Memories.ai ✓

Memories.ai is a multimodal video-intelligence infrastructure that gives machines the ability to see, remember, and act on video at scale. Its Large Visual Memory Model turns raw footage into structured, searchable data — indexing every frame, face, scene, and spoken word — then builds a persistent visual memory that organizes people, events, actions, and context across time and improves the more it sees.

On top of that memory sit agents that complete end-to-end tasks: analyzing hours of footage in seconds, flagging safety incidents in real time, auto-tagging and organizing video libraries, and surfacing exact moments.

It's delivered through two APIs shaped around two usage patterns — a Video Intelligence API (captioning, streaming understanding, human re-identification) for getting answers about a piece of footage now, and a Visual Search API (host, index, search at scale) for products that need to remember across years.

It targets media and entertainment, robotics and physical AI, and security and safety, deploys on edge, on-prem, or cloud, and is SOC 2 Type II certified.

The company spun out of Cambridge research with a team of ex-Meta, Google, and Nvidia scientists, publishes actively, and also builds LUCI, an on-device system for human-centric AI training data.

Sold via self-serve API and enterprise sales.