Instructor teaches. Agents run production. Class runs on-prem.
The complete teaching stack — scene generation, stream, capture, and delivery — runs on a single on-premises GPU box with no cloud routing in the path. Deterministic, replayable, and no student data leaves the room.
A production agent drives the scene and the live stream through a chat command surface. The instructor never stops teaching to operate software.
A typed command puts an asset or an environment into the live 3D scene mid-lesson. Geometry builds in 1.5 s; texture applies in 1.6 s from cache.
Instructor in mixed reality, one student in XR, the rest on desktop — switching between greenscreen webcam, an external 3D game view of the XR avatar, and the player's own first-person view.
Capture, live markers, captioned cut, delivered. The same workflow that runs the class produces the coursework.
LMS-compatible, full-page LTI launch, vendor-neutral xAPI statements, and zero personally identifying data held on the platform. The learning record is the procurement argument.
Determinism is the closer. A hosted competitor is structurally unable to offer it because their pipeline passes through ephemeral cloud state. This one does not.
Multi-room deployment. Student roster persistence. Shared asset libraries across sessions. Grade passback to LMS. The infrastructure exists; the allocation does not.