$\delta$-mem: Efficient Online Memory for Large Language Models

ArXi:2605.12357v1 Announce Type: new Large language models increasingly need to accumulate and reuse historical information in long-term assistants and agent systems. Simply expanding the context window is costly and often fails to ensure effective context utilization. We propose $\delta$-mem, a lightweight memory mechanism that augments a frozen full-attention backbone with a compact online state of associative memory.