Reproduce “ACE-Router: Generalizing History-Aware Routing from MCP Tools to the Agent Web” 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
Zhiyuan Yao, Zishan Xu, Yifu Guo, Zhiguang Han, Cheng Yang, Shuo Zhang, Weinan Zhang, Xingshan Zeng and 1 more
Researchers from Shanghai Jiao Tong University and Huawei Technologies developed ACE-Router, a framework that leverages graph-based data synthesis and multi-agent simulation to train a history-aware router for dynamic tool and agent selection in large-scale agent ecosystems. This specialized router achieved superior performance compared to larger generalist LLMs like GPT-4o on real-world benchmarks, exhibiting strong scalability and robustness.
Key insights
- Specialized, history-aware routing models can surpass larger generalist LLMs in precise tool and agent selection, especially in large and dynamic candidate spaces.
- Synthesizing behaviorally diverse tools/agents and context-rich, multi-turn interaction trajectories through graph-based mutation and multi-agent simulation provides effective and reliable supervision for robust router training.
- Incorporating dialogue history is crucial for accurate tool and agent selection, enabling sequential dependency reasoning and error recovery in complex, multi-step tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.