Memories AI Is Giving Robots and Wearables the Power to Remember What They See
A startup founded by two former Meta engineers is solving one of AI's most overlooked problems: the inability to remember the physical world. Memories AI is building the infrastructure that allows AI-powered wearables and robots to store, search, and recall visual memories in real time. If their vision succeeds, the next wave of AI devices will not just see, but actually remember.
 |
| Credit: Memories.ai |
The Problem That Started Everything
Most people think of AI as a purely digital tool. But as smart glasses, humanoid robots, and AI-powered cameras move into daily life, a glaring limitation has emerged. These devices can capture hours of video but cannot meaningfully recall what they recorded five minutes ago, let alone five days ago.
That frustration is exactly what led Shawn Shen and Ben Zhou to build Memories AI. The two co-founders were previously deep in the work of building the AI system behind a major tech company's smart glasses. As they developed the technology, a critical question kept surfacing: what good is a device that records everything if it cannot remember anything? Users could not search, surface, or make sense of the footage being collected.
When Shen and Zhou looked around to see who was already solving this problem, they found nobody. So they made the decision to leave their roles and build the solution themselves.
From Smart Glasses to a New Category of AI Infrastructure
The insight behind Memories AI is deceptively simple but technically complex. In Shen's own words: "AI is already doing really well in the digital world. What about the physical world? AI wearables, robotics need memories as well."
That observation points to a gap that most AI companies have not prioritized. Large language models and generative AI tools have transformed how we interact with text and images on screens. But the physical world moves differently. It is continuous, messy, and context-dependent. A robot navigating a warehouse or a nurse wearing an AI headset needs to recall specific events, recognize recurring patterns, and connect past observations to present actions.
Memories AI is building the layer of infrastructure that makes this possible. Think of it as long-term visual memory for machines — the equivalent of episodic memory in the human brain, but for AI hardware operating in the real world.
A Major Partnership That Signals Industry Confidence
The startup has just announced a significant collaboration that puts them squarely on the map. Memories AI revealed a partnership with semiconductor giant Nvidia at Nvidia's flagship GTC conference, one of the most prominent annual events in the AI and computing industry.
Through this partnership, Memories AI is integrating two of Nvidia's most advanced technologies into its platform. The first is Cosmos-Reason 2, a reasoning vision language model that allows AI systems to not just see but understand and reason about what they observe. The second is Nvidia Metropolis, an application framework built for video search and summarization at scale.
Together, these tools give Memories AI the processing backbone needed to make visual memory retrieval fast, accurate, and practical for real-world deployment. The Nvidia collaboration is not just a technical upgrade. It is a signal that the broader AI hardware ecosystem is beginning to recognize visual memory as a critical missing layer.
Why Visual Memory Is the Missing Link in Physical AI
To understand why this matters, consider how current AI wearables actually work. A pair of smart glasses might record everything you see throughout the day. But without a memory architecture, that footage is essentially a raw, unsearchable file. You cannot ask the device what someone said at your morning meeting, where you left your keys, or what the label on that medicine bottle said last week.
For robotics, the stakes are even higher. A robot working in a complex environment needs to remember spatial layouts, recognize objects it has interacted with before, and build a coherent model of its surroundings over time. Without visual memory, every interaction starts from scratch. That is not just inefficient. In many applications, it is dangerous.
Memories AI is essentially creating the episodic memory layer that current AI hardware is missing. By pairing edge-level capture with intelligent retrieval and reasoning, they are trying to give machines the same kind of contextual awareness that allows humans to navigate daily life with confidence.
The Founders' Edge: They Built It Before
One of the most compelling aspects of the Memories AI story is the specific experience Shen and Zhou bring to the problem. Having worked on the AI system for smart glasses at scale, they understand firsthand how visual data flows from device to cloud, where the friction points are, and what product teams actually need from a memory layer.
That practical, product-level knowledge is rare in a space often dominated by researchers working at a distance from real hardware. Shen and Zhou did not arrive at this problem through a thesis or a market report. They ran into it while building something people would actually use, and they felt the gap sharply enough to dedicate a company to closing it.
Their decision to spin out and build independently also reflects a belief that this infrastructure is too important to be siloed inside a single company. If visual memory for AI is going to scale across the industry — across wearables, robotics, automotive, healthcare, and beyond — it needs to be built as a platform, not a feature.
What Comes Next for AI That Remembers
The implications of Memories AI's work stretch well beyond smart glasses and warehouse robots. As AI becomes embedded in more physical contexts, the ability to remember will become a baseline requirement, not a premium feature.
Imagine a surgical assistant that recalls exactly how a procedure was performed three months ago. A personal AI companion that remembers your preferences, your routines, and the names of people you encounter. An autonomous vehicle that builds a persistent map of the roads it travels most often, getting safer and more efficient with every trip.
Each of these use cases depends on the same foundational capability: the ability to store, index, and retrieve visual experiences in a way that is both fast and meaningful. That is the infrastructure Memories AI is building.
The company is still early, but the Nvidia partnership gives them significant technical credibility and a path to deploying their system at scale. As wearables and robotics move from novelty to necessity, the question of visual memory is going to move from the edges of the conversation to the center of it.
Memories AI is positioning itself to be the answer before most people have even thought to ask the question.
The Physical World Is AI's Next Frontier
The AI revolution to date has largely played out on screens. Chatbots, image generators, coding assistants — these are powerful tools, but they live in the digital domain. The next chapter is different. It is happening in hospitals, factories, homes, and streets. And in that world, intelligence without memory is just a very expensive camera.
What Shawn Shen and Ben Zhou are building at Memories AI is not a feature or a product improvement. It is the infrastructure layer that the entire physical AI ecosystem will eventually need. Their bet is that the race to give machines a sense of place, time, and experience is just beginning — and that the company that solves visual memory will have a foundational role in how AI operates in the world for decades to come.
The startup is still writing its story. But if the Nvidia partnership is any indication, the industry is starting to read along very closely.
