Jev-Mem: A System-One-Controlled Memory System for Efficient AI Agents

Summary

Jev-Mem is an agentic memory architecture that separates fast memory control from slower reasoning. It uses three planes: a System-One control plane, a structured multi-relational memory plane, and a System-Two reasoning plane. During memory construction, the lightweight controller assigns memory types and relations instead of relying on an autoregressive LLM. During retrieval, it performs query routing, allocates the retrieval budget, traverses the memory graph, scores candidates, and decides when to stop. System Two is reserved for complex reasoning and answer synthesis, keeping LLM generation off the critical path of ordinary memory operations. On LoCoMo, Jev-Mem achieves an overall LLM-as-a-Judge score of 0.777, an 11.0% relative improvement over the strongest baseline. Memory construction takes 158 seconds, reported as 6.6 times faster than the fastest competing system, while average query latency falls 36.7% to 0.93 seconds.