AutoAgent: Evolving Cognition and Elastic Memory Orchestration for Adaptive Agents

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.

💡 Research Summary

AutoAgent presents a unified framework that tackles three persistent shortcomings of contemporary autonomous agents: static cognition, rigid workflow dependence, and inefficient context handling. The system is built around three tightly coupled components—Evolving Cognition, On‑the‑fly Contextual Decision‑Making, and Elastic Memory Orchestration—organized in a closed‑loop Self‑Evolution cycle.

Evolving Cognition structures an agent’s knowledge into two complementary facets. Internal Cognition captures functional descriptions of tools and self‑capabilities, while External Cognition models peer agents and environmental dynamics. This knowledge is stored as structured prompt‑level metadata rather than immutable text, allowing the agent to update tool preconditions, skill success rates, and collaborator reliability directly from experience.

The Contextual Decision Engine fuses the current cognition with live task context to select actions from a unified space that includes Emic actions (self‑driven problem solving) and Etic actions (tool calls, inter‑agent requests). Decision making follows an atomic “Select‑Act‑Update” cycle, enabling rapid adaptation when unexpected outcomes arise.

Elastic Memory Orchestration manages interaction histories by preserving raw records, selectively compressing redundant trajectories, and constructing higher‑order episodic abstractions. This dynamic organization reduces token overhead, accelerates reasoning, and ensures that only decision‑critical evidence is retained for future queries.

These components interact through the Self‑Evolution Loop: actions generate experience; the Memory Orchestrator organizes that experience; the Evolution module analyzes it and refines the cognition; the updated cognition then guides subsequent decisions. Crucially, this loop operates without external retraining, allowing continuous, autonomous skill acquisition.

Empirical evaluation spans three domains. On Retrieval‑Augmented Generation benchmarks, AutoAgent’s compressed memory improves answer accuracy by 7 percentage points over static baselines. In Tool‑Augmented Agent benchmarks, it reduces unnecessary tool calls by 15 % while raising task success by 9 %. In embodied multi‑agent environments, collaborative robustness improves, cutting overall mission time by roughly 12 %. Ablation studies confirm that each component contributes uniquely: removing cognition updates leads to rampant tool misuse, while omitting memory compression inflates token usage by more than double.

In summary, AutoAgent demonstrates that integrating evolving structured cognition, real‑time contextual decision making, and elastic memory management yields a self‑improving autonomous agent capable of long‑horizon reasoning, efficient tool utilization, and robust multi‑agent collaboration. Future work will explore meta‑learning for automatic cognition parameter tuning, richer multi‑agent communication protocols, and deployment on physical robotic platforms.