Agentic Edge Intelligence: Integrating Large Language Models, Federated Learning, and Edge Computing for Autonomous Decision Systems

Abstract

The convergence of large language models (LLMs), federated learning (FL), and edge computing offers a pathway to autonomous decision systems that can reason near data sources while limiting reliance on centralized infrastructure. Edge-deployed LLMs can support latency-sensitive and privacy-conscious applications, although their computational demands challenge resource-constrained devices. FL enables collaborative adaptation without directly pooling private data, but heterogeneous devices, non-identically distributed data, and communication constraints complicate training. This paper proposes an agentic edge-intelligence framework that combines local LLM inference, federated model adaptation, and multi-agent coordination for sensing, planning, and decision execution. The framework identifies interfaces among reasoning, learning, and edge-resource management, and articulates design considerations for privacy, communication efficiency, and device heterogeneity. It provides a systems-oriented basis for developing and evaluating autonomous edge applications under practical network and resource constraints.