AgentNet: Decentralized Evolutionary Coordination for LLM-based Multi-Agent Systems

TL;DR

AgentNet addresses centralization bottlenecks in LLM-based multi-agent systems by introducing a fully decentralized framework with DAG-based task routing, retrieval-augmented memory, and adaptive evolution. It enables autonomous specialization and dynamic topology reconfiguration, reducing data exposure and improving robustness across tasks. Empirical results show superior task accuracy over single-agent and centralized baselines, with notable gains from decentralized coordination and continual skill refinement. The approach holds practical potential for privacy-preserving, cross-organizational collaboration in large-scale AI ecosystems.

Abstract

The rapid advancement of large language models (LLMs) has enabled the development of multi-agent systems where multiple LLM-based agents collaborate on complex tasks. However, existing systems often rely on centralized coordination, leading to scalability bottlenecks, reduced adaptability, and single points of failure. Privacy and proprietary knowledge concerns further hinder cross-organizational collaboration, resulting in siloed expertise. We propose AgentNet, a decentralized, Retrieval-Augmented Generation (RAG)-based framework that enables LLM-based agents to specialize, evolve, and collaborate autonomously in a dynamically structured Directed Acyclic Graph (DAG). Unlike prior approaches with static roles or centralized control, AgentNet allows agents to adjust connectivity and route tasks based on local expertise and context. AgentNet introduces three key innovations: (1) a fully decentralized coordination mechanism that eliminates the need for a central orchestrator, enhancing robustness and emergent intelligence; (2) dynamic agent graph topology that adapts in real time to task demands, ensuring scalability and resilience; and (3) a retrieval-based memory system for agents that supports continual skill refinement and specialization. By minimizing centralized control and data exchange, AgentNet enables fault-tolerant, privacy-preserving collaboration across organizations. Experiments show that AgentNet achieves higher task accuracy than both single-agent and centralized multi-agent baselines.

Figures (11)

• Figure 1: The illustration contrasts Pre-Defined Multi-Agents (hierarchical, static, with centralized control and single point of failure) against Self-Evolving Agents/AgentNet (adaptive, decentralized, and fault-tolerant with dynamic expertise development).
• Figure 2: Illutration of AgentNet. Initially, agents are fully connected and equipped with executors and routers. The system eliminates the need for a central controller, using a DAG for dynamic task routing and agents leveraging RAG pools and few-shot learning. In the evolved phase, the network adapts with agents developing private trajectories and diversified abilities, ensuring scalability, fault tolerance, and continuous evolution of expertise..
• Figure 3: Dual-role agent architecture.
• Figure 4: Details of Dynamic Task Allocation.
• Figure 5: AgentNet's Router Performance on the BBH (Backbone: gpt-4o-mini)