Federated Agent Optimization

2026-10-01

AI Summary

This study addresses the collaboration bottleneck arising from the inability to share cross-organizational experiences among LLM agents under privacy constraints by proposing a federated agent optimization framework. Methodologically, it constructs a unified multi-objective federated optimization space encompassing policies, memory, tools, and knowledge. Through mechanisms of experience abstraction, privacy-preserving aggregation, and transfer adaptation, the framework achieves an optimal trade-off among utility, privacy leakage, and communication costs. The primary contributions lie in enabling the co-evolution of distributed agents without exposing any raw data, while establishing a theoretical paradigm for trustworthy federated agent systems that clarifies key challenges and delineates future research directions.

Abstract

Large language model (LLM) agents increasingly operate in private environments and accumulate valuable experience from task execution, tool use, feedback, and local knowledge. Yet such experience is distributed across organizations and cannot be directly shared because of privacy and proprietary constraints. Conventional federated learning is insufficient for this setting, as agent capabilities extend beyond model parameters to memory, tools, rewards, skills, and structured knowledge. In this paper, we formulate \textbf{Federated Agent Optimization (FAO)}, which studies how distributed agents can collaboratively improve through controlled information exchange while keeping raw data, complete trajectories, and private knowledge local. We define FAO as a multi-objective problem balancing agent utility, privacy leakage, and communication cost, and organize its optimization space across policy, memory, tool use, reward, and structured knowledge and skills. We further characterize how private experience can be abstracted, protected, aggregated, and adapted into transferable capabilities, providing a unified view of how agents can benefit from one another without direct experience sharing. Finally, we identify the key challenges of FAO and outline several promising directions for future research toward trustworthy federated agent systems.

Problem

• Federated Agent Optimization
• Large Language Model Agents
• Privacy Preservation
• Distributed Experience Sharing
• Multi-objective Optimization

Innovation

• Federated Agent Optimization
• Large Language Model Agents
• Multi-objective Optimization
• Privacy-preserving Collaboration
• Transferable Capabilities