How it fits together
The stack keeps improving
Together, the seven layers take an AI project from a conversation to a running solution that keeps improving. The improvement comes from four places.
The solution learns from use
Once a solution is live, it learns from real use and starts its own improvements. The same seven layers plan each improvement, build it, check it and run it, and Maestro tells the owner what changed.
ZB Model can test a change to the solution's instructions, tools or handling of exceptions, check the revised version against the case it missed, and replay the cases that already work.
ZB Model learns from every project
Each finished project teaches it how to plan and route the next one: which models were used, what was checked and what was fixed.
The network grows
Every new model and agent raises what every project can do. A new model joins on the day it is released.
The compute grows
ZB Compute starts with 64 GPUs and expands with the work.
Our research
Our research is how models and agents reason, act and learn together. It goes into ZB Model, the Harness, and the compute infrastructure that lets many models work together at scale. Our mission: AI that improves itself, available to anyone.