Memories.ai partners with Nvidia to build visual memory for wearables

In short: Memories.ai says it is building a “visual memory” system for AI, and it announced a new partnership with Nvidia to help power it.

What happened

Memories.ai is developing what it calls a Large Visual Memory Model, or LVMM. In simple terms, it is software that can turn long videos into something an AI can search, like looking up a moment in a photo library, but for first-person video.

The company says the goal is to help AI systems that live in the real world, like smart glasses and robots. These devices often see more than they can keep in mind at once, so they need a way to store and later recall what happened.

On March 16, 2026, Memories.ai announced a partnership with Nvidia at Nvidia’s GTC conference. The companies plan to use Nvidia tools, including Cosmos Reason 2 (a vision model, which means AI that can interpret images) and Metropolis (software for searching and summarizing video), to improve the “visual memory” infrastructure.

Memories.ai says it launched an initial LVMM in July 2025. It also released LVMM 2.0 in collaboration with Qualcomm, with the aim of running parts of the system directly on Qualcomm processors starting in 2026, including on PCs, wearables, and robots.

Why it matters

Today, many AI assistants mainly work with text, so their “memory” is closer to notes than lived experience. A system like LVMM is meant to help devices remember what they saw, like a person recalling a detail when given a hint (for example, “the red door from earlier”). That could make wearable assistants and robots more useful in everyday tasks, but it also raises practical questions about what gets recorded and how long it is kept.

Source: TechCrunch AI