Memory for machines that act. Coming back, substantially rebuilt.
Embodied is not available at the moment. We have taken it off release while the work goes into NeuraFrame Studio and the Gateway, and it will return as a significantly upgraded product rather than a maintained one. We are not putting a date on that.
The idea is unchanged: put the memory on the machine, not the model. Train with a heavy model where compute is cheap, ship the learned experience as a portable file, and let the machine act from memory in milliseconds, keep learning from corrections on-device, and escalate rather than guess when a situation is new or a safety rule would be broken.
That thinking is not going anywhere. What is changing is the depth of it, and it is worth doing properly rather than shipping the version we had.
The engine underneath Embodied is the same one that powers what you can use right now.
The reuse engine: preserved verified work and corrections so repeated model work is not recomputed, from exact repeats up to things followed across a video. Studio docs
The drop-in: sits in front of any model provider with a one-line change and serves repeats from memory, with freshness windows and pinned answers you teach it. Gateway docs
Code footage against your own standard: your codes, your rules, your grading, one considered model call per real finding. How it works
Send us a message and we will get in touch when Embodied returns. Tell us what you are building and what the machine needs to do; it genuinely shapes what we prioritise.