TechCrunch Industry News Memories AI is building the visual memory layer for wearables and robotics; plus, Nvidia’s version of OpenClaw could solve its biggest problem: security
Mar 17, 2026
They discuss building a visual memory layer that indexes and retrieves video memories for wearables and robots. The conversation covers a custom wearable recorder and partnerships to run models on edge hardware. They also break down a new enterprise AI agent platform designed to bring OpenClaw-style agents with stronger security and local control.
AI Snips
Chapters
Transcript
Episode notes
Visual Memory Is The Missing Layer For Physical AI
- Visual memory is essential for physical AI to operate effectively beyond text-based capabilities.
- Memories.ai spun out from work on Meta's Ray-Ban glasses and focuses on indexing and recalling video-recorded memories for wearables and robots.
Founders Started Memories.ai After Ray-Ban Project
- The founders conceived Memories.ai while building Meta's Ray-Ban glasses and realized users needed recall, not just recording.
- They couldn't find an existing visual memory solution, so they spun out of Meta to build one themselves.
Text Memory Progress Doesn't Solve Visual Needs
- Text memory systems like ChatGPT and Gemini are improving but are insufficient for devices that interact via vision.
- Visual data is unstructured and harder to index, so Memories.ai built a large visual memory model launched in July.