Small Model as Master Orchestrator: Learning Unified Agent-Tool Orchestration with Parallel Subtask Decomposition

TL;DR

This paper introduces a unified parallel orchestration framework for multi-agent systems, enabling flexible, learnable coordination of agents and tools through a lightweight, robust orchestrator trained with supervised fine-tuning and reinforcement learning.

Contribution

It proposes Agent-as-Tool, a standardized paradigm for agent-tool orchestration, and develops ParaManager, a lightweight, state-aware orchestrator trained with a novel two-stage pipeline.

Findings

ParaManager achieves strong performance across multiple benchmarks.

It exhibits robust generalization under unseen model pools.

The approach improves system extensibility and coordination flexibility.

Abstract

Multi-agent systems (MAS) demonstrate clear advantages in tackling complex problems by coordinating diverse agents and external tools. However, most existing orchestration methods rely on static workflows or serial agent scheduling, and are further constrained by heterogeneous interface protocols between tools and agents. This leads to high system complexity and poor extensibility. To mitigate these issues, we propose Agent-as-Tool, a unified parallel orchestration paradigm that abstracts both agents and tools into a standardized, learnable action space with protocol normalization and explicit state feedback. Building on this paradigm, we train a lightweight orchestrator, ParaManager, which decouples planning decisions from subtask solving, enabling state-aware parallel subtask decomposition, delegation, and asynchronous execution. For training, we adopt a two-stage ParaManager training pipeline. It improves robustness by incorporating supervised fine-tuning (SFT) trajectories equipped with recovery mechanisms, and further applies reinforcement learning (RL) to achieve an optimal balance among task success, protocol compliance, diversity, and reasoning efficiency. Experiments show that ParaManager achieves strong performance across multiple benchmarks and exhibits robust generalization under unseen model pools.