ObjectGraph: A Native File Format for AI Agents
Researchers have introduced ObjectGraph (.og), a file format tailored for autonomous LLM agents rather than for human consumption. This format reimagines documents as typed, directed knowledge graphs, facilitating traversal and eliminating the inefficiencies associated with placing entire documents into context windows. ObjectGraph is a comprehensive superset of Markdown, necessitating only a basic two-primitive query protocol and remains accessible to both agents and humans. The authors define the Document Consumption Problem in their paper, asserting that it stems from format limitations rather than issues related to prompt engineering, retrieval, or compression. This research is available on arXiv with the identifier 2604.27820.
Key facts
- ObjectGraph (.og) is a file format for LLM agents
- It reconceives documents as typed, directed knowledge graphs
- It is a strict superset of Markdown
- Requires no infrastructure beyond a two-primitive query protocol
- Readable by both humans and agents without tooling
- Addresses the Document Consumption Problem
- Published on arXiv: 2604.27820
- Argues the problem is a format problem, not engineering or retrieval
Entities
Institutions
- arXiv