NanoResearch: Co-Evolving Skills, Memory, and Policy for Personalized Research Automation

NanoResearch introduces a multi-agent LLM framework that personalizes research automation through three co-evolving components: a skill bank for reusable procedural knowledge, a memory module for user-specific experience, and label-free policy learning for preference internalization. The system addresses the gap between uniform AI outputs and diverse researcher needs, demonstrating substantial improvements over existing AI research systems while reducing costs across successive cycles.

NanoResearch tackles a critical limitation in current LLM-powered research automation: the one-size-fits-all approach that ignores individual researcher contexts. Different researchers operate with distinct resource constraints, methodological preferences, and output requirements, yet existing systems produce standardized results regardless of these differences. This framework's innovation lies in its tri-level architecture that captures and evolves based on actual usage patterns rather than relying solely on initial training data.

The system's design reflects a broader shift in AI development toward personalization and adaptability. Previous research automation tools focused on task completion without regard for how outputs align with individual workflows or preferences. NanoResearch's skill bank mechanism extracts reusable procedural knowledge from repeated operations, creating a growing repository of researcher-specific optimizations. The memory module grounds planning decisions in historical context, while the label-free policy learning converts informal user feedback into persistent parameter adjustments without requiring explicit formalization of preferences.

For the AI research community and tool developers, this work demonstrates that meaningful personalization in agent systems requires architectural support for continuous learning and memory retention. The co-evolution mechanism—where improved skills generate richer memory, which informs better planning—creates a positive feedback loop that compounds benefits over time. This suggests future research automation tools should prioritize user-specific adaptation as a primary feature rather than an afterthought.

The progressive refinement approach, delivering better outputs at lower computational cost across successive cycles, positions NanoResearch as a template for cost-effective AI systems. Researchers building on this framework should focus on generalizing the personalization mechanisms across different research domains and exploring how skill transfer works between fundamentally different research methodologies.

• →NanoResearch enables personalized research automation through co-evolving skills, memory, and policy components tailored to individual researcher preferences.
• →The skill bank distills recurring operations into reusable procedural rules, reducing redundant work across multiple research projects.
• →Label-free policy learning converts informal user feedback into persistent updates without requiring explicit preference formalization.
• →The system demonstrates substantial performance gains over existing AI research tools while progressively reducing computational costs.
• →The tri-level co-evolution architecture creates a positive feedback loop where each component improves the others over successive cycles.