DKGI: agents treat knowledge graphs as files and co-evolve their own tools, +6.7 Hit@1
yuz9yuz · x · 2026-10-19
- arXiv paper 2610.18616 (extending the DCI work, a NeurIPS'26 Spotlight): DKGI (Direct Knowledge Graph Interaction) drops predefined graph interfaces and instead hands the agent complete node and edge files.
- The agent writes, runs and revises its own Python code to follow relations, combine node sets and select textual evidence, interleaving code execution with shell search and file inspection; tool creation and tool use co-evolve, with execution feedback guiding tool revision.
- Results: training-free DKGI beats the strongest published baseline by at least 6.7 Hit@1 on every graph across three domains; ablations show multi-turn programming outperforms shell-only access and predefined graph tools; trajectories show agents revising code to extend searches or fix errors. Code and data are open-sourced.
More from coding & agent
- x402 agent-payment standard logs 75.4M transactions, $24.2M in 30 days — AiJohnAllen · 2026-10-19
- Dev runs 300 Claude Code agents in parallel with a verification layer — ctjlewis · 2026-10-19
- From BabyAGI to ChatGPT dots: 12 posts that map 4 years of personal agents — heyneighbor · 2026-10-19
- WebMCP directory indexes 38,000+ tools across 6,400+ sites for AI agents — AiJohnAllen · 2026-10-19
- Dev building agent-first GitHub alternative to kill manual merges to main — fforres · 2026-10-19
- A seven-layer knowledge graph stack for a law firm to align model behavior — SnooPeripherals5313 · 2026-10-19