Grammar Search for Multi-Agent Systems

2025-12-15

AI Summary

To address the low efficiency, high computational cost, and poor interpretability of large language model (LLM)-driven free-form code search in automated multi-agent system construction, this paper proposes a syntax-driven structured search framework. It formalizes the agent component space via a compositional grammar and enables efficient, modular multi-agent architecture search through deterministic candidate generation and domain-adapted evaluation—specifically tailored for mathematical reasoning and question answering. This work is the first to introduce formal grammatical modeling into multi-agent architecture search, markedly reducing search overhead while enhancing reusability and debuggability. Evaluated on five benchmark tasks, the method achieves state-of-the-art performance on four, demonstrating superior trade-offs among effectiveness, cost-efficiency, and interpretability.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Multiagent Systems: Agent/AI Theories and Architectures
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Automatic search for Multi-Agent Systems has recently emerged as a key focus in agentic AI research. Several prior approaches have relied on LLM-based free-form search over the code space. In this work, we propose a more structured framework that explores the same space through a fixed set of simple, composable components. We show that, despite lacking the generative flexibility of LLMs during the candidate generation stage, our method outperforms prior approaches on four out of five benchmarks across two domains: mathematics and question answering. Furthermore, our method offers additional advantages, including a more cost-efficient search process and the generation of modular, interpretable multi-agent systems with simpler logic.

Problem

• Structured search for multi-agent systems using composable components
• Outperforms LLM-based methods on math and QA benchmarks
• Enables cost-efficient, modular, and interpretable agent system generation

Innovation

• Structured framework with fixed composable components
• Outperforms LLM-based search on most benchmarks
• Generates cost-efficient modular interpretable systems