RE-TRAC: REcursive TRAjectory Compression for Deep Search Agents
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , , , , , , , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866908806307381248 |
|---|---|
| author | Zhu, Jialiang Zhang, Gongrui Ma, Xiaolong Xu, Lin Zhang, Miaosen Yang, Ruiqi Wang, Song Qiu, Kai Wu, Zhirong Dai, Qi Ma, Ruichun Liu, Bei Yang, Yifan Luo, Chong Yang, Zhengyuan Li, Linjie Wang, Lijuan Chen, Weizhu Geng, Xin Guo, Baining |
| author_facet | Zhu, Jialiang Zhang, Gongrui Ma, Xiaolong Xu, Lin Zhang, Miaosen Yang, Ruiqi Wang, Song Qiu, Kai Wu, Zhirong Dai, Qi Ma, Ruichun Liu, Bei Yang, Yifan Luo, Chong Yang, Zhengyuan Li, Linjie Wang, Lijuan Chen, Weizhu Geng, Xin Guo, Baining |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_02486 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | RE-TRAC: REcursive TRAjectory Compression for Deep Search Agents Zhu, Jialiang Zhang, Gongrui Ma, Xiaolong Xu, Lin Zhang, Miaosen Yang, Ruiqi Wang, Song Qiu, Kai Wu, Zhirong Dai, Qi Ma, Ruichun Liu, Bei Yang, Yifan Luo, Chong Yang, Zhengyuan Li, Linjie Wang, Lijuan Chen, Weizhu Geng, Xin Guo, Baining Computation and Language Artificial Intelligence 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. |
| title | RE-TRAC: REcursive TRAjectory Compression for Deep Search Agents |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2602.02486 |