LDP: An Identity-Aware Protocol for Multi-Agent LLM Systems

2026-03-11 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, extended

Summary

The LLM Delegate Protocol (LDP) is an AI-native communication protocol designed for multi-agent LLM systems, addressing limitations in existing protocols like Google's A2A and Anthropic's MCP by exposing model-level properties as first-class primitives. LDP introduces five mechanisms: rich delegate identity cards, progressive payload modes with negotiation and fallback, governed sessions for persistent context, structured provenance tracking, and trust domains for security. Implemented as a plugin for the JamJet agent runtime, LDP was empirically studied against A2A and random baselines using local Ollama models and LLM-as-judge evaluation. Key findings include a ~12x lower latency on easy tasks due to identity-aware routing, a 37% reduction in token count with semantic frame payloads, and a 39% token overhead elimination in governed sessions over 10 rounds. Notably, noisy provenance degraded synthesis quality below no-provenance baselines, emphasizing the need for verification.

Key takeaway

For AI Architects designing multi-agent LLM systems, adopting an AI-native protocol like LDP can significantly improve operational efficiency and system governance. Your teams should consider LDP's structured identity, payload negotiation, and governed sessions to reduce latency and token costs, especially for high-volume or multi-round delegation tasks. Be cautious with provenance metadata, ensuring verification mechanisms are in place, as unverified confidence signals can degrade decision quality.

Key insights

AI-native protocols exposing model properties enhance multi-agent LLM system efficiency, governance, and reliability.

Principles

• AI delegates have rich, actionable identity.
• Communication overhead is a first-order cost.
• Governance requires protocol support.

Method

LDP uses identity cards, payload negotiation, governed sessions, structured provenance, and trust domains to enable efficient, governable, and secure multi-agent LLM delegation.

In practice

• Route easy tasks to lightweight models for latency reduction.
• Use semantic frames to reduce token count by 37%.
• Implement governed sessions to eliminate token overhead.

Topics

• LLM Delegate Protocol
• Multi-Agent LLM Systems
• AI-Native Communication
• Protocol Efficiency
• AI Governance & Security

Code references

• sunilp/ldp-protocol
• sunilp/ldp-research
• agent-network-protocol/AgentNetworkProtocol
• a2aproject/A2A
• jamjet-labs/jamjet

Best for: AI Architect, Machine Learning Engineer, AI Scientist, AI Researcher, AI Engineer, MLOps Engineer

Related on AIssential

• The Provenance Paradox in Multi-Agent LLM Routing: Delegation Contracts and Attested Identity in LDP — Unverified self-claimed quality in multi-agent LLM routing leads to systematically selecting the worst delegates.
• A Technical Taxonomy of LLM Agent Communication Protocols — A new taxonomy classifies LLM agent communication protocols, revealing architectural patterns and future trends.
• How AI Agents Use MCPs to Deliver Stronger Results Than Standalone LLMs — AI Agents leveraging Model Context Protocol (MCP) transform LLMs from conversational assistants into action-oriented problem solvers by integrating external systems.
• Adversarial Consensus Protocol (ACP): Validation by Contradiction in Multi-LLM Systems — Real architectural divergence between LLMs, not staged conflict, is key to robust validation.
• Papers You Should Know About — AI research is advancing LLM efficiency and agent autonomy through novel architectures, self-improvement mechanisms, and real-world applications.
• DIY #20 - Routing LLM Agent with LangChain — The Routing pattern efficiently directs LLM queries to specialized handlers, optimizing cost, speed, and safety.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.