ModelsResearch

Microsoft Research introduces Memora: memory system for long-horizon AI agents

Microsoft Research has developed Memora, a memory framework for AI agents that can track long-term projects over weeks or months without forcing a choice between detail preservation and efficient storage.

In detail

  • Memora decouples memory content from retrieval mechanism: information remains rich and expressive (e.g., project timelines, multi-turn discussions), while a separate structural layer handles indexing and retrieval.
  • Solves the core problem of today's LLMs: they are stateless, must re-read entire conversation history in long sessions, and lose either details (when compressed) or structure (when stored as raw text).
  • Enables agents to navigate their own history without re-reading everything – relevant for copilots tracking projects over months or research agents building domain expertise over time.

Why it matters

For mid-market companies deploying AI assistants on complex, long-term projects, a reliable memory system is critical – without it, agents remain ineffective at tracking decision paths, stakeholder preferences, and project history.

For you Watch Memora and similar memory systems: they will become foundational for practical AI copilots in your business – without them, AI agents are not yet production-ready for real project work.

← All news

Summaries are generated automatically and link to the original source.