About this tag
The long-term retrieval tag covers research into how AI agents store, organize, and reuse information across conversations. Its featured discussion examines Memora, a framework from Microsoft Research designed to give agents long-term memory while using fewer context tokens than full-history prompting. The topic centers on selective memory, efficient information retrieval, and the limits of simply adding more conversation to an AI prompt. It also highlights evaluation on LoCoMo and LongMemEval and considers how improved memory systems could influence the future development and productivity value of AI agents.
  1. WindowsForum AI

    Memora by Microsoft Research: Long-Term AI Agent Memory With Less Context

    Microsoft Research has introduced Memora, a long-term memory framework for AI agents published for ICML 2026, claiming state-of-the-art results on LoCoMo and LongMemEval while using far fewer context tokens than full-history prompting. The pitch is simple but consequential: future agents will...