SSR：面向LLM多智能体系统的可扩展Agent发现检索器
原标题：SSR: Structural Signal Retriever for Scalable Agent Discovery in LLM-Orchestrated Multi-Agent Systems
AI 摘要
论文提出 SSR（Structural Signal Retriever），一种确定性、免训练的检索方法，用于 LLM 编排的多智能体系统中 Agent Card 的发现。SSR 从 Agent Card 的能力字段提取结构化词汇信号，用 IDF 加权，并通过倒排信号索引（ISI）进行选择性候选评分。在 1000 个智能体规模下，SSR 相比全目录编排减少 98.8% 的 LLM 输入并降低 70.9% 的编排延迟；在 MS MARCO 上 ISI 减少 96.8% 的显式评分段落且保持检索效果。
正文节选
SSR: Structural Signal Retriever for Scalable Agent Discovery in LLM-Orchestrated Multi-Agent Systems Abstract Large-scale multi-agent ecosystems require mechanisms for discovering suitable agents without exposing entire registries to a large language model (LLM) orchestrator. The challenge is not only to retrieve relevant Agent Cards from a growing registry, but also to maintain a compact candidate context, since each additional candidate increases prompt size, inference cost, latency, and the