Decoupled Planning and Execution: A Hierarchical Reasoning Framework for Deep Search
Abstract
A hierarchical framework for deep search tasks separates strategic planning from specialized execution, improving answer quality and efficiency over traditional retrieval-augmented generation and agent-based systems.
Complex information needs in real-world search scenarios demand deep reasoning and knowledge synthesis across diverse sources, which traditional retrieval-augmented generation (RAG) pipelines struggle to address effectively. Current reasoning-based approaches suffer from a fundamental limitation: they use a single model to handle both high-level planning and detailed execution, leading to inefficient reasoning and limited scalability. In this paper, we introduce HiRA, a hierarchical framework that separates strategic planning from specialized execution. Our approach decomposes complex search tasks into focused subtasks, assigns each subtask to domain-specific agents equipped with external tools and reasoning capabilities, and coordinates the results through a structured integration mechanism. This separation prevents execution details from disrupting high-level reasoning while enabling the system to leverage specialized expertise for different types of information processing. Experiments on four complex, cross-modal deep search benchmarks demonstrate that HiRA significantly outperforms state-of-the-art RAG and agent-based systems. Our results show improvements in both answer quality and system efficiency, highlighting the effectiveness of decoupled planning and execution for multi-step information seeking tasks. Our code is available at https://github.com/ignorejjj/HiRA.
Community
Introduction
We present HiRA, a hierarchical reasoning framework that separates high-level planning from execution, enabling more efficient and scalable deep search tasks. By integrating specialized expert agents, HiRA enhances both answer quality and system efficiency, outperforming traditional methods and current agent-based systems across multiple benchmarks. Explore the full potential of dynamic, tool-augmented reasoning with our approach.
🎯 Key points: Instead of forcing one model to handle everything, HiRA uses specialized expert agents coordinated by a meta-planner - think of it as having a strategic commander directing domain experts! The results? Shorter reasoning chains, fewer environment interactions, and superior performance across complex multi-modal search tasks.
âš¡ Plug-and-play architecture - existing search agents can be directly integrated without retraining! Works seamlessly with search, code execution, and multimodal capabilities through the innovative Adaptive Reasoning Coordinator with dual-channel memory.
Github Repo: https://github.com/ignorejjj/HiRA
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper