Agentic-R: Learning to Retrieve for Agentic Search

Overview

• Paper: Agentic-R: Learning to Retrieve for Agentic Search
• Authors: Wenhan Liu, Xinyu Ma, Yutao Zhu, Yuchen Li, Daiting Shi, Dawei Yin, et al. (7 authors)
• arXiv: https://arxiv.org/abs/2601.11888
• Code: https://github.com/8421BCD/Agentic-R
• Topics: Agentic Search, Retrieval, RAG

Key Points

• Problem: Agentic search—where an agent interleaves multi-step reasoning with on-demand retrieval—has emerged as a powerful paradigm, but retriever design for it remains underexplored. Existing agents use similarity-based retrievers, yet similar passages are not always useful for final answer generation.
• Proposal: A novel retriever training framework tailored for agentic search. Unlike single-turn RAG retrievers that rely only on local passage utility, Agentic-R measures passage utility using both local query-passage relevance and global answer correctness in multi-turn agentic search.
• Iterative training: The search agent and the retriever are optimized bidirectionally and iteratively. Instead of being trained once with fixed questions, the retriever is continuously improved using evolving, higher-quality queries generated by the agent.
• Results: Extensive experiments on seven single-hop and multi-hop QA benchmarks show that Agentic-R consistently outperforms strong baselines across different search agents.

Abstract (verbatim)

> Agentic search has recently emerged as a powerful paradigm, where an agent interleaves multi-step reasoning with on-demand retrieval to solve complex questions. Despite its success, how to design a retriever for agentic search remains largely underexplored. Existing search agents typically rely on similarity-based retrievers, while similar passages are not always useful for final answer generation. In this paper, we propose a novel retriever training framework tailored for agentic search. Unlike retrievers designed for single-turn retrieval-augmented generation (RAG) that only rely on local passage utility, we propose to use both local query-passage relevance and global answer correctness to measure passage utility in a multi-turn agentic search. We further introduce an iterative training strategy, where the search agent and the retriever are optimized bidirectionally and iteratively. Different from RAG retrievers that are only trained once with fixed questions, our retriever is continuously improved using evolving and higher-quality queries from the agent. Extensive experiments on seven single-hop and multi-hop QA benchmarks demonstrate that our retriever, termed \ours{}, consistently outperforms strong baselines across different search agents. Our codes are available at: https://github.com/8421BCD/Agentic-R.

Context and Significance

In classic search stacks, recall provides coverage, re-ranking provides discrimination, and generation presents results. The LLM era adds two new variables: reasoning budget and action space (whether to retrieve, how many times, which tools to call). Agentic search extends external knowledge access from a one-shot retrieval step into an iterative, verifiable, plannable process, shifting evaluation from static nDCG toward task success rate, citation accuracy, and multi-hop reasoning chain completeness.

Implications for Search / Rec / Personalization:

1. Architecture: Cascaded retrieve-rerank-generate remains mainstream, but the agentic paradigm turns "when and how to retrieve" itself into a learnable object. 2. Data: High-quality instruction data and synthetic data help, but must guard against knowledge leakage and distribution shift. 3. Evaluation: The gap between offline metrics and online satisfaction is widening; LLM-as-judge should be cross-validated with human assessment. 4. Deployment: Latency, cost, interpretability, and safety are hard constraints in production—do not optimize academic benchmarks alone.

Limitations

Typical caveats for this line of work include experiment scale limited by GPU budget, benchmarks that may not match real user distributions, English-centric data with unknown cross-lingual generalization, and security risks when agents retrieve from the open web. Quantitative results should be verified against the original PDF tables.

Related Entries

• A Systematic Framework for Enterprise Knowledge Retrieval (arXiv:2512.05411)
• Retrieval Augmented Generation and Understanding in Vision: A Survey (arXiv:2503.18016)
• Synergizing RAG and Reasoning: A Systematic Review (arXiv:2504.15909)
• AceSearcher: Bootstrapping Reasoning and Search for LLMs via RL (arXiv:2509.24193)
• AgentX: Agent-Driven Self-Iteration of Industrial Recommender Systems (arXiv:2606.26859)
• Agentic Information Retrieval (arXiv:2410.09713)

Glossary

| Term | Meaning | |------|---------| | RAG | Retrieval-Augmented Generation | | nDCG | Normalized Discounted Cumulative Gain, a ranking quality metric | | Agentic Search | Modeling search as sequential decision-making and tool invocation by an agent | | Gen-IR | Generative Information Retrieval |