Project UmbraRun AI where it belongs.
An early Moonwind research project for deciding where AI inference should run: on-device, on private infrastructure, or in the cloud.
Moonwind Labs / early research
Not publicly available
Status
Early research
Scope
Local and remote inference
Availability
Not public yet
One workload. The right place.
Run each workload in the right place.
Some work belongs on-device. Some belongs on private infrastructure. Some can use the cloud. Umbra explores how to choose between them.
Before the workload crosses a boundary.
What should stay near the source?
How much of the path should be visible, and to whom?
When does a fallback become a boundary crossed?
Why this exists
Location, privacy, cost, and control.
Proximity
Keep sensitive work close
Some inference and context should stay on-device or inside private infrastructure.
Early research. Real infrastructure questions.
If you are working on local inference, private compute, or workload routing, we would like to compare notes.