Mem-T: Densifying Rewards for Long-Horizon Memory Agents

Memory agents, which depart from predefined memory-processing pipelines by endogenously managing the processing, storage, and retrieval of memories, have garnered increasing attention for their autonomy and adaptability. However, existing training paradigms remain constrained: agents often traverse long-horizon sequences of memory operations before receiving sparse and delayed rewards, which hinders truly end-to-end optimization of memory management policies. To address this limitation, we introduce Mem-T, an autonomous memory agent that interfaces with a lightweight hierarchical memory database to perform dynamic updates and multi-turn retrieval over streaming inputs. To effectively train long-horizon memory management capabilities, we further propose MoT-GRPO, a tree-guided reinforcement learning framework that transforms sparse terminal feedback into dense, step-wise supervision via memory operation tree backpropagation and hindsight credit assignment, thereby enabling the joint optimization of memory construction and retrieval. Extensive experiments demonstrate that Mem-T is **(1) high-performing**, surpassing frameworks such as A-Mem and Mem0 by up to $14.94\\%$, and **(2) economical**, operating on a favorable accuracy-efficiency Pareto frontier and reducing inference tokens per query by $\sim24.45\\%$ relative to GAM without sacrificing performance.

It is crucial to construct a memory system that encompasses a wide variety of functions and to train it on diverse memory operations. However, the processes of memory construction and retrieval are excessively long-range and sparse, making it difficult for feedback from a single question to reflect the quality of relevant memory construction. To address this, we propose Mem-T, an autonomous memory agent that interfaces with a lightweight hierarchical memory database. Furthermore, we introduce MoT-GRPO, which constructs dense rewards through a Memory Operation Tree (MoT) and identifies critical retrieval operations. By using memory items and evidence as pivots, this approach propagates reasoning feedback to evaluate the quality of the corresponding memory construction operations. Consequently, this achieves optimization for both memory construction and memory retrieval operations within the hierarchical memory systems.