LLM-based deep research agents are largely built on the ReAct framework. This linear design makes it difficult to revisit earlier states, branch into alternative search directions, or maintain global awareness under long contexts, often leading to local optima, redundant exploration, and inefficient search. We propose Re-TRAC, an agentic framework that performs cross-trajectory exploration by generating a structured state representation after each trajectory to summarize evidence, uncertainties, failures, and future plans, and conditioning subsequent trajectories on this state representation. This enables iterative reflection and globally informed planning, reframing research as a progressive process. Empirical results show that Re-TRAC consistently outperforms ReAct by 15-20% on BrowseComp with frontier LLMs. For smaller models, we introduce Re-TRAC-aware supervised fine-tuning, achieving state-of-the-art performance at comparable scales. Notably, Re-TRAC shows a monotonic reduction in tool calls and token usage across rounds, indicating progressively targeted exploration driven by cross-trajectory reflection rather than redundant search.
@article{arxiv.2602.02486,
title = {RE-TRAC: REcursive TRAjectory Compression for Deep Search Agents},
author = {Jialiang Zhu and Gongrui Zhang and Xiaolong Ma and Lin Xu and Miaosen Zhang and Ruiqi Yang and Song Wang and Kai Qiu and Zhirong Wu and Qi Dai and Ruichun Ma and Bei Liu and Yifan Yang and Chong Luo and Zhengyuan Yang and Linjie Li and Lijuan Wang and Weizhu Chen and Xin Geng and Baining Guo},
journal= {arXiv preprint arXiv:2602.02486},
year = {2026}
}