Certified Memory Consolidation 可认证的记忆压缩
Every time an AI rewrites its memory, something can quietly disappear. AgentMem is the first memory system with a mathematical warranty — it proves which memories are safe to condense, and refuses to condense the rest. AI 每改写一次自己的记忆，就可能有东西悄悄消失。AgentMem 是第一个自带数学质保书的记忆系统：它能证明哪些记忆可以安全压缩，而证明不了的，它会拒绝压缩。
An Index Built for Robots' Long-Term Memory 为机器人长期记忆设计的索引
A robot's memories live in three worlds at once — where it was, when it happened, and what it meant. TriDex folds space, time and meaning into one joint index tree, so recalling "the cozy coffee shop I passed last spring" takes a single query, not three systems stitched together. 机器人的记忆同时活在三个世界里——在哪里发生、什么时候发生、意味着什么。TriDex 把空间、时间与语义收进同一棵索引树：想起"去年春天路过的那家温馨咖啡店"，只需一次查询，而不是把三套系统缝在一起。
Billion-Scale Vector Search on Hybrid Memory 混合内存上的十亿级向量检索
A graph-based vector search framework built for the DRAM + SSD hierarchy: a sparse in-memory graph steers each query in a few cheap hops, and the full graph on SSD finishes it with a handful of reads. Faster, more accurate, on less memory. 为 DRAM + SSD 层级而生的图式向量检索框架：内存中的稀疏导航图用几次廉价跳转引导查询，SSD 上的完整图再用寥寥数次读取完成收尾。更快、更准、更省内存。