VoiceMem introduces a dual-brain memory architecture for real-time conversational systems, with a left brain for factual information and a right brain for emotion and persona. It uses a streaming retrieval pipeline that completes in 134 ms, within typical voice activity detection latency. In tests, the left brain achieved top-5 accuracy nearly 30 points higher than Mem0's top-200, and the right brain improved persona benchmark scores by 1.89 points over prior state-of-the-art.
This matters because existing memory systems often add seconds of latency or lack emotional context. VoiceMem bridges that gap, enabling conversational agents to recall precise information and personal emotional history without delaying response. Its decoupled design also allows swapping the underlying memory engine as technology evolves.