Survey of 150+ Agent Memory Architectures: Self-Evolving Designs Boost Retention by 50%

blaizedsouza · x · 2026-08-12

A massive 90-page computer science survey maps over 150 agent memory architectures to build self-evolving long-horizon agents.

The paper introduces a 3D taxonomy, demonstrating that action-based memory and self-evolving structures boost long-horizon retention by 50%.

Crucially, the research argues that memory is no longer just a passive database lookup. It functions as an active operating system that trains models to execute tool actions, update parametric weights, and consolidate episodic traces into reusable skills.

Original post →

More from coding & agent

coding & agent channel →