NapMem：将记忆重构为智能体的行动空间

// Memory becomes an action space //

Great paper on long-term memory for agents.

(bookmark it)

In short, it's discusses the use of a learned policy for using memory at the right granularity.

Most memory systems still hand the model whatever a retriever selected, leaving it a passive consumer of pre-picked evidence. NapMem reframes memory as a structured action space the agent navigates on its own.

It organizes user history into a multi-granularity pyramid, raw conversations, typed memory records, topic tracks, and user profiles, linked by provenance relations and exposed as tools.

Trained with memory-tool RL, the agent chooses which granularity to inspect before answering.

NapMem stays competitive across PersonaMem-v2, LongMemEval, and LoCoMo while largely preserving general reasoning and tool-use ability.

Paper: https://arxiv.org/abs/2607.05794

Learn to build effective AI agents in our academy: https://academy.dair.ai/