发表机构
Microsoft（微软）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
Stashbird通过显式溯源和分层记忆组织，实现高效说话人索引的代理记忆，显著降低提示词元消耗，在多个长期记忆基准上取得与现有方法相当或更优的准确率。
AI 中文摘要
AI代理需要具备记忆能力，以在用户与代理的交互、用户之间的对话以及有或无代理参与的群组对话中保留信息，同时支持在证据变化或被移除时进行更新。我们提出了Stashbird，一种代理记忆系统，通过显式溯源将源片段与派生记忆状态关联起来。Stashbird将记忆组织为情节记录、语义关系、社区摘要和持久化图状态，并提供支持增量更新和情节级删除的生命周期操作。我们在四个长期记忆基准上评估了问答准确性和面向模型的负载。在LoCoMo上，Stashbird使用的摄入提示词元比Graphiti少76.4倍。与在同一基准上复现的Hindsight相比，其检索提示词元使用量少8.1倍，准确率低1.6个百分点。在LongMemEval-S和GroupMemBench上，Stashbird的准确率高于Hindsight，在EverMemBench上准确率相当。
英文摘要
AI agents require memory that preserves information across user-agent exchanges, user-to-user conversations, and group conversations with or without agent participation, while supporting updates as evidence changes or is removed. We present Stashbird, an agent memory system that links source episodes to derived memory state through explicit provenance. Stashbird organizes memory into episodic records, semantic relations, community summaries, and persisted graph state, with lifecycle operations for incremental updates and episode-level deletion. We evaluate question-answering accuracy and model-facing workload across four long-term memory benchmarks. On LoCoMo, Stashbird uses 76.4x fewer ingestion prompt tokens than Graphiti. Compared with reproduced Hindsight on the same benchmark, it uses 8.1x fewer retrieval prompt tokens, with accuracy 1.6 percentage points lower. It achieves higher accuracy than Hindsight on LongMemEval-S and GroupMemBench and comparable accuracy on EverMemBench.
