Metis: The First Memory Foundation Model Architecture
MemTensor · hf · 2026-07-31
While current AI agents rely on external modules for memory, this paper proposes the concept of Memory Foundation Models to internalize native memory capabilities. Key innovations include:
- Native Memory State: Maintains a persistent, dynamically evolving state within the backbone.
- Native Memory Procedures: Autonomously stores and utilizes information through model computation.
The authors introduce Metis, the first prototype. It compresses historical information into the model, accessed via memory attention. Metis's online memory maintenance is gradient-free, requiring only a forward pass. During inference, model weights are frozen, and memory states are transformed autonomously via standard forward computation.
More from Research
- Meta Proposes OneShot Retrieval Framework, Deployed in Instagram — _reachsumit · 2026-07-31
- Google Proposes HA-MoE Architecture to Boost Cross-Content Ranking Fairness — _reachsumit · 2026-07-31
- Research: Enhancing Generative Recommendation with LLM-Derived Language Tags — _reachsumit · 2026-07-31
- Meta's ROCS Paradigm Boosts Recommendation Retrieval QPS Up to 3x — _reachsumit · 2026-07-31
- With Mandatory Reviewing, Low-Quality Reviews in AI Conferences Are No Longer Justifiable — Kwangryeol · 2026-07-31
- HiLaR: Optimizing LLM Recommendation Reasoning via Hierarchical RL — _reachsumit · 2026-07-31