A Scalable Communication Protocol for Networks of Large Language Models (2410.11905v1)

Abstract: Communication is a prerequisite for collaboration. When scaling networks of AI-powered agents, communication must be versatile, efficient, and portable. These requisites, which we refer to as the Agent Communication Trilemma, are hard to achieve in large networks of agents. We introduce Agora, a meta protocol that leverages existing communication standards to make LLM-powered agents solve complex problems efficiently. In Agora, agents typically use standardised routines for frequent communications, natural language for rare communications, and LLM-written routines for everything in between. Agora sidesteps the Agent Communication Trilemma and robustly handles changes in interfaces and members, allowing unprecedented scalability with full decentralisation and minimal involvement of human beings. On large Agora networks, we observe the emergence of self-organising, fully automated protocols that achieve complex goals without human intervention.

• The paper introduces Agora, a meta-protocol that resolves the Agent Communication Trilemma by combining structured data and natural language for scalable LLM communication.
• The methodology employs a multi-tiered approach with Protocol Documents to manage diverse agent interactions and significantly reduce API costs.
• Experiments with up to 100 agents demonstrate Agora's ability to lower communication expenses while supporting self-organizing protocols in large-scale networks.

A Scalable Communication Protocol for Networks of LLMs

The paper, "A Scalable Communication Protocol for Networks of LLMs," introduces a protocol named Agora, designed to address the complex communication challenges among large networks of LLM-powered agents. Within this work, the authors tackle what they refer to as the Agent Communication Trilemma, which involves achieving versatility, efficiency, and portability within agent communication systems. Agora is proposed as a meta-protocol that leverages a combination of structured data and natural language to facilitate efficient and scalable communication among heterogeneous LLMs.

Context and Objectives

Human language serves as the foundation for collaboration between agents, yet translating this fluid communication into structured protocols presents inherent challenges. The emergence of LLMs has reinvigorated interest in agent networks capable of complex problem-solving. Previous paradigms, such as rule-based agents, often struggled with adaptability and versatility. LLMs, with their ability to understand natural language and invoke APIs, offer a pathway towards more dynamic interactions.

Agora addresses the communication challenges faced by LLMs categorized into three categories: heterogeneity, general-purpose capabilities, and computational expense. These challenges comprise the Agent Communication Trilemma, where achieving compatibility and efficiency without sacrificing versatility remains difficult.

Methodology

Agora adopts a multi-tiered approach to communication. Frequent interactions are managed through pre-existing, efficient protocols, while less frequent tasks are handled with structured data, leveraging LLM-written routines when feasible. In rare scenarios where routines are infeasible, LLMs use natural language for negotiation and communication. At the core of Agora are Protocol Documents (PDs), which facilitate the sharing and negotiation of communication protocols among agents.

Findings and Implications

The authors conducted experiments with two setups: a small-scale interaction between two agents and a more complex network of 100 agents. In both cases, Agora demonstrated a significant reduction in costs compared to natural language-only communications. The paper also highlights the emergence of self-organizing protocols and behaviors within large-scale agent networks, showcasing the potential for LLMs to autonomously negotiate and optimize communications in real-time.

The reduction in API costs, particularly in the larger network, underscores the efficiency gains achieved by implementing structured communication protocols in place of relying solely on natural language. These findings illustrate Agora's scalability and its potential to facilitate a decentralized and automated network of agents that can perform complex tasks without extensive human intervention.

Future Directions

As computational power continues to grow and LLMs become increasingly sophisticated, the demand for scalable communication strategies across diverse sectors will undoubtedly increase. Agora provides a foundation for developing higher-order communication protocols and frameworks that support decentralized, efficient, and adaptive agent networks. Future advancements may explore enhancing Agora's capabilities through exploring more emergent behavior in agentic-LLM settings or integrating additional machine learning techniques to further optimize communication and processing efforts.

The work encapsulates a step towards a future where LLM-powered networks could automate complex tasks, pushing the boundaries of multi-agent interactions and collaboration. By overcoming the Agent Communication Trilemma, Agora paves the way for more resilient, flexible, and efficient AI-driven ecosystems.