MyAG: A Graph-Based Framework for Designing and Analyzing Composable LLM Agent Systems

2026-07-15

AI Summary

This work addresses the lack of flexible, composable, and analyzable design frameworks in current large language model (LLM) agent systems. It proposes a novel construction methodology based on a three-layer graph abstraction: a component graph defines reusable modules, a workflow graph orchestrates task logic, and a search graph enables exploration of reasoning paths. These layers operate synergistically to support recursive composition, strategy decoupling, and runtime monitoring. Notably, this framework is the first to unify hierarchical agent construction and visual analysis within a single graph-based model. Empirical evaluation on representative agent tasks demonstrates its advantages in design flexibility and the trade-off between performance and efficiency. The implementation has been open-sourced to facilitate further research and adoption.

Abstract

We present MyAG, a graph-based framework for designing and analyzing composable LLM agent systems. Our framework separates agent system construction into three graph abstractions: a component graph for agents, environments, and modules; a workflow graph for execution control; and a search graph for runtime execution. This separation allows users to flexibly reuse the same components with different strategies. We further support hierarchical composition through recursive system nodes and provide monitoring and visualization tools for inspecting agent execution. Experiments on representative agent applications show that our framework supports flexible agent system design and helps analyze performance-efficiency tradeoffs. Our framework is publicly available and fully open-source.

Problem

• composable LLM agent systems
• graph-based framework
• agent system design
• performance-efficiency tradeoffs
• hierarchical composition

Innovation

• graph-based framework
• composable LLM agents
• component graph
• workflow graph
• hierarchical composition