Memora: A Harmonic Memory Representation Balancing Abstraction and Specificity
Microsoft
Microsoft Research introduces Memora, an agentic memory system that decouples rich memory content from lightweight retrieval abstractions, achieving state-of-the-art performance on long-context benchmarks while using up to 98% fewer tokens. The system uses primary abstractions and cue anchors to balance detail and efficiency, enabling AI agents to sustain long-term collaboration.
Microsoft Research has unveiled Memora, a scalable memory system designed to overcome the stateless nature of current AI agents. Memora decouples what is stored (rich memory content) from how it is retrieved (lightweight primary abstractions and cue anchors), resolving the tradeoff between specificity and abstraction. It achieves new state-of-the-art results on LoCoMo and LongMemEval benchmarks, outperforming Mem0, RAG, and full-context inference while using up to 98% fewer tokens. The system was published at ICML 2026 and its code is available on GitHub. Memora's design enables agents to sustain long-term collaboration by maintaining detailed, accessible memories across sessions.
- Сокращения
- LLM = Large Language Model — большая языковая модель
- RAG = Retrieval-Augmented Generation — генерация с дополнением извлечением
- ICML = International Conference on Machine Learning — Международная конференция по машинному обучению
Source: Microsoft Research —
original