Reinforced Collaboration in Multi-Agent Flow Networks

TL;DR

MANGO is a reinforcement learning framework that optimizes multi-agent collaboration in flow networks, significantly improving task performance and efficiency across multiple benchmarks.

Contribution

The paper introduces MANGO, a novel data-driven approach that refines agent workflows using reinforcement learning and textual gradients, with a skipping mechanism for efficiency.

Findings

Achieves up to 12.8% performance improvement over baselines.

Enhances efficiency by 47.4%.

Generalizes well to unseen domains.

Abstract

Multi-agent systems provide a powerful way to extend large language models (LLMs) by decomposing a complex task into specialized subtasks handled by different agents. However, their performance is often hindered by error propagation, arising from suboptimal workflow design or inaccurate agent outputs, which can propagate through the agent collaboration process and degrade final results. To address the challenges, we present MANGO (Multi-Agent Network Gradient Optimization), a data-driven framework that organizes and refines agent collaboration via a flow network constructed from past successful workflows. MANGO integrates reinforcement learning and textual gradients to jointly optimize workflow paths and agent behaviors, while a skipping mechanism prevents redundant updates to well-optimized agents for improving efficiency. Extensive experiments on seven benchmarks show that MANGO achieves up to 12.8% performance improvement over state-of-the-art baselines, enhances efficiency by 47.4%, and generalizes effectively to unseen domains. Our code and datasets are publicly available at https://github.com/openJiuwen-ai/agent-store/tree/main/community/mango.