Agentic-R: Learning to Retrieve for Agentic Search
Paper: arXiv:2601.11888 · Code: github.com/8421BCD/Agentic-R
Authors: Wenhan Liu, Xinyu Ma, Yutao Zhu, Yuchen Li, Daiting Shi, Dawei Yin, et al. (7 authors total)
Background
Agentic search has emerged as a powerful paradigm in which an LLM agent interleaves multi-step reasoning with on-demand retrieval to solve complex questions. However, how to design a retriever for agentic search remains largely underexplored. Existing search agents typically rely on similarity-based retrievers, yet similar passages are not always useful for the final answer generation.
Key Contributions
- A retriever training framework tailored for agentic search. Unlike retrievers designed for single-turn retrieval-augmented generation (RAG) that rely only on local passage utility, Agentic-R measures passage utility in multi-turn agentic search using both local query-passage relevance and global answer correctness.
- An iterative, bidirectional training strategy. The search agent and the retriever are optimized jointly and iteratively. Instead of being trained once with fixed questions like RAG retrievers, Agentic-R is continuously improved using evolving, higher-quality queries produced by the agent.
- Architecture: cascaded retrieve-rerank-generate remains mainstream, but the agentic paradigm makes retrieval *count and policy* itself learnable.
- Data: high-quality instruction data and interaction logs matter; synthetic data requires care against knowledge leakage and distribution shift.
- Evaluation: the gap between offline metrics and online satisfaction is widening; LLM-as-judge should be cross-validated with human evaluation.
- Deployment: latency, cost, interpretability, and safety are hard constraints that academic benchmarks alone cannot capture.
- A Systematic Framework for Enterprise Knowledge Retrieval
- RAG and Understanding in Vision: A Survey
- Synergizing RAG and Reasoning: A Systematic Review
- AceSearcher: Bootstrapping Reasoning and Search for LLMs via RL
- Agentic Information Retrieval
Results
Extensive experiments on seven single-hop and multi-hop QA benchmarks demonstrate that Agentic-R consistently outperforms strong baselines across different search agents.
Context and Takeaways
Within the broader search/retrieval landscape, this work sits at the intersection of agentic search and large-scale IR systems. It addresses how to redistribute the responsibilities of retrieval, ranking, generation, and tool use in the LLM era — where the new variables are the reasoning budget and action space (whether to retrieve, how many times, and which tools to call). Relevant discussion points for practitioners:
Related Work
Original 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 Agentic-R, consistently outperforms strong baselines across different search agents. Our codes are available at: https://github.com/8421BCD/Agentic-R.
*Note: for exact quantitative results, refer to the tables in the original PDF.*