Upload your memories. Train your robots.
Memories Upload turns everyday first-person video into training data for physical AI.
Captured on Luci Pin, indexed by our own Omni Model.
Upload your memories. Train your robots.
Memories Upload turns everyday first-person video into training data for physical AI.
Captured on Luci Pin, indexed by our own Omni Model.
Language models learned from nearly every word on the internet. World models need millions of years of human hands at work. That experience has never been recorded.
Billions of hours of video exist online, but very little is shot from the point of view of the person doing the work. Today's capture rigs, a helmet, a backpack, a tangle of cables, work for a few hundred people, not a few million. So we put the capture device on people, not in labs.
One camera. Five mounts.
The same Pin snaps onto each mount, so one device covers wrist, chest, head, stereo head and fixed third-person capture.
Open Memories Upload and pick a task: dishes, breakfast, a shift in a store or on a site. Wear the Pin and work as you normally would. Footage uploads on its own, and you are paid for every accepted hour.
Join the waitlist
A global effort to capture real-world tasks
Each point of light is a real contribution to the dataset, indexed by our Omni Model and added to the Video Data Lake.
Hours of first-person video
1,000,000+
Pins recording worldwide
5,000+
Every clip lands in the Memories.ai Video Data Lake: quality checked, indexed by our own Omni Model, annotated, and delivered to the labs training world models and robots.
Pipeline
Quality gates
Automated and human checks. We count accepted hours only
Omni Model indexing
Captions and embeddings in one pass, keyed to timestamps
Five annotation layers
Hand action boundaries, action class, dense captions, 3D hand pose, task labels
3D hand pose
10.4 mm wrist error on Meta's HOT3D, vs 16.8 mm for WiLoR
Search moments, not files
Teams and their agents query the lake through the DataLake API
Robotics & world-model labs
Start with a free sample, then a paid pilot, then volume. Specify tasks, scenes and camera setup, and we run the collection for you.
Every scene, every viewpoint
Homes, laundry, kitchens, retail, factories. Wrist, chest, head, stereo head and fixed third person. We are filling the matrix one cell at a time.
Contributors
Paid only for accepted hours: hands in frame, idle time removed. A rolling quality grade sets your rate.