LLM-X: Scalable Negotiation Protocol for Personal AI Agents
A new scalable negotiation-focused exchange, named LLM-X, has been unveiled by researchers, facilitating structured communication between personal LLM agents that represent individual users. In contrast to existing protocols that center on agent-API interactions, LLM-X offers a message bus and routing framework for LLM-to-LLM coordination, ensuring schema validity and policy enforcement. Its architecture features federated gateways, topic-based routing, and policy enforcement mechanisms. A typed message protocol enables capability negotiation and contract-net-style coordination. The initial large-scale empirical evaluation of LLM-based multi-agent negotiation was performed, involving 5, 9, and 12 agents under varying negotiation policies (Low, Medium, High) and short (minutes) and long (2h, 12h) load conditions. Findings reveal distinct policy-performance trade-offs. The paper can be accessed on arXiv with the identifier 2605.11376.
Key facts
- LLM-X is a scalable negotiation-oriented exchange for personal LLM agents.
- It enables direct, structured communication across populations of personal agents.
- Unlike tool-centric protocols, it focuses on LLM-to-LLM coordination.
- Architecture includes federated gateways, topic-based routing, and policy enforcement.
- Typed message protocol supports capability negotiation and contract-net-style coordination.
- First empirical evaluation of LLM-based multi-agent negotiation at scale.
- Experiments with 5, 9, and 12 agents under Low, Medium, High policies.
- Tests conducted over short-run (minutes) and long-run (2h, 12h) loads.
Entities
Institutions
- arXiv