TALM: Dynamic Tree-Structured Multi-Agent Framework with Long-Term Memory for Scalable Code Generation

TL;DR

TALM introduces a dynamic, tree-structured multi-agent framework with long-term memory that improves complex code generation by enhancing reasoning flexibility, error correction, and knowledge reuse.

Contribution

It presents a novel framework combining structured task decomposition, localized re-reasoning, and long-term memory for scalable, efficient code generation.

Findings

Outperforms existing models on HumanEval, BigCodeBench, ClassEval benchmarks.

Achieves higher reasoning accuracy and token efficiency.

Demonstrates robustness in complex code generation tasks.

Abstract

Agentic code generation requires large language models (LLMs) capable of complex context management and multi-step reasoning. Prior multi-agent frameworks attempt to address these challenges through collaboration, yet they often suffer from rigid workflows and high reasoning recovery costs. To overcome these limitations, we propose TALM (Tree-Structured Multi-Agent Framework with Long-Term Memory), a dynamic framework that integrates structured task decomposition, localized re-reasoning, and long-term memory mechanisms. TALM employs an extensible tree-based collaboration structure. The parent-child relationships, when combined with a divide-and-conquer strategy, enhance reasoning flexibility and enable efficient error correction across diverse task scopes. Furthermore, a long-term memory module enables semantic querying and integration of prior knowledge, supporting implicit self-improvement through experience reuse. Experimental results on HumanEval, BigCodeBench, and ClassEval benchmarks demonstrate that TALM consistently delivers strong reasoning performance and high token efficiency, highlighting its robustness and practical utility in complex code generation tasks.