Memories.ai and Qualcomm Partner on On-Device Visual Memory Model

Memories.ai has introduced its Large Visual Memory Model 2.0 (LVMM 2.0), designed to give AI systems on-device visual memory, announced on the company's website. The company also revealed a collaboration with Qualcomm Technologies to run the model natively on Qualcomm processors beginning in 2026.

The LVMM 2.0 converts raw video into structured, searchable memory directly on the device. It encodes frames, compresses data, and creates an index that supports sub-second search. Users can query video content in natural language or with image prompts to locate specific moments quickly. Running the model on Qualcomm hardware reduces latency, eliminates cloud dependency, and keeps data local.

The system integrates video, audio, and images into a unified memory format, enabling consistent performance across smartphones, cameras, and edge devices. Developers will have access to SDKs and reference designs for integrating capture, indexing, and retrieval capabilities. According to the company, as more video is indexed, LVMM 2.0 will improve its recall and precision.

Planned applications include AI-powered video albums, enhanced recall for smart glasses and wearables, real-time analysis for security devices, and improved contextual understanding for robotics systems.