Jev-Mem: System-One/Two-Inspired Memory Architecture Builds Agent Memory 6.6x Faster

omarsar0 · x · 2026-09-26

A new report proposes Jev-Mem, an agentic memory architecture inspired by System-One/System-Two cognition. A lightweight controller makes memory decisions—assigning types and relations on write, and handling routing, retrieval budget, graph traversal, scoring, and stopping on read—so the LLM is only called for final reasoning.

Results: memory construction is 6.6x faster, query latency drops 36.7% (to 0.93s on average), and it scores 0.777 on LoCoMo with an LLM judge, an 11.0% relative improvement over the strongest baseline. Construction takes 158 seconds. A solid weekend read for anyone building agent memory layers.

Original post →

More from coding & agent

coding & agent channel →