This paper proposes TAPE (Trajectory Alignment and exPerience pool Evolution), a novel self-evolving fine-tuning framework designed to enhance the generalization and robustness of web search agents without requiring large-scale human annotation.
Abstract
Large language model (LLM)-based web search agents have demonstrated remarkable capabilities in autonomous information retrieval and multi-step reasoning. However, their robustness under real-world distribution shifts, such as evolving webpage structures, dynamic content layouts, and unseen task environments, remains a critical open challenge. Existing approaches predominantly rely on static supervised fine-tuning with human-annotated trajectories, which not only incurs substantial labeling cost but also lacks the adaptive capacity to handle the inherent stochasticity of live web environments. In this paper, we propose TAPE (Trajectory Alignment and exPerience pool Evolution), a novel self-evolving fine-tuning framework designed to enhance the generalization and robustness of web search agents without requiring large-scale human annotation. TAPE introduces a dual-stream experience pool that continuously accumulates both successful and failed agent trajectories during self-play execution. A trajectory alignment module maps heterogeneous execution paths into a unified semantic space, enabling contrastive learning to identify failure-inducing operations and reward generalizable search strategies. The framework further incorporates an adaptive pool evolution mechanism that filters, reweights, and distills experiences to prevent knowledge stagnation and distributional collapse. Extensive experiments on the GAIA benchmark and BrowseComp dataset demonstrate that TAPE consistently outperforms strong baselines across all three difficulty levels, achieving up to 6.2% absolute improvement in task success rate on GAIA Level-2 tasks (a 134.8% relative gain over the SFT-Only baseline on the GAIA validation split) and 1.3% absolute (217% relative) improvement on BrowseComp hard-tier queries, while exhibiting significantly greater resilience to webpage structure perturbations. These improvements are consistent across three open base models spanning two model families and the 7B–14B scale range. Our analysis further suggests that the contrastive self-play signal can serve as a useful partial surrogate for human preference labeling in agent trajectory optimization, substantially reducing, though not eliminating, reliance on human-annotated trajectories.
This work introduces SearchOS, a system-level multi-agent framework that turns fragile, implicit search progress into explicit, persistent, and shared state, and introduces a Search Tool Middleware Harness that intercepts model and tool interactions to record grounded evidence and react to stalls or budget exhaustion.
Recently, Diffusion Large Language Models (dLLMs) have demonstrated unique efficiency advantages, enabled by their inherently parallel decoding mechanism and flexible generation paradigm. Meanwhile, despite the rapid advancement of Search Agents, their practical deployment is constrained by a fundamental limitation, termed as 1) Latency Challenge : the serial execution of multi-round reasoning, tool calling, and tool response waiting under the ReAct agent paradigm induces severe end-to-end latency. Intuitively, dLLMs can leverage their distinctive strengths to optimize the operational efficiency of agents under the ReAct agent paradigm. Practically, existing dLLM backbones face the 2) Agent Ability Challenge. That is, existing dLLMs exhibit remarkably weak reasoning and tool-calling capabilities, preventing these advantages from being effectively realized in practice. In this paper, we propose DLLM-Searcher, an optimization framework for dLLM-based Search Agents. To solve the Agent Ability Challenge, we design a two-stage post-training pipeline encompassing Agentic Supervised Fine-Tuning (Agentic SFT) and Agentic Variance-Reduced Preference Optimization (Agentic VRPO), which enhances the backbone dLLM's information seeking and reasoning capabilities. To mitigate the Latency Challenge, we leverage the flexible generation mechanism of dLLMs and propose a novel agent paradigm termed Parallel-Reasoning and Acting (P-ReAct). P-ReAct guides the model to prioritize decoding tool_call instructions, thereby allowing the model to keep thinking while waiting for the tool's return. Experimental results demonstrate that DLLM-Searcher achieves performance comparable to mainstream LLM-based search agents and P-ReAct delivers approximately 15% inference acceleration. Our code is available at https://github.com/bubble65/DLLM-Searcher
Jiahao Zhao, Shaoxuan Xu, ZhongXiang Sun et al.· Annual International ACM SIG...· 0 citations
Continuous self-improvement requires an ever-expanding pool of self-generated, diverse, adaptive goals. For language agents, existing training environment pools (hand-curated, statically synthesized, or frozen-verifier) keep the goal distribution fixed as the learner scales. We introduce SPADE (Self-Play in Adaptive Synthetic Executable Environments), a self-play RL framework in which a single LLM plays two roles: an Environment Designer that writes complete, long-horizon training environments as executable code with an OpenAI Gym-style reset()/step() interface, and a Reasoning Agent that learns to act in them. Each is a stateful, multi-turn environment (state transitions, reward functions, and verification code), so one interface spans reasoning problems and multi-step agentic tool use. The Reasoning Agent's regret is estimated using the gap between its reward with and without privileged hints; in optimizing this regret signal the Environment Designer learns to target environments at the edge of the agent's capabilities while keeping them feasible. Through extensive experimentation, we find several components critical to success: grounding the Environment Designer on documents sampled from a large pretraining corpus, and giving it an accumulated environment memory. Scaling to 30B-parameter models, SPADE improves over the strongest fixed-environment baseline by +5.3 on average across eight held-out math, science, code, and reasoning benchmarks, and lifts the tool-use setting by +5.7 on BFCL-v4 multi-turn and +13.9 on ACEBench-Agent; on the games setting, the margin over the strongest baseline grows with model scale. By making environment design itself a learnable component, SPADE takes a concrete step toward open-ended self-improvement.
Bo Liu, Simon Yu, Yiding Jiang et al.· 2 citations
Large language model (LLM) agents can self-evolve by continually improving from their own accumulated experience. However, existing studies predominantly adopt independent evaluation. Consequently, the behavior of self-evolving agents in realistic streaming settings, where agents adapt to diverse and complex task streams, remains poorly understood. To address this gap, we introduce AgentStream, a unified framework that evaluates self-evolving agents spanning diverse evolution components by organizing agentic benchmarks into a configurable task stream and instantiating the \texttt{Isolated}, \texttt{Sequential}, and \texttt{Interleaved} streaming scenarios at test time, which progressively vary the scope and domain composition of the stream. Over these scenarios, we combinatorially evaluate five representative self-evolving methods across three frontier foundation models, disentangling how model capability, method architecture, and streaming scenario jointly shape self-evolution. Our results show that self-evolution reliability varies across streaming scenarios, the benefit of self-evolution is gated by model capability and non-monotonic in model strength, and no single method dominates across models and scenarios. These findings offer concrete guidance for selecting self-evolving methods across models and streaming scenarios. Overall, we advocate that self-evolving agents should be evaluated under realistic task streams rather than isolated single-task settings.
Dong Yan, Jian Liang, Dapeng Hu et al.· 0 citations
This work proposes that LLM web agents can learn simple environment observations at test time, and introduces trial steps for agents to decompose a complex environment observation into sub-modules, and implements a label-free learning method, Test-Time Environment Decomposition (TTED), to adapt agent behaviors with experience during inference.
Jun-Xuan Li, Zijun Liu, Zi-Yi Huang et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
