EvoAgentX: An Automated Framework for Evolving Agentic Workflows

Source: arXiv Authors: Yingxu Wang, Siwei Liu, Jinyuan Fang, Zaiqiao Meng Date: 2025-07-04 Primary category: cs.AI All categories: cs.AI

Abstract

EvoAgentX presents a unified framework for generating, executing, evaluating, and evolutionarily optimizing multi-agent workflows. Its evolving layer explicitly refines prompts, tool configurations, and workflow topologies, rather than treating workflow design as static setup. This is close to the practical control-plane problem because it joins workflow execution with the evaluation and promotion loop required for durable improvement.