We propose RLAnything, a reinforcement learning framework that dynamically forges environment, policy, and reward models through closed-loop optimization, amplifying learning signals and strengthening the overall RL system for any LLM or agentic scenarios. Specifically, the policy is trained with integrated feedback from step-wise and outcome signals, while the reward model is jointly optimized via consistency feedback, which in turn further improves policy training. Moreover, our theory-motivated automatic environment adaptation improves training for both the reward and policy models by leveraging critic feedback from each, enabling learning from experience. Empirically, each added component consistently improves the overall system, and RLAnything yields substantial gains across various representative LLM and agentic tasks, boosting Qwen3-VL-8B-Thinking by 9.1% on OSWorld and Qwen2.5-7B-Instruct by 18.7% and 11.9% on AlfWorld and LiveBench, respectively. We also that optimized reward-model signals outperform outcomes that rely on human labels. Code: https://github.com/Gen-Verse/Open-AgentRL
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.
AI agents often fail in ways that are difficult to localize because executions are probabilistic, long-horizon, multi-agent, and mediated by noisy tool outputs. We address this gap by manually annotating failed agent runs and release a novel benchmark of 115 failed trajectories spanning structured API workflows, incident management, and open-ended web/file tasks. Each trajectory is annotated with a critical failure step and a category from a grounded-theory derived, cross-domain failure taxonomy. To mitigate the human cost of failure attribution, we present AGENTRX, an automated domain-agnostic diagnostic framework that pinpoints the critical failure step in a failed agent trajectory. It synthesizes constraints, evaluates them step-by-step, and produces an auditable validation log of constraint violations with associated evidence; an LLM-based judge uses this log to localize the critical step and category. Our framework improves step localization and failure attribution over existing baselines across three domains.
Most Large Language Model (LLM) agent memory systems rely on a small set of static, hand-designed operations for extracting memory. These fixed procedures hard-code human priors about what to store and how to revise memory, making them rigid under diverse interaction patterns and inefficient on long histories. To this end, we present \textbf{MemSkill}, which reframes these operations as learnable and evolvable memory skills, structured and reusable routines for extracting, consolidating, and pruning information from interaction traces. Inspired by the design philosophy of agent skills, MemSkill employs a \emph{controller} that learns to select a small set of relevant skills, paired with an LLM-based \emph{executor} that produces skill-guided memories. Beyond learning skill selection, MemSkill introduces a \emph{designer} that periodically reviews hard cases where selected skills yield incorrect or incomplete memories, and evolves the skill set by proposing refinements and new skills. Together, MemSkill forms a closed-loop procedure that improves both the skill-selection policy and the skill set itself. Experiments on LoCoMo, LongMemEval, HotpotQA, and ALFWorld demonstrate that MemSkill improves task performance over strong baselines and generalizes well across settings. Further analyses shed light on how skills evolve, offering insights toward more adaptive, self-evolving memory management for LLM agents.
Representing and understanding 3D environments in a structured manner is crucial for autonomous agents to navigate and reason about their surroundings. While traditional Simultaneous Localization and Mapping (SLAM) methods generate metric reconstructions and can be extended to metric-semantic mapping, they lack a higher level of abstraction and relational reasoning. To address this gap, 3D scene graphs have emerged as a powerful representation for capturing hierarchical structures and object relationships. In this work, we propose an enhanced hierarchical 3D scene graph that integrates open-vocabulary features across multiple abstraction levels and supports object-relational reasoning. Our approach leverages a Vision Language Model (VLM) to infer semantic relationships. Notably, we introduce a task reasoning module that combines Large Language Models (LLM) and a VLM to interpret the scene graph's semantic and relational information, enabling agents to reason about tasks and interact with their environment more intelligently. We validate our method by deploying it on a quadruped robot in multiple environments and tasks, highlighting its ability to reason about them.
As Large Language Models transition to autonomous agents, user inputs frequently violate cooperative assumptions (e.g., implicit intent, missing parameters, false presuppositions, or ambiguous expressions), creating execution risks that text-only evaluations do not capture. Existing benchmarks typically assume well-specified instructions or restrict evaluation to text-only, single-turn clarification, and thus do not measure multi-turn disambiguation under grounded execution risk. We introduce \textbf{Drift-Bench}, the first diagnostic benchmark that evaluates agentic pragmatics under input faults through multi-turn clarification across state-oriented and service-oriented execution environments. Grounded in classical theories of communication, \textbf{Drift-Bench} provides a unified taxonomy of cooperative breakdowns and employs a persona-driven user simulator with the \textbf{Rise} evaluation protocol. Experiments show substantial performance drops under these faults, with clarification effectiveness varying across user personas and fault types. \MethodName bridges clarification research and agent safety evaluation, enabling systematic diagnosis of failures that can lead to unsafe executions.
Large language model (LLM) agents are increasingly equipped with memory, which are stored experience and reusable guidance that can improve task-solving performance. Recent \emph{self-evolving} systems update memory based on interaction outcomes, but most existing evolution pipelines are developed for static train/test splits and only approximate online learning by folding static benchmarks, making them brittle under true distribution shift and continuous feedback. We introduce \textsc{Live-Evo}, an online self-evolving memory system that learns from a stream of incoming data over time. \textsc{Live-Evo} decouples \emph{what happened} from \emph{how to use it} via an Experience Bank and a Meta-Guideline Bank, compiling task-adaptive guidelines from retrieved experiences for each task. To manage memory online, \textsc{Live-Evo} maintains experience weights and updates them from feedback: experiences that consistently help are reinforced and retrieved more often, while misleading or stale experiences are down-weighted and gradually forgotten, analogous to reinforcement and decay in human memory. On the live \textit{Prophet Arena} benchmark over a 10-week horizon, \textsc{Live-Evo} improves Brier score by 20.8\% and increases market returns by 12.9\%, while also transferring to deep-research benchmarks with consistent gains over strong baselines. Our code is available at https://github.com/ag2ai/Live-Evo.
Assessing soft skills such as empathy, ethical judgment, and communication is essential in competitive selection processes, yet human scoring is often inconsistent and biased. While Large Language Models (LLMs) have improved Automated Essay Scoring (AES), we show that state-of-the-art rationale-based fine-tuning methods struggle with the abstract, context-dependent nature of Multiple Mini-Interviews (MMIs), missing the implicit signals embedded in candidate narratives. We introduce a multi-agent prompting framework that breaks down the evaluation process into transcript refinement and criterion-specific scoring. Using 3-shot in-context learning with a large instruct-tuned model, our approach outperforms specialised fine-tuned baselines (Avg QWK 0.62 vs 0.32) and achieves reliability comparable to human experts. We further demonstrate the generalisability of our framework on the ASAP benchmark, where it rivals domain-specific state-of-the-art models without additional training. These findings suggest that for complex, subjective reasoning tasks, structured prompt engineering may offer a scalable alternative to data-intensive fine-tuning, altering how LLMs can be applied to automated assessment.
Multi-Agent Discussion (MAD) has garnered increasing attention very recently, where multiple LLM instances collaboratively solve problems via structured discussion. However, we find that current MAD methods easily suffer from discussion inconsistency, LLMs fail to reach a coherent solution, due to the misalignment between their individual contexts.In this paper, we introduce a multi-LLM context learning method (M2CL) that learns a context generator for each agent, capable of dynamically generating context instructions per discussion round via automatic information organization and refinement. Specifically, inspired by our theoretical insights on the context instruction, M2CL train the generators to control context coherence and output discrepancies via a carefully crafted self-adaptive mechanism.It enables LLMs to avoid premature convergence on majority noise and progressively reach the correct consensus. We evaluate M2CL on challenging tasks, including academic reasoning, embodied tasks, and mobile control. The results show that the performance of M2CL significantly surpasses existing methods by 20%--50%, while enjoying favorable transferability and computational efficiency.
LLM-powered coding agents are redefining how real-world software is developed. To drive the research towards better coding agents, we require challenging benchmarks that can rigorously evaluate the ability of such agents to perform various software engineering tasks. However, popular coding benchmarks such as HumanEval and SWE-Bench focus on narrowly scoped tasks such as competition programming and patch generation. In reality, software engineers have to handle a broader set of tasks for real-world software development. To address this gap, we propose OmniCode, a novel software engineering benchmark that contains a broader and more diverse set of task categories beyond code or patch generation. Overall, OmniCode contains 1794 tasks spanning three programming languages (Python, Java, and C++) and four key categories: bug fixing, test generation, code review fixing, and style fixing. In contrast to prior software engineering benchmarks, the tasks in OmniCode are (1) manually validated to eliminate ill-defined problems, and (2) synthetically crafted or recently curated to avoid data leakage issues, presenting a new framework for synthetically generating diverse software tasks from limited real-world data. We evaluate OmniCode with popular agent frameworks such as SWE-Agent and show that while they may perform well on bug fixing for Python, they fall short on tasks such as Test Generation and in languages such as C++ and Java. For instance, SWE-Agent achieves a maximum of 20.9% with DeepSeek-V3.1 on Java Test Generation tasks. OmniCode aims to serve as a robust benchmark and spur the development of agents that can perform well across different aspects of software development. Code and data are available at https://github.com/seal-research/OmniCode.
The effectiveness of LLM-based agents is often limited not by model capacity alone, but by how efficiently contextual information is utilized at runtime. Existing agent frameworks rely on rigid, syntax-heavy state representations such as nested JSON, which require models to devote a substantial portion of their limited attention to syntactic processing rather than semantic reasoning. In this paper, we propose Fat-Cat, a document-driven agent architecture that improves the signal-to-noise ratio of state management. By integrating three key components: (1) a Semantic File System that represents agent state as Markdown documents aligned with common pre-training corpora, (2) a Textual Strategy Evolution module that accumulates task-solving knowledge without parameter updates, and (3) a Closed-Loop Watcher that monitors reasoning trajectories to reduce hallucinations. Extensive reasoning, retrieval, and coding benchmarks, Fat-Cat consistently improves agent performance. It enables the Kimi-k2 model to outperform the proprietary GPT-4o baseline on HotPotQA. Replacing the document-based state with JSON leads to performance drop, while empirically validating the critical necessity of document-driven state modeling over rigid syntax. The code is available at https://github.com/answeryt/Fat-Cat.
Recent advances in autonomous LLM agents demonstrate their ability to improve performance through iterative interaction with the environment. We define this paradigm as Test-Time Improvement (TTI). However, the mechanisms under how and why TTI succeed or fail remain poorly understood, and existing evaluation metrics fail to capture their task optimization efficiency, behavior adaptation after erroneous actions, and the specific utility of working memory for task completion. To address these gaps, we propose Test-time Improvement Diagnostic Evaluation (TIDE), an agent-agnostic and environment-agnostic framework that decomposes TTI into three comprehensive and interconnected dimensions. The framework measures (1) the overall temporal dynamics of task completion and (2) identifies whether performance is primarily constrained by recursive looping behaviors or (3) by burdensome accumulated memory. Through extensive experiments across diverse agents and environments, TIDE highlights that improving agent performance requires more than scaling internal reasoning, calling for explicitly optimizing the interaction dynamics between the agent and the environment.
Large language models (LLMs) have shown promise in assisting cybersecurity tasks, yet existing approaches struggle with automatic vulnerability discovery and exploitation due to limited interaction, weak execution grounding, and a lack of experience reuse. We propose Co-RedTeam, a security-aware multi-agent framework designed to mirror real-world red-teaming workflows by integrating security-domain knowledge, code-aware analysis, execution-grounded iterative reasoning, and long-term memory. Co-RedTeam decomposes vulnerability analysis into coordinated discovery and exploitation stages, enabling agents to plan, execute, validate, and refine actions based on real execution feedback while learning from prior trajectories. Extensive evaluations on challenging security benchmarks demonstrate that Co-RedTeam consistently outperforms strong baselines across diverse backbone models, achieving over 60% success rate in vulnerability exploitation and over 10% absolute improvement in vulnerability detection. Ablation and iteration studies further confirm the critical role of execution feedback, structured interaction, and memory for building robust and generalizable cybersecurity agents.
Despite the remarkable success that Multi-Agent Code Generation Systems (MACGS) have achieved, the inherent complexity of multi-agent architectures produces substantial volumes of intermediate outputs. To date, the individual importance of these intermediate outputs to the system correctness remains opaque, which impedes targeted optimization of MACGS designs. To address this challenge, we propose CAM, the first \textbf{C}ausality-based \textbf{A}nalysis framework for \textbf{M}ACGS that systematically quantifies the contribution of different intermediate features for system correctness. By comprehensively categorizing intermediate outputs and systematically simulating realistic errors on intermediate features, we identify the important features for system correctness and aggregate their importance rankings. We conduct extensive empirical analysis on the identified importance rankings. Our analysis reveals intriguing findings: first, we uncover context-dependent features\textemdash features whose importance emerges mainly through interactions with other features, revealing that quality assurance for MACGS should incorporate cross-feature consistency checks; second, we reveal that hybrid backend MACGS with different backend LLMs assigned according to their relative strength achieves up to 7.2\% Pass@1 improvement, underscoring hybrid architectures as a promising direction for future MACGS design. We further demonstrate CAM's practical utility through two applications: (1) failure repair which achieves a 73.3\% success rate by optimizing top-3 importance-ranked features and (2) feature pruning that reduces up to 66.8\% intermediate token consumption while maintaining generation performance. Our work provides actionable insights for MACGS design and deployment, establishing causality analysis as a powerful approach for understanding and improving MACGS.
Modern data centers (DCs) hosting artificial intelligence (AI)-dedicated devices operate at high power densities with rapidly varying workloads, making minute-level adaptation essential for safe and energy-efficient operation. However, manually designing piecewise deep reinforcement learning (DRL) agents cannot keep pace with frequent dynamics shifts and service-level agreement (SLA) changes of an evolving DC. This specification-to-policy lag causes a lack of timely, effective control policies, which may lead to service outages. To bridge the gap, we present DCoPilot, a hybrid framework for generative control policies in dynamic DC operation. DCoPilot synergizes two distinct generative paradigms, i.e., a large language model (LLM) that performs symbolic generation of structured reward forms, and a hypernetwork that conducts parametric generation of policy weights. DCoPilot operates through three coordinated phases: (i) simulation scale-up, which stress-tests reward candidates across diverse simulation-ready (SimReady) scenes; (ii) meta policy distillation, where a hypernetwork is trained to output policy weights conditioned on SLA and scene embeddings; and (iii) online adaptation, enabling zero-shot policy generation in response to updated specifications. Evaluated across five control task families spanning diverse DC components, DCoPilot achieves near-zero constraint violations and outperforms all baselines across specification variations. Ablation studies validate the effectiveness of LLM-based unified reward generation in enabling stable hypernetwork convergence.
Tool-using agents based on Large Language Models (LLMs) excel in tasks such as mathematical reasoning and multi-hop question answering. However, in long trajectories, agents often trigger excessive and low-quality tool calls, increasing latency and degrading inference performance, making managing tool-use behavior challenging. In this work, we conduct entropy-based pilot experiments and observe a strong positive correlation between entropy reduction and high-quality tool calls. Building on this finding, we propose using entropy reduction as a supervisory signal and design two reward strategies to address the differing needs of optimizing tool-use behavior. Sparse outcome rewards provide coarse, trajectory-level guidance to improve efficiency, while dense process rewards offer fine-grained supervision to enhance performance. Experiments across diverse domains show that both reward designs improve tool-use behavior: the former reduces tool calls by 72.07% compared to the average of baselines, while the latter improves performance by 22.27%. These results position entropy reduction as a key mechanism for enhancing tool-use behavior, enabling agents to be more adaptive in real-world applications.
The agency expected of Agentic Large Language Models goes beyond answering correctly, requiring autonomy to set goals and decide what to explore. We term this investigatory intelligence, distinguishing it from executional intelligence, which merely completes assigned tasks. Data Science provides a natural testbed, as real-world analysis starts from raw data rather than explicit queries, yet few benchmarks focus on it. To address this, we introduce Deep Data Research (DDR), an open-ended task where LLMs autonomously extract key insights from databases, and DDR-Bench, a large-scale, checklist-based benchmark that enables verifiable evaluation. Results show that while frontier models display emerging agency, long-horizon exploration remains challenging. Our analysis highlights that effective investigatory intelligence depends not only on agent scaffolding or merely scaling, but also on intrinsic strategies of agentic models.
Large language model (LLM)-based agents are increasingly used to perform complex, multi-step workflows in regulated settings such as compliance and due diligence. However, many agentic architectures rely primarily on prompt engineering of a single agent, making it difficult to observe or compare how models handle uncertainty and coordination across interconnected decision stages and with human oversight. We introduce a multi-agent system formalized as a finite-horizon Markov Decision Process (MDP) with a directed acyclic structure. Each agent corresponds to a specific role or decision stage (e.g., content, business, or legal review in a compliance workflow), with predefined transitions representing task escalation or completion. Epistemic uncertainty is quantified at the agent level using Monte Carlo estimation, while system-level uncertainty is captured by the MDP's termination in either an automated labeled state or a human-review state. We illustrate the approach through a case study in AI safety evaluation for self-harm detection, implemented as a multi-agent compliance system. Results demonstrate improvements over a single-agent baseline, including up to a 19\% increase in accuracy, up to an 85x reduction in required human review, and, in some configurations, reduced processing time.
Automatically formulating optimization models from natural language descriptions is a growing focus in operations research, yet current LLM-based approaches struggle with the composite constraints and appropriate modeling paradigms required by complex operational rules. To address this, we introduce the Canonical Intermediate Representation (CIR): a schema that LLMs explicitly generate between problem descriptions and optimization models. CIR encodes the semantics of operational rules through constraint archetypes and candidate modeling paradigms, thereby decoupling rule logic from its mathematical instantiation. Upon a newly generated CIR knowledge base, we develop the rule-to-constraint (R2C) framework, a multi-agent pipeline that parses problem texts, synthesizes CIR implementations by retrieving domain knowledge, and instantiates optimization models. To systematically evaluate rule-to-constraint reasoning, we test R2C on our newly constructed benchmark featuring rich operational rules, and benchmarks from prior work. Extensive experiments show that R2C achieves state-of-the-art accuracy on the proposed benchmark (47.2% Accuracy Rate). On established benchmarks from the literature, R2C delivers highly competitive results, approaching the performance of proprietary models (e.g., GPT-5). Moreover, with a reflection mechanism, R2C achieves further gains and sets new best-reported results on some benchmarks.
Agent memory systems often adopt the standard Retrieval-Augmented Generation (RAG) pipeline, yet its underlying assumptions differ in this setting. RAG targets large, heterogeneous corpora where retrieved passages are diverse, whereas agent memory is a bounded, coherent dialogue stream with highly correlated spans that are often duplicates. Under this shift, fixed top-$k$ similarity retrieval tends to return redundant context, and post-hoc pruning can delete temporally linked prerequisites needed for correct reasoning. We argue retrieval should move beyond similarity matching and instead operate over latent components, following decoupling to aggregation: disentangle memories into semantic components, organise them into a hierarchy, and use this structure to drive retrieval. We propose xMemory, which builds a hierarchy of intact units and maintains a searchable yet faithful high-level node organisation via a sparsity--semantics objective that guides memory split and merge. At inference, xMemory retrieves top-down, selecting a compact, diverse set of themes and semantics for multi-fact queries, and expanding to episodes and raw messages only when it reduces the reader's uncertainty. Experiments on LoCoMo and PerLTQA across the three latest LLMs show consistent gains in answer quality and token efficiency.
Existing Tool-Integrated Reasoning (TIR) models have effectively extended the question-answering capabilities of LLMs by incorporating external tools. However, real-world scenarios present numerous open-ended problems where fixed tools often fail to meet task requirements. Furthermore, the lack of self-optimization mechanisms means that erroneous tool outputs can mislead the LLM's responses. Additionally, the construction of existing tools entails significant manual effort, which consequently constrains their applicability. Recognizing that the reasoning traces of LLMs encapsulate implicit problem-solving capabilities, we propose UCT, a novel training-free framework that transforms agents from tool users to tool creators. This approach harvests reasoning experiences and distills them into reusable assets. This method transforms the agent from a mere tool user into a tool creator, enabling adaptive tool creation and self-updating during the inference process. We also introduce a memory consolidation mechanism to maintain the tool library, ensuring high reusability of retained experiential memory for subsequent reasoning tasks. This novel automated tool construction paradigm continuously improves tool quality during reasoning, allowing the overall agent system to progress without additional training. Extensive experiments demonstrate that our method serves as a novel paradigm for enhancing the capabilities of TIR models. In particular, the significant performance gains achieved +20.86%$\uparrow$ and +23.04%$\uparrow$ on benchmarks across multi-domain mathematical and scientific reasoning tasks validate the self-evolving capability of the agent.
While large language model (LLM) agents have demonstrated impressive problem-solving capabilities, they typically operate as static systems, lacking the ability to evolve through lifelong interaction. Existing attempts to bridge this gap primarily rely on retrieving successful past trajectories as demonstrations. However, this paradigm faces two critical limitations. First, by focusing solely on success, agents overlook the rich pedagogical value embedded in failed attempts, preventing them from identifying and avoiding recurrent pitfalls. Second, continually accumulating textual experiences not only increases the time consumption during retrieval but also inevitably introduces noise and exhausts the largest context window of current LLMs. To address these challenges, we propose a novel self-evolving framework for LLM agents that introduces a complementary evolution mechanism: First, a contrastive reflection strategy is introduced to explicitly summarize error-prone patterns and capture reusable insights. Second, we propose a self-consolidation mechanism that distills non-parametric textual experience into compact learnable parameters. This enables the agent to internalize extensive historical experience directly into its latent space. Extensive experiments demonstrate the advantages of our method in long-term agent evolution.
People navigate complex environments using cues, heuristics, and other strategies, which are often adaptive in stable settings. However, as AI increasingly permeates society's information environments, those become more adaptive and evolving: LLM-based chatbots participate in extended interaction, maintain conversational histories, mirror social cues, and can hypercustomize responses, thereby shaping not only what information is accessed but how questions are framed, how evidence is interpreted, and when action feels warranted. Here we propose a framework for sustained human-AI interaction that rests on invariant features of human cognition and human--AI interaction and centers on three interlinked phenomena: entanglement between users and AI systems, the emergence of cognitive and behavioral drift over repeated interactions, and the role of metacognition in the awareness and regulation of these dynamics. As conversational agents provide cues (e.g., fluency, coherence, responsiveness) that people treat as informative, subjective confidence and action readiness may increase without corresponding gains in epistemic reliability, making drift difficult to detect and correct. We describe these dynamics across micro-, meso-, and macro-levels. The framework identifies four metacognitive intervention points and psychologically informed interventions that provide metacognitive scaffolding (boosting and self-nudging). Finally, we outline a long-horizon research agenda for scientific foresight.
Large Language Models have recently shown impressive capabilities in reasoning and code generation, making them promising tools for natural language interfaces to relational databases. However, existing approaches often fail to generalize in complex, real-world settings due to the highly database-specific nature of SQL reasoning, which requires deep familiarity with unique schemas, ambiguous semantics, and intricate join paths. To address this challenge, we introduce a novel two-stage LLM-based framework that decouples knowledge acquisition from query generation. In the Exploration Stage, the system autonomously constructs a database-specific knowledge base by navigating the schema with a Monte Carlo Tree Search-inspired strategy, generating triplets of schema fragments, executable queries, and natural language descriptions as usage examples. In the Deployment Stage, a dual-agent system leverages the collected knowledge as in-context examples to iteratively retrieve relevant information and generate accurate SQL queries in response to user questions. This design enables the agent to proactively familiarize itself with unseen databases and handle complex, multi-step reasoning. Extensive experiments on large-scale benchmarks demonstrate that our approach significantly improves accuracy over strong baselines, highlighting its effectiveness and generalizability.
Model serving costs dominate AI systems, making compiler optimization essential for scalable deployment. Recent works show that a large language model (LLM) can guide compiler search by reasoning over program structure and optimization history. However, using a single large model throughout the search is expensive, while smaller models are less reliable when used alone. Thus, this paper seeks to answer whether multi-LLM collaborative reasoning relying primarily on small LLMs can match or exceed the performance of a single large model. As such, we propose a lightweight collaborative multi-LLM framework, dubbed COLT, for compiler optimization that enables coordinated reasoning across multiple models within a single Monte Carlo tree search (MCTS) process. A key contribution is the use of a single shared MCTS tree as the collaboration substrate across LLMs, enabling the reuse of transformation prefixes and cross-model value propagation. Hence, we circumvent both heavy internal reasoning mechanisms and conventional agentic machinery that relies on external planners, multiple concurrent LLMs, databases, external memory/versioning of intermediate results, and controllers by simply endogenizing model selection within the lightweight MCTS optimization loop. Every iteration, the acting LLM proposes a joint action: (compiler transformation, model to be queried next). We also introduce a model-aware tree policy that biases search toward smaller models while preserving exploration, and a course-alteration mechanism that escalates to the largest model when the search exhibits persistent regressions attributable to smaller models.
Recent advances in Large Language Models (LLMs) and Vision-Language Models (VLMs) have enabled powerful semantic and multimodal reasoning capabilities, creating new opportunities to enhance sample efficiency, high-level planning, and interpretability in reinforcement learning (RL). While prior work has integrated LLMs and VLMs into various components of RL, the replay buffer, a core component for storing and reusing experiences, remains unexplored. We propose addressing this gap by leveraging VLMs to guide the prioritization of experiences in the replay buffer. Our key idea is to use a frozen, pre-trained VLM (requiring no fine-tuning) as an automated evaluator to identify and prioritize promising sub-trajectories from the agent's experiences. Across scenarios, including game-playing and robotics, spanning both discrete and continuous domains, agents trained with our proposed prioritization method achieve 11-52% higher average success rates and improve sample efficiency by 19-45% compared to previous approaches. https://esharony.me/projects/vlm-rb/
Large Language Models (LLMs) have shown strong potential as conversational agents. Yet, their effectiveness remains limited by deficiencies in robust long-term memory, particularly in complex, long-term web-based services such as online emotional support. However, existing long-term dialogue benchmarks primarily focus on static and explicit fact retrieval, failing to evaluate agents in critical scenarios where user information is dispersed, implicit, and continuously evolving. To address this gap, we introduce ES-MemEval, a comprehensive benchmark that systematically evaluates five core memory capabilities: information extraction, temporal reasoning, conflict detection, abstention, and user modeling, in long-term emotional support settings, covering question answering, summarization, and dialogue generation tasks. To support the benchmark, we also propose EvoEmo, a multi-session dataset for personalized long-term emotional support that captures fragmented, implicit user disclosures and evolving user states. Extensive experiments on open-source long-context, commercial, and retrieval-augmented (RAG) LLMs show that explicit long-term memory is essential for reducing hallucinations and enabling effective personalization. At the same time, RAG improves factual consistency but struggles with temporal dynamics and evolving user states. These findings highlight both the potential and limitations of current paradigms and motivate more robust integration of memory and retrieval for long-term personalized dialogue systems.
LLM-driven agents demonstrate strong performance in sequential decision-making but often rely on on-the-fly reasoning, re-deriving solutions even in recurring scenarios. This insufficient experience reuse leads to computational redundancy and execution instability. To bridge this gap, we propose ProcMEM, a framework that enables agents to autonomously learn procedural memory from interaction experiences without parameter updates. By formalizing a Skill-MDP, ProcMEM transforms passive episodic narratives into executable Skills defined by activation, execution, and termination conditions to ensure executability. To achieve reliable reusability without capability degradation, we introduce Non-Parametric PPO, which leverages semantic gradients for high-quality candidate generation and a PPO Gate for robust Skill verification. Through score-based maintenance, ProcMEM sustains compact, high-quality procedural memory. Experimental results across in-domain, cross-task, and cross-agent scenarios demonstrate that ProcMEM achieves superior reuse rates and significant performance gains with extreme memory compression. Visualized evolutionary trajectories and Skill distributions further reveal how ProcMEM transparently accumulates, refines, and reuses procedural knowledge to facilitate long-term autonomy.
Standard Operating Procedures (SOPs) are essential for ensuring operational safety and consistency in industrial environments. However, retrieving and following these procedures presents unique challenges, such as rigid proprietary structures, condition-dependent relevance, and actionable execution requirement, which standard semantic-driven Retrieval-Augmented Generation (RAG) paradigms fail to address. Inspired by the Mixture-of-Experts (MoE) paradigm, we propose SOPRAG, a novel framework specifically designed to address the above pain points in SOP retrieval. SOPRAG replaces flat chunking with specialized Entity, Causal, and Flow graph experts to resolve industrial structural and logical complexities. To optimize and coordinate these experts, we propose a Procedure Card layer that prunes the search space to eliminate computational noise, and an LLM-Guided gating mechanism that dynamically weights these experts to align retrieval with operator intent. To address the scarcity of domain-specific data, we also introduce an automated, multi-agent workflow for benchmark construction. Extensive experiments across four industrial domains demonstrate that SOPRAG significantly outperforms strong lexical, dense, and graph-based RAG baselines in both retrieval accuracy and response utility, achieving perfect execution scores in real-world critical tasks.
Recent advances in large-scale language models (LLMs) have made multi-agent architectures attractive for challenging reasoning tasks. However, many existing systems rely on stochastic routing or ad-hoc heuristics, making their behavior difficult to reproduce and their decision process hard to interpret. We propose ORCH, a deterministic coordination framework for discrete-choice reasoning that orchestrates heterogeneous LLMs. ORCH follows a ``many analyses, one decision'' paradigm: multiple base models independently produce structured analyses, and a dedicated merge agent outputs the final choice. The framework uses fixed rules for task decomposition and answer aggregation, keeping the pipeline predictable, reproducible, and training-free. Determinism here refers to fixed routing and aggregation rules under a fixed evaluation protocol, rather than strict bit-level reproducibility across deployments. To exploit model complementarity, we optionally introduce an EMA-guided router that updates agent selection using historical accuracy, latency, or cost; since it relies on answer-based feedback, it is mainly intended for benchmarking, controlled evaluation, or delayed-feedback settings. Experiments on MMLU, MMLU-Pro, and GSM8K show that ORCH consistently outperforms single-model baselines and a majority-vote ensemble. On MMLU-Pro, ORCH improves accuracy by over 10 points compared to the strongest baseline, and on GSM8K it yields gains exceeding 50 points; McNemar tests confirm statistical significance. The EMA router provides an additional 0.7--2.0 point accuracy boost, and ablations show that both multi-agent collaboration and routing contribute substantially. Overall, ORCH offers a practical path toward controllable, interpretable, and deployment-ready LLM-based agent systems for discrete-choice reasoning.