Microsoft Research has developed Memora, a memory system designed to provide more scalable and reliable long-term recall for AI agents, addressing the problem of fragmented memory and duplicate information. According to Microsoft, Memora decouples what the AI remembers from how it looks up that information, reducing context token usage by up to 98% while matching or exceeding full-context accuracy, as claimed in a blog post. The system addresses limitations of existing approaches like Mem0, retrieval-augmented generation (RAG), and graph-based systems, which fall into extremes of content-fragmentation or coarse-abstraction. Memora’s architecture stores each memory entry with two components: a primary abstraction, a short phrase of 6–8 words capturing the memory’s essence, and a memory value holding the rich content. New information about an evolving topic is merged into the existing entry under the same primary abstraction. Additionally, cue anchors are short, context-aware tags extracted from each memory’s value, providing alternative access paths and functioning as organically-generated metadata.