Reproduce “Rethinking the Role of Tool Graph Memory in Multi-Turn Tool Use” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Uiji Hwang, Richeng Xuan, Taeuk Kim
The authors compare tool-graph memory with natural-language insight memory for multi-turn tool agents. Graph methods do not reliably outperform memory-free agents, and H-EPM shows statistically reliable gains in only four settings. On ToolSandbox, Mistral-Small-3.2-24B scores 0.545 with Base, 0.603 with H-EPM, and 0.620 with Insight; broader performance beyond evaluated models and tasks remains unestablished.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.