$\"main_figure.jpg\"$

Github Repo: https://github.com/ignorejjj/HiRA

\n","updatedAt":"2025-07-04T01:37:46.550Z","author":{"_id":"6695f14df0ffd8e3a379ad61","avatarUrl":"/avatars/5ebb7e55ee9c2d93850b279f440675b0.svg","fullname":"Jiajie Jin","name":"jinjiajie","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":7,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8500083684921265},"editors":["jinjiajie"],"editorAvatarUrls":["/avatars/5ebb7e55ee9c2d93850b279f440675b0.svg"],"reactions":[],"isReport":false}},{"id":"68b88d3ca88c3a27e6c85d94","author":{"_id":"65becfb637491e7adc55ef42","avatarUrl":"/avatars/7ea50a002946be47eb3d12da8df018da.svg","fullname":"Noah Lyons","name":"lyonsno","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2025-09-03T18:47:24.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Tldr?","html":"

Tldr?

\n","updatedAt":"2025-09-03T18:47:24.249Z","author":{"_id":"65becfb637491e7adc55ef42","avatarUrl":"/avatars/7ea50a002946be47eb3d12da8df018da.svg","fullname":"Noah Lyons","name":"lyonsno","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"sv","probability":0.3160574436187744},"editors":["lyonsno"],"editorAvatarUrls":["/avatars/7ea50a002946be47eb3d12da8df018da.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2507.02652","authors":[{"_id":"6867282b9db35afc9c304c83","name":"Jiajie Jin","hidden":false},{"_id":"6867282b9db35afc9c304c84","name":"Xiaoxi Li","hidden":false},{"_id":"6867282b9db35afc9c304c85","user":{"_id":"61cd4b833dd34ba1985e0753","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/61cd4b833dd34ba1985e0753/BfHfrwotoMESpXZOHiIe4.png","isPro":false,"fullname":"KABI","user":"dongguanting","type":"user"},"name":"Guanting Dong","status":"claimed_verified","statusLastChangedAt":"2025-07-29T12:51:21.544Z","hidden":false},{"_id":"6867282b9db35afc9c304c86","name":"Yuyao Zhang","hidden":false},{"_id":"6867282b9db35afc9c304c87","name":"Yutao Zhu","hidden":false},{"_id":"6867282b9db35afc9c304c88","name":"Yang Zhao","hidden":false},{"_id":"6867282b9db35afc9c304c89","name":"Hongjin Qian","hidden":false},{"_id":"6867282b9db35afc9c304c8a","user":{"_id":"66f0bf59e9d50ec57febf751","avatarUrl":"/avatars/be97941e60064e5dd806c6fe9db3c537.svg","isPro":false,"fullname":"Zhicheng Dou","user":"douzc","type":"user"},"name":"Zhicheng Dou","status":"claimed_verified","statusLastChangedAt":"2025-11-06T11:50:49.761Z","hidden":false}],"publishedAt":"2025-07-03T14:18:08.000Z","submittedOnDailyAt":"2025-07-04T00:07:46.543Z","title":"Decoupled Planning and Execution: A Hierarchical Reasoning Framework for\n Deep Search","submittedOnDailyBy":{"_id":"6695f14df0ffd8e3a379ad61","avatarUrl":"/avatars/5ebb7e55ee9c2d93850b279f440675b0.svg","isPro":false,"fullname":"Jiajie Jin","user":"jinjiajie","type":"user"},"summary":"Complex information needs in real-world search scenarios demand deep\nreasoning and knowledge synthesis across diverse sources, which traditional\nretrieval-augmented generation (RAG) pipelines struggle to address effectively.\nCurrent reasoning-based approaches suffer from a fundamental limitation: they\nuse a single model to handle both high-level planning and detailed execution,\nleading to inefficient reasoning and limited scalability. In this paper, we\nintroduce HiRA, a hierarchical framework that separates strategic planning from\nspecialized execution. Our approach decomposes complex search tasks into\nfocused subtasks, assigns each subtask to domain-specific agents equipped with\nexternal tools and reasoning capabilities, and coordinates the results through\na structured integration mechanism. This separation prevents execution details\nfrom disrupting high-level reasoning while enabling the system to leverage\nspecialized expertise for different types of information processing.\nExperiments on four complex, cross-modal deep search benchmarks demonstrate\nthat HiRA significantly outperforms state-of-the-art RAG and agent-based\nsystems. Our results show improvements in both answer quality and system\nefficiency, highlighting the effectiveness of decoupled planning and execution\nfor multi-step information seeking tasks. Our code is available at\nhttps://github.com/ignorejjj/HiRA.","upvotes":26,"discussionId":"6867282c9db35afc9c304c8b","githubRepo":"https://github.com/ignorejjj/HiRA","githubRepoAddedBy":"auto","ai_summary":"A hierarchical framework separates strategic planning from specialized execution in deep search tasks, improving answer quality and system efficiency over traditional RAG and agent-based systems.","ai_keywords":["retrieval-augmented generation","RAG","hierarchical framework","strategic planning","specialized execution","domain-specific agents","external tools","reasoning capabilities","structured integration mechanism","complex search tasks","cross-modal deep search benchmarks","multi-step information seeking tasks"],"githubStars":1},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6695f14df0ffd8e3a379ad61","avatarUrl":"/avatars/5ebb7e55ee9c2d93850b279f440675b0.svg","isPro":false,"fullname":"Jiajie Jin","user":"jinjiajie","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"664dc41d534a4afa606966bf","avatarUrl":"/avatars/c53e2df9dab8e045f413541415594960.svg","isPro":false,"fullname":"Nguyen The Hiep","user":"hiepnguyen3872","type":"user"},{"_id":"66e03eace17fb5ff054b7686","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66e03eace17fb5ff054b7686/PpSV0Qo5lwTyxIZMp57xq.jpeg","isPro":false,"fullname":"Xiaoxi Li","user":"lixiaoxi45","type":"user"},{"_id":"67809011c5273cefdd3c7dcf","avatarUrl":"/avatars/b7e8fa0705d7c3f0728466a78bede2aa.svg","isPro":false,"fullname":"Ethan Jorvik","user":"EthanJorvik","type":"user"},{"_id":"642cd105ad221e8f41d2e669","avatarUrl":"/avatars/831cc1fb4852f2b42469868b326d8195.svg","isPro":false,"fullname":"xiaoxi li","user":"Sunnynexus","type":"user"},{"_id":"61cd4b833dd34ba1985e0753","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/61cd4b833dd34ba1985e0753/BfHfrwotoMESpXZOHiIe4.png","isPro":false,"fullname":"KABI","user":"dongguanting","type":"user"},{"_id":"681354792a6960a1b77c9df7","avatarUrl":"/avatars/a557bcbec425b6deddab7bf61c09b841.svg","isPro":false,"fullname":"Nathaniel Grey","user":"Nathaniel002","type":"user"},{"_id":"684d57f26e04c265777ead3f","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/cuOj-bQqukSZreXgUJlfm.png","isPro":false,"fullname":"Joakim Lee","user":"Reinforcement4All","type":"user"},{"_id":"6347ebfdaae3cb2f9b01b181","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6347ebfdaae3cb2f9b01b181/EbkZaI0drwfsDcOSc7lSf.jpeg","isPro":false,"fullname":"Li","user":"Laurie","type":"user"},{"_id":"648eb1eb59c4e5c87dc116e0","avatarUrl":"/avatars/c636cea39c2c0937f01398c94ead5dad.svg","isPro":false,"fullname":"fdsqefsgergd","user":"T-representer","type":"user"},{"_id":"66d1892971ba7a722e0c0d78","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/pxLH2RFr3TNMP0NVSREMU.jpeg","isPro":false,"fullname":"Arthur EDMOND","user":"Shumatsurontek","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":0}">

Papers

arxiv:2507.02652

Decoupled Planning and Execution: A Hierarchical Reasoning Framework for Deep Search

Published on Jul 3, 2025

· Submitted by

Jiajie Jin on Jul 4, 2025

Upvote

Authors:

Guanting Dong ,

Zhicheng Dou

Abstract

A hierarchical framework separates strategic planning from specialized execution in deep search tasks, improving answer quality and system efficiency over traditional RAG and agent-based systems.

AI-generated summary

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.

View arXiv page View PDF GitHub 1 auto Add to collection

Community

jinjiajie

Paper submitter Jul 4, 2025

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.
⚡ 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.

Framework