EverMemOS: A Self-Organizing Memory Operating System for Structured Long-Horizon Reasoning

TL;DR

EverMemOS introduces an engram-inspired memory operating system for long-horizon LLM agents, modeling memory as a lifecycle with three phases: Episodic Trace Formation, Semantic Consolidation, and Reconstructive Recollection. The MemCell and MemScene abstractions enable stable semantic consolidation and grounded, necessity-driven retrieval, outperforming state-of-the-art memory systems on LoCoMo and LongMemEval. Empirical results show strong gains on multi-hop and temporal reasoning, along with a favorable accuracy-cost trade-off and a notable improvement in profile consistency when a consolidated user profile is added (e.g., PersonaMem v2). The work demonstrates the practical impact of lifecycle-based memory organization for coherent long-term interaction, while acknowledging limitations in modality scope and latency, and suggesting future multimodal extensions and new benchmarks for ultra-long timelines.

Abstract

Large Language Models (LLMs) are increasingly deployed as long-term interactive agents, yet their limited context windows make it difficult to sustain coherent behavior over extended interactions. Existing memory systems often store isolated records and retrieve fragments, limiting their ability to consolidate evolving user states and resolve conflicts. We introduce EverMemOS, a self-organizing memory operating system that implements an engram-inspired lifecycle for computational memory. Episodic Trace Formation converts dialogue streams into MemCells that capture episodic traces, atomic facts, and time-bounded Foresight signals. Semantic Consolidation organizes MemCells into thematic MemScenes, distilling stable semantic structures and updating user profiles. Reconstructive Recollection performs MemScene-guided agentic retrieval to compose the necessary and sufficient context for downstream reasoning. Experiments on LoCoMo and LongMemEval show that EverMemOS achieves state-of-the-art performance on memory-augmented reasoning tasks. We further report a profile study on PersonaMem v2 and qualitative case studies illustrating chat-oriented capabilities such as user profiling and Foresight. Code is available at https://github.com/EverMind-AI/EverMemOS.

Figures (6)

• Figure 1: Comparison of typical fragment-based memory and EverMemOS in an interactive chat scenario.
• Figure 2: The EverMemOS workflow mirrors an engram-inspired memory lifecycle: (1) Episodic Trace Formation segments continuous dialogue into MemCells with episodes, atomic facts, and time-bounded Foresight. (2) Semantic Consolidation organizes MemCells into MemScenes and updates a user profile. (3) Reconstructive Recollection performs MemScene-guided retrieval to compose the necessary and sufficient context.
• Figure 3: Ablation results (overall accuracy) on LoCoMo and LongMemEval.
• Figure 4: Sensitivity analysis of the MemScene count ($N$).
• Figure 5: Performance vs. cost frontier on LoCoMo by varying the retrieved episode count ($K$).