Enterprise AI. Sovereign Compute. Build entirely in your data center.
Zero-trust AI deployments: Your models, your data, completely air-gapped inside your private data center. Kudoson AI runs its orchestration, retrieval, data and model layers inside the customer's own racks. No inference leaves the premises, no tenant data crosses a public boundary. Move the controls below and the elevation on the left redraws to match.
Size the deployment
Platform design
The same six layers run on any of three substrates. Pick one and every component below shows how it stays up when something underneath it fails. One fixed seam runs through the middle: above it is platform code, below it is settings a customer can change while the platform is running.
What runs on which host
The platform splits into four planes plus a model layer. Orchestration and data services sit on the general-purpose hosts; everything that needs a GPU is pinned to the accelerated pair.
Follow a question through the building
One orchestration graph handles every query. The router classifies intent and re-enters after each agent, so a composite question can cross several lanes before it exits. Pick a question and watch which physical hosts it touches.
Execution trace
How the hosts reach their storage
Two independent fibre-channel fabrics, no shared component between them. Every host has one port on each, so a switch can be taken out for maintenance without pausing a single container.
What the backup appliance is actually holding
Four retention policies stack into the same repository. Deduplication is what makes it fit — turn policies on and off, or drag the reduction ratio, to see how close the sizing runs to the edge.
What this sizing supports
Derived from the current configuration, not from a datasheet. Change the sliders and these move with them.