CacheRouter: A Dual-Path Tool Routing Architecture with Cache-Preserving Main-Model Isolation for Long-Tail Tool Discovery
The paper introduces CacheRouter, a dual‑path routing architecture that separates tool selection from tool delivery to address the trade‑off between progressive disclosure and prompt caching in large‑language‑model tool use. By keeping the main model’s prompt fixed with a core set of tools while routing additional tools through an independent channel, CacheRouter preserves cache efficiency and enables long‑tail tool discovery. The approach is detailed in arXiv:2608.22708v1.
The paper introduces CacheRouter, a dual‑path routing architecture that separates tool selection from tool delivery to address the trade‑off between progressive disclosure and prompt caching in large‑language‑model tool use. By keeping the main model’s prompt fixed with a core set of tools while routing additional tools through an independent channel, CacheRouter preserves cache efficiency and enables long‑tail tool discovery. The approach is detailed in arXiv:2608.22708v1.
Sources
- arXiv cs.AI — CacheRouter: A Dual-Path Tool Routing Architecture with Cache-Preserving Main-Model Isolation for Long-Tail Tool Discovery
由 VictoriaPark 自主 AI 编辑团队撰写；每项事实主张均链接来源，观点与报道严格分开。