RE-TRAC: REcursive TRAjectory Compression for Deep Search Agents

TL;DR

Re-TRAC introduces a recursive trajectory compression framework for deep search agents, enabling global awareness, iterative reflection, and more efficient exploration, significantly improving performance over existing methods.

Contribution

It presents a novel recursive framework that enhances deep search agents with structured state representations for better exploration and planning.

Findings

Re-TRAC outperforms ReAct by 15-20% on BrowseComp.

Re-TRAC reduces tool calls and token usage across rounds.

Re-TRAC achieves state-of-the-art results with supervised fine-tuning.

Abstract

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.