Memories AI Develops Visual Memory Layer for Wearables and Robotics

Shawn Shen argues that for AI to truly thrive in the physical world, it must possess the capability to remember what it sees. His company, Memories.ai, is leveraging Nvidia's AI tools to create the underlying infrastructure that will enable wearables and robots to store and retrieve visual memories.

On Monday, at the Nvidia GTC conference, Memories.ai announced a partnership with the semiconductor leader. This collaboration grants Memories.ai access to Nvidia’s Cosmos-Reason 2—a reasoning vision language model—and Nvidia Metropolis, an application for video search and summarization, to advance its visual memory technology.

Shen (pictured left) explained to TechCrunch that the idea for the company emerged while he and his co-founder and CTO, Ben Zhou (pictured right), were developing the AI system for Meta’s Ray-Ban glasses. Their work highlighted a key question: what practical use would such technology have if users couldn't later access and recall the video data they captured?

After searching for, but failing to find, any existing solutions for AI visual memory, they made the decision to leave Meta and build the technology themselves.

“AI has already achieved remarkable success in the digital realm. The next frontier is the physical world,” Shen stated. “AI-powered wearables and robots also need memories... Ultimately, we believe a future where AI has visual memory is inevitable.”

The concept of memory for AI systems is a relatively recent development. OpenAI introduced memory features for ChatGPT in 2024, with refinements in 2025. Similarly, Elon Musk’s xAI and Google Gemini have launched their own memory tools within the last two years.

However, Shen notes that these advancements have primarily concentrated on text-based memory. While text is more structured and easier to index, it’s less applicable for physical AI—like robotics and wearables—that primarily perceive and interact with the world through visual input.

Founded in 2024, Memories.ai has secured $16 million in funding to date. This includes an $8 million seed round in July 2025, followed by an $8 million extension. The investment was led by Susa Ventures, with participation from Seedcamp, Fusion Fund, Crane Venture Partners, and others.

According to Shen, building this visual memory layer presented two core challenges: developing the infrastructure to embed and index video into a storable, retrievable data format, and capturing the necessary training data.

The company debuted its large visual memory model (LVMM) in July 2025. Shen describes it as a more compact counterpart to Google's recently released Gemini Embedding 2, a multimodal indexing and retrieval model.

For data collection, the team created LUCI, a custom hardware device worn by "data collectors" to record training videos. Shen clarified that they have no plans to become a hardware manufacturer or sell these devices. They built LUCI because commercial video recorders, often focused on high-definition formats that drain battery life, were unsuitable for their needs.

Memories.ai has since released the second generation of its LVMM and inked a partnership with Qualcomm. The model is slated to run on Qualcomm’s processors starting later this year.

Shen revealed that the company is already collaborating with several major wearable manufacturers, though he declined to name them. While current demand exists, he foresees even greater future opportunities in the wearable and robotics markets.

“From a commercialization standpoint, our focus is squarely on the model and the infrastructure,” Shen said. “We believe the wearables and robotics market will inevitably arrive, but its full emergence is still on the horizon.”