openJiuwen AgentMemory 0.2.0 Adds OmniMem-Lens for Long-Video Memory
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
- Who
- openJiuwen
- What
- released OmniMem-Lens multimodal memory capability in AgentMemory 0.2.0
- When
- Quote
openJiuwen AgentMemory 正式上线了多模态记忆能力 OmniMem-Lens
What happened
openJiuwen AgentMemory 0.2.0 has added OmniMem-Lens, which uses visual and audio cues to organize long-video memories around events and retrieve them at different levels of detail. The article reports higher recall accuracy across three benchmarks, including a 45.7% increase on EgoLife-QA, and says online latency on LVbench fell from 34.5 seconds with Qwen3-VL-8B direct answering to 1.72 seconds with OmniMem-Lens. The project is available at https://gitcode.com/openJiuwen/agent-memory/.
Read next
See the full storyopenJiuwen Adds OmniMem-Lens Multimodal Memory to AgentMemory 0.2.0The story so far
openJiuwen added OmniMem-Lens, a multimodal memory capability for long videos, to AgentMemory 0.2.0. It uses visual and audio cues to organize memories around events and retrieve them at different levels of detail. The report says recall accuracy improved across three benchmarks, including a 45.7% increase on EgoLife-QA. On LVbench, it reports online latency of 1.72 seconds with OmniMem-Lens, compared with 34.5 seconds for direct answering with Qwen3-VL-8B.
Keep up with AI, one email a day
Model releases, price changes and compute developments, with original sources. Free, confirmed by email, and easy to unsubscribe. Read a past issue · 2026-11-04
