent, predictable format, regardless of which tool processed them.

What DocLang encodes?

DocLang enables seamless interoperability across AI tools by standardizing document representation.

Semantic and structural information

Documents processed to the DocLang standard carry explicit semantic roles and a defined reading order, giving AI models the structural context they need to interpret content accurately.

Geometric accuracy

DocLang preserves geometric precision for tables and layout-dependent content, ensuring that spatial relationships within a document are not lost during processing.

Governance controls

Embedded governance controls allow downstream systems to enforce policies related to:

• Privacy and data scope
• Model training permissions
• Compliance requirements

Why DocLang matters for AI workflows?

Adopting the DocLang standard delivers measurable benefits across large language model (LLM), RAG, and agentic AI workflows:

• More reliable AI interpretation: Consistent document structure reduces ambiguity for AI models.
• Fewer hallucinations: Preserving structural context minimizes errors caused by missing or misread layout information.
• Lower computational costs: Standardized inputs reduce the processing overhead required to normalize documents across different tools and pipelines.

Frequently asked questions

What's the history of DocLang?

DocLang was developed to address the lack of uniformity across document formats as AI models took on a more critical role in document processing. It provides a consistent framework for structured data interpretation, improving reliability, minimizing errors, and optimizing resource use in large-scale document processing environments.

What is the DocLang format?

DocLang is an open, AI-native XML markup format (using the .dclg or .dclg.xml file extension) designed to make documents machine-readable for large language models. It maps layout, semantics, and geometry directly to tokens, and features one-to-one token alignment, compact table encoding via OTSL, and embedded governance metadata. This structure helps AI models interpret and process information accurately, while ensuring compatibility across a variety of tools and workflows.