量子位· 思邈·· 2 days agoSelectedAI score61
openJiuwen AgentMemory 0.2.0 launches OmniMem-Lens multimodal video memory capability
让Agent首次拥有“视听双流记忆”：openJiuwen Agent-Memory看懂长视频，关键事件一个不落
AI guide
openJiuwen AgentMemory 0.2.0 has officially rolled out the OmniMem-Lens multimodal memory capability, leveraging audiovisual dual-stream coordination to segment long videos by real-world events and enable hierarchical retrieval.
Why it matters
Addressing context fragmentation and noise interference caused by fixed-interval chunking in long videos, it introduces audiovisual dual-stream event segmentation and a three-tier retrieval architecture, offering a practical engineering blueprint for multimodal agent memory design.
Source:量子位 · qbitai.com