📄 中文摘要
随着多智能体人工智能系统的复杂性增加，连接它们的协议限制了其能力。现有的协议如A2A和MCP未能将模型级属性作为首要原语，忽视了有效委托所需的基本属性：模型身份、推理特征、质量校准和成本特征。提出了LLM委托协议（LDP），这是一种人工智能原生通信协议，引入了五种机制：（1）包含质量提示和推理特征的丰富委托身份卡；（2）具有协商和回退的渐进式负载模式；（3）具有持久上下文的受控会话；（4）结构化的来源追踪，确保信心和验证状态；（5）强制执行安全边界的信任域。
LDP：一种面向身份的多智能体大语言模型系统协议
出处: LDP: An Identity-Aware Protocol for Multi-Agent LLM Systems
发布: 2026年3月11日
📄 English Summary
LDP: An Identity-Aware Protocol for Multi-Agent LLM Systems
As multi-agent AI systems become increasingly complex, the protocols that connect them limit their capabilities. Current protocols like A2A and MCP fail to expose model-level properties as first-class primitives, overlooking essential attributes for effective delegation: model identity, reasoning profiles, quality calibration, and cost characteristics. The LLM Delegate Protocol (LDP) is introduced as an AI-native communication protocol that incorporates five mechanisms: (1) rich delegate identity cards with quality hints and reasoning profiles; (2) progressive payload modes with negotiation and fallback; (3) governed sessions with persistent context; (4) structured provenance tracking confidence and verification status; (5) trust domains enforcing security boundaries.