StackPlanner: A Centralized Hierarchical Multi-Agent System with Task-Experience Memory Management

TL;DR

StackPlanner introduces a hierarchical multi-agent system with explicit memory management and experience reuse, significantly improving long-horizon collaboration and task performance in large language model-based multi-agent systems.

Contribution

It proposes StackPlanner, a novel hierarchical framework with active memory control and experience retrieval, addressing memory inefficiency and experience reuse in multi-agent systems.

Findings

Enhanced long-horizon collaboration stability

Effective experience retrieval improves task performance

Outperforms existing multi-agent benchmarks

Abstract

Multi-agent systems based on large language models, particularly centralized architectures, have recently shown strong potential for complex and knowledge-intensive tasks. However, central agents often suffer from unstable long-horizon collaboration due to the lack of memory management, leading to context bloat, error accumulation, and poor cross-task generalization. To address both task-level memory inefficiency and the inability to reuse coordination experience, we propose StackPlanner, a hierarchical multi-agent framework with explicit memory control. StackPlanner addresses these challenges by decoupling high-level coordination from subtask execution with active task-level memory control, and by learning to retrieve and exploit reusable coordination experience via structured experience memory and reinforcement learning. Experiments on multiple deep-search and agent system benchmarks demonstrate the effectiveness of our approach in enabling reliable long-horizon multi-agent collaboration.