UniMem: Self-Routing Episodic-to-Parametric Memory for Boundary-Agnostic LLM Agent Task Streams

Forum topic · 小凯 · 2026-07-30

Summary

UniMem is a self-routing framework for autonomous memory management in LLM agents, proposed by researchers in a 2026 arXiv paper (2607.26017). It addresses the stability-plasticity dilemma that arises when agents operate on boundary-agnostic, evolving task streams. Retrieval-based external memory absorbs new evidence quickly but fails to internalize recurring execution patterns and adds inference-time retrieval overhead, while parametric memory offers stable, efficient execution but typically requires explicit task boundaries and fixed parameter budgets. Inspired by the brain's complementary episodic storage and gradual consolidation, UniMem employs learnable routing tokens as memory controllers to coordinate two complementary pathways: novel or sparse tasks are retained in an episodic buffer for retrieval-augmented execution, while recurring, reliable patterns are consolidated into expandable parametric memory. By decoupling task identification from task execution via routing tokens and parametric memory blocks, UniMem expands memory on demand during deployment without task labels or uncontrolled parameter growth. Experiments on long-horizon streaming task sequences show consistent improvements over baselines across three backbone models, with an average gain of 4.0 EM points while preserving execution fidelity.

Paper Overview

Field: NLP Authors: Siyu Xia, Chenheng Zhang, Yanting Wu, Haoxuan Li, Jiajun Chai, Xiaohan Wang, Guojun Yin, Wei Lin, Zhouchen Lin, Haifeng Zhang, Jun Wang Published: 2026-07-28 arXiv: 2607.26017

Abstract (English)

Memory is essential for LLM agents to accumulate task experience and reuse task-specific execution strategies. However, real-world deployment over boundary-agnostic and evolving task streams exposes a fundamental stability-plasticity dilemma. External retrieval-based memory can rapidly absorb new evidence, but it often fails to internalize recurring execution patterns and incurs inference-time retrieval overhead. Parametric memory enables stable and efficient execution once learned, but typically relies on explicit task boundaries and fixed parameter budgets.

Inspired by the human brain, which balances plasticity and stability through complementary episodic storage and gradual consolidation, the authors propose UniMem, a self-routing framework for autonomous memory management. UniMem uses learnable routing tokens as memory controllers to enable adaptive coordination of complementary memory pathways: novel or sparse tasks are retained in an episodic buffer for retrieval-augmented execution, while recurring and reliable patterns are consolidated into expandable parametric memory.

Key Contributions

Learnable routing tokens act as memory controllers, adaptively routing tasks between episodic retrieval and parametric consolidation.
By decoupling task identification from task execution (via routing tokens and parametric memory blocks), UniMem expands memory on demand during deployment—no task labels required, and no uncontrolled parameter growth.
Experiments on long-horizon streaming task sequences show UniMem consistently outperforms baselines while maintaining execution fidelity, achieving an average improvement of 4.0 EM points across three backbone models.

Reference

arXiv page: https://arxiv.org/abs/2607.26017