Researchers at the National University of Singapore developed MRAgent, a framework that abandons the static "retrieve-then-reason" approach to address AI agents' context window limitations and retrieval noise in long-horizon tasks. MRAgent uses a mechanism allowing an agent to dynamically develop its memory based on accumulating evidence, integrating multi-step memory reconstruction into the reasoning process of the large language model (LLM). The framework treats memory as an interactive environment rather than a static database, where the agent uses the backbone LLM’s reasoning abilities to explore multiple candidate retrieval paths across a structured memory graph. At each step, the LLM evaluates intermediate evidence to iteratively optimize its search, inferring new constraints and pruning irrelevant branches. MRAgent significantly reduces token consumption and runtime costs compared to other agentic memory management approaches.