Knowledge Graphs as Files: An Agent Harness Where Tool Creation and Tool Use Co-Evolve
What it is
This paper proposes Direct Knowledge Graph Interaction (DKGI), a harness where language model agents treat knowledge graphs as files and write Python code to create custom tools for accessing nodes and edges. The agent iteratively writes, executes, and revises code to follow relations and select textual evidence, allowing tool creation and use to co-evolve during the search process.
The problem it solves
Predefined graph interfaces constrain the type of evidence agents can access, limiting their ability to effectively search and reason over complex, text-rich knowledge graphs.
What you could use it for
- Searching large, text-rich knowledge graphs where predefined APIs are too restrictive
- Performing multi-hop reasoning over graph structures by programmatically combining node sets
- Extracting specific textual evidence from graph nodes for LLM assessment
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.