Small, efficient AI systems meant to live on ordinary consumer devices instead of a data center.
The successor to Astar Technologies
Astar Labs is the spiritual successor to Astar Technologies, but instead of building rockets, I build AI models. Same restlessness, different vehicle. I am no longer trying to put a rocket on a suborbital trajectory. I am trying to make small systems that are actually usable on the machines people already own.
Small models, local devices
Astar Labs's goal is to develop small-scale, efficient AI systems that can be integrated locally onto basically any consumer device, from phones to desktops, and still deliver helpful, accurate answers. If it only works next to a cluster, it is a different project than the one I want to run.
Why local
I like problems with an unforgiving budget. A microcontroller, a rocket, a laptop that has to think for itself: they all force the same habit. Look at the whole system first, then spend the scarce resource on the part that actually matters.
Where it is
The project started in March 2026 and is still early. I would rather publish more here when there is something I would install on my own phone, than pretend a research sketch is a finished product.