ObjectGraph: From Document Injection to Knowledge Traversal -- A Native File Format for the Agentic Era

Researchers introduce ObjectGraph (.og), a new file format designed specifically for how AI agents consume documents through retrieval rather than linear reading. The format reduces token consumption by up to 95.3% while maintaining task accuracy, addressing a fundamental architectural mismatch between traditional documents and LLM agent workflows.

ObjectGraph represents a paradigm shift in how information structures are designed for artificial intelligence consumption. Traditional document formats—PDFs, Markdown, Word documents—evolved for human readers who process information sequentially. LLM agents operate fundamentally differently, retrieving relevant information from context windows, yet current systems force agents to ingest entire documents inefficiently. This creates cascading problems: wasted tokens on irrelevant content, compounded state in multi-turn loops, and indiscriminate information broadcasting across different agent roles.

The innovation addresses what the researchers frame as a format problem rather than a prompt engineering or retrieval challenge. ObjectGraph functions as a typed, directed knowledge graph rather than linear text, enabling agents to traverse information strategically. Critically, it maintains backward compatibility as a strict Markdown superset, eliminating adoption friction. The format introduces native primitives including the Progressive Disclosure Model, Role-Scoped Access Protocol, and Executable Assertion Nodes—features purpose-built for agentic systems.

Empirical validation demonstrates substantial efficiency gains with up to 95.3% token reduction across five document classes and eight agent task types, with transpiler fidelity reaching 98.7% content preservation. This has immediate implications for developers building AI systems: reduced computational costs, improved agent reasoning through targeted information access, and better role-based permissions for multi-agent frameworks.

As autonomous agents become increasingly central to enterprise and consumer AI infrastructure, format standards that bridge human and machine information consumption will become critical infrastructure. ObjectGraph's success could catalyze broader format standardization, similar to how JSON transformed data interchange.

• →ObjectGraph achieves 95.3% token reduction while maintaining task accuracy, directly improving computational efficiency for LLM agents.
• →The format is backward-compatible with Markdown, requiring no infrastructure beyond a two-primitive query protocol.
• →Role-Scoped Access Protocol enables agents to access information based on assigned roles, improving security and relevance.
• →Traditional document formats were designed for sequential human reading, creating inefficiencies when consumed by retrieval-based agents.
• →Format standardization for agentic systems could become as foundational as JSON was for web development.