AutoAgent: Evolving Cognition and Elastic Memory Orchestration for Adaptive Agents

Xiaoxing Wang; Ning Liao; Shikun Wei; Chen Tang; Feiyu Xiong

cs.AIMar 10, 2026

AutoAgent: Evolving Cognition and Elastic Memory Orchestration for Adaptive Agents

Authors: Xiaoxing Wang, Ning Liao, Shikun Wei, Chen Tang, Feiyu Xiong

Organizations: MemTensor (Shanghai) Technology Co., Ltd. · Institute for Advanced Algorithms Research, Shanghai. · Shanghai Jiao Tong University

Abstract

Autonomous agent frameworks still struggle to reconcile long-term experiential learning with real-time, context-sensitive decision-making. In practice, this gap appears as static cognition, rigid workflow dependence, and inefficient context usage, which jointly limit adaptability in open-ended and non-stationary environments. To address these limitations, we present AutoAgent, a self-evolving multi-agent framework built on three tightly coupled components: evolving cognition, on-the-fly contextual decision-making, and elastic memory orchestration. At the core of AutoAgent, each agent maintains structured prompt-level cognition over tools, self-capabilities, peer expertise, and task knowledge. During execution, this cognition is combined with live task context to select actions from a unified space that includes tool calls, LLM-based generation, and inter-agent requests. To support efficient long-horizon reasoning, an Elastic Memory Orchestrator dynamically organizes interaction history by preserving raw records, compressing redundant trajectories, and constructing reusable episodic abstractions, thereby reducing token overhead while retaining decision-critical evidence. These components are integrated through a closed-loop cognitive evolution process that aligns intended actions with observed outcomes to continuously update cognition and expand reusable skills, without external retraining. Empirical results across retrieval-augmented reasoning, tool-augmented agent benchmarks, and embodied task environments show that AutoAgent consistently improves task success, tool-use efficiency, and collaborative robustness over static and memory-augmented baselines. Overall, AutoAgent provides a unified and practical foundation for adaptive autonomous agents that must learn from experience while making reliable context-aware decisions in dynamic environments.

Figures & tables

Figure 1: The AutoAgent architecture. The Execution Cycle (solid black arrows) handles real-time task progression. The Evolution Cycle (dashed red arrows) drives long-term adaptation. The Elastic Memory Orchestrator couples both cycles by managing experience.

Figure 2: Cognition as the interface between the LLM and the action space. Tool calls, collaboration, and self-driven generations are treated as actions; their descriptions and beliefs are part of cognition and are updated from practice through execution outcomes and memory updates.

Figure 3: The framework of the proposed Elastic Memory Orchestration (EMO) module. At each step of history, action information will be preserved in the Memory Pool in two forms: the complete and lossless raw data, and the summarized action abstracts. At the beginning of each current step, the EMO module first concatenates the action abstracts from all historical steps as input to the selector. The selector determines whether the actions in each round should use the raw information (False), the summarized information (True), or discard the corresponding step’s information (None). After this round of judgment, the historical action information is reorganized accordingly. Subsequently, the selector further decides whether to fully summarize multiple historical steps by executing MemFold, or proceed with other actions such as web search, and so on.

Method	HotpotQA		2Wiki		Bamboogle		Musique		Avg
	Acc	LLM	Acc	LLM	Acc	LLM	Acc	LLM	Acc	LLM
Naive Generation	0.324	0.404	0.348	0.346	0.240	0.280	0.134	0.170	0.2615	0.3000
Standard RAG	0.342	0.450	0.344	0.292	0.272	0.328	0.172	0.188	0.2825	0.3145
SuRe	0.270	0.380	0.244	0.264	0.168	0.208	0.128	0.146	0.2025	0.2495
REPLUG	0.350	0.428	0.296	0.254	0.224	0.256	0.132	0.138	0.2505	0.2690
LongLLMLingua	0.358	0.450	0.324	0.316	0.248	0.288	0.150	0.172	0.2700	0.3065

Table 1: Performance comparison on multi-hop QA benchmarks. The best result in each column is boldfaced ; the second best is underlined . ‘Avg’ denotes the average score across the four datasets.

Model	GAIA				HLE-500			ALFWorld
Model	text	mm	file	all	text	mm	all	Success	Path
gpt-4o
Direct Answer	10.7	0.0	2.6	7.3	2.8	4.4	3.2	–	–
ReAct	26.2	20.8	26.3	25.5	3.4	5.3	3.8	24.6	23.6
DeepAgent	42.5	24.6	31.6	37.6	4.9	5.3	5.0	85.1	90.5
AutoAgent	43.7	25.0	31.6	38.2	5.4	7.1	5.8	85.1	82.8

Table 2: Performance comparison on tool-use and embodied agent benchmarks for closed-source models. For each model backbone, the best result in each column is boldfaced . ‘–’ indicates the baseline is not applicable.

Model	GAIA				HLE-500			ALFWorld
Model	text	mm	file	all	text	mm	all	Success	Path
DeepSeek-R1
Direct Answer	22.3	8.3	0.0	15.2	14.7	2.7	12.0	–	–
ReAct	46.6	29.2	34.2	41.2	15.2	11.5	14.4	94.0	90.3
DeepAgent	29.1	4.2	7.9	20.6	8.3	8.0	8.2	40.3	62.4
AutoAgent	54.4	54.2	44.7	52.1	29.5	23.9	28.2	91.0	87.3

Table 3: Performance comparison on tool-use and embodied agent benchmarks for open-source models. For each model backbone, the best result in each column is boldfaced . ‘–’ indicates the baseline is not applicable.

Benchmarks	GAIA				HLE
Settings	Text	MM	File	All	Text	MM	All
w/o EMO	48.5	41.7	39.5	45.5	19.9	16.8	19.2
w/ EMO	54.4	54.2	44.7	52.1	29.5	23.9	28.2

Table 4: The ablation study on the Elastic Memory Orchestration module.

Action Type	Setting	F1 (%)	EM (%)
Web search Tool	Initial cognition	24.07	22.00
	Evolved cognition	29.67	28.00
Python code generation & execution	Initial cognition	25.55	24.00
	Evolved cognition	37.31	36.00

Table 5: Ablation on evolving cognition under simulated tool instability on the first 50 GAIA questions.

Self-Evolving Agents

Self-evolving agents improve their skills, memory, tools, policies, or environments by learning from interaction and evaluation feedback.

See all 421 papers

Oct 16, 2026

Darwin-Claw: Environment-Adaptive Experience Consolidation for Recursively Self-Evolving Agents

While Large Language Model (LLM) agents have demonstrated remarkable capabilities across complex tasks, they predominantly operate in isolated sessions, lacking the ability to continually learn from historical interactions or adapt to user-specific environments. Existing self-evolution paradigms often indiscriminately inject raw interaction trajectories into memory or skill modules, leading to obscured behavioral patterns and compromised update controllability. Furthermore, as foundational LLM reasoning improves, the bottleneck for agentic evolution shifts from accumulating generic knowledge to effectively managing situational experience. To address this, we propose Darwin-Claw, an environment-adaptive self-evolving framework that systematically distills episodic experiences to drive the comprehensive evolution of operational skills and durable memory. Theoretically grounded in an iterative agent-actor and evolver-critic co-adaptation, Darwin-Claw introduces a Hierarchical Experience Consolidation System that extracts fine-grained error attributions and generalizes them into dynamically updated prototype vectors. This structured situational experience serves as interaction patches for context-sensitive perception and explicit optimization signals for skill refinement. Practically, Darwin-Claw is a readily deployable, automated framework compatible with mainstream agent systems without requiring manual curation. Extensive experiments on WildClawBench demonstrate that Darwin-Claw yields substantial performance improvements, establishing a controllable and stable pathway for the continual, user-centric adaptation of autonomous agents in real-world ecosystems.

Mingze Yin, Xiaohan Wang, Dian Li +4

Zhejiang University · Tencent · Zhongguancun Academy

Self-Evolving Agents Continual Learning for LLM Agents LLM Agent Skill Learning Memory Management for LLM Agents

Ask PDF

Aug 12, 2026

MindMemOS: A Portable and Self-Evolving Memory Operating Layer for AI Agents

Memory is a core component of AI agents, enabling them to accumulate experience, maintain personalization, and adapt over long-term interactions. However, existing memory systems often remain fixed after development, limiting their ability to adapt their memory models, organization strategies, and procedural knowledge through continued use. We present MindMemOS, a portable and self-evolving memory operating layer that organizes open-world information using a unified entity property timestructure. MindMemOS supports scenario-adaptive memory modeling, higher-order pattern discovery, autonomous memory refinement, and continuous skill evolution. Its MindMemEvolve algorithm employs validation-driven evolutionary search to optimize memory schemas for target scenarios, whiledreaming consolidates accumulated memories by merging redundant records and resolving conflicts. In addition, implicit corrective feedback serves as a human-in-the-loop signal for identifying and revising potentially inaccurate or misaligned memories. Its MindSkillEvolve algorithm further transforms agent execution trajectories into reusable and progressively refined skills. MindMemOS achieves 94.03% accuracy on LOCOMO and 70.63% on PersonaMem. MindSkillEvolve improves SpreadsheetBench success by 9.2 percentage points over the initial-skill baseline.

Kaichao Liang, Yuqi Cui, Hao Kong +13

Noah’s Ark Lab, Huawei Technologies

Agent Memory Agent Skill Learning Agent Memory Management Self-Evolving Agents

Ask PDF

Jul 4, 2026

SelfMem: Self-Optimizing Memory for AI Agents

While current AI agents support increasingly long context windows, tool use, and skill execution for long-horizon tasks, they still require memory systems to effectively leverage historical experience. Existing memory frameworks typically rely on fixed storage, retrieval, and summarization mechanisms, which can be rigid across different tasks and often require manual tuning. To address this limitation, we propose SelfMem, a self-optimizing memory framework. Inspired by prior work on self-improving AI, we follow the principle of "teaching an agent to fish rather than giving it a fish." Instead of forcing the model to follow a predefined memory strategy or format, SelfMem provides an environment with memory tools and feedback signals that allow the agent to explore, evaluate, and refine its own memory strategy. Our results show that SelfMem consistently outperforms retrieval, compression, and agent-memory baselines on BEAM across conversation scales from 100K to 1M tokens. Compared with the strongest baseline, SelfMem improves the official score by 48.7%, 40.8%, and 41.9% at 100K, 500K, and 1M, respectively. Further question-type analysis shows broad robustness across diverse memory demands, and our optimization study shows that model-guided strategy refinement further improves performance.

Shu Yang, Junchao Wu, Derek F. Wong +1

1PRADA Lab, King Abdullah University of Science and Technology · NLP2CT Lab, University of Macau

LLM Agent Memory Long-Term Conversational Memory Self-Improving Agents Agent Memory Management

Ask PDF

Abstract

Self-Evolving Agents

Explore similar work

Darwin-Claw: Environment-Adaptive Experience Consolidation for Recursively Self-Evolving Agents

MindMemOS: A Portable and Self-Evolving Memory Operating Layer for AI Agents

SelfMem: Self-Optimizing Memory for AI Agents