Memories AI is developing a layer of memory for wearables and robotics

Shawn Shen believes that AI will need to remember what it sees to succeed in the world. Shen’s company Memories.ai is using Nvidia AI tools to build a foundation for wearables and robotics to be able to remember and recall visuals.

Memories.ai announced the partnership with semiconductor giant Nvidia at its GTC conference on Monday. Through this partnership, Memories.ai uses Nvidia’s Cosmos-Reason 2, a model of the language of vision, and Nvidia Metropolis, a video search and summary service, to continue to develop its memory-based capabilities.

Shen (pictured above left) told TechCrunch that he and co-founder and CTO, Ben Zhou (pictured above right), came up with the idea for the company while building the AI ​​system behind Ray-Ban’s Meta sunglasses. The creation of AI glasses led him to imagine how people would use the technology in real life if users could not remember what they were photographing.

He looked around to see if he could find anyone who was already building an AI memory system. When they couldn’t, they decided to get out of Meta and build it themselves.

“AI is doing very well in the digital world. What about the physical?” Shen said. “AI clothes, robotics also need memory. … Ultimately, you need AI to have memory.

The ability of AI systems to remember, for the most part, is new. OpenAI has updated ChatGPT to start remembering past chats in 2024 and improve this in 2025. Elon Musk’s xAI and Google Gemini they have re-launched their commemorative kits for the past two years.

But these advances have focused on text-based memories, Shen said. Text memory is compact and easy to describe but it is not useful for AI-type programs that interact more with the world through vision and visualization.

Memories.ai was launched in 2024 and has raised $16 million to date, through an $8 million seed in July 2025 and an additional $8 million. The round was led by Susa Ventures and included Seedcamp, Fusion Fund, and Crane Venture Partners, among others.

Shen said that successfully creating this memory layer requires two things: building the infrastructure needed to set and target videos in a data format that can be stored and remembered, and capturing the data needed to train the model to do the same.

The company launched its own big visual memory model (LVMM) in July 2025. Shen said it can be compared to a smaller version of Gemini Embedding 2multimodal indexing and retrieving model, which was released earlier this month.

To collect the data, the company developed LUCI, a hardware device worn by the company’s “data collectors” who record the videos used to train the model. Shen said that he does not plan to become a hardware company, or to sell these devices, but, instead, he built them himself because he was not satisfied with video cameras that focused on high-definition videos and battery consumption.

The company released the second generation of this LVMM and signed a partnership with Qualcomm running on Qualcomm processors from the end of this year.

Memories.ai is also working with other major wearable companies already, Shen said, but declined to reveal which ones. Despite the current demand, Shen sees an even bigger opportunity for wearables and robotics to come.

“In terms of marketing, we focus on modeling and architecture, because eventually we think that the market for wearables and robotics will come, but maybe not now,” said Shen.