Understand the device
Citadel profiles the hardware and normal resource usage of each device—not a theoretical specification, but the machine as it is actually used.
Citadel · Edge intelligence
One fleet.
Many brains.
Citadel turns the devices your organisation already runs—from smartphones to satellites—into an intelligent edge layer, then gives every prompt exactly as much compute as it needs.
Try the routing lab ↓Routing lab
One request.
The whole fleet.
Citadel connects a library of right-sized open-weight models to every device in your fleet. Your people ask once. Citadel picks the model and the machine, staying on the device when it can and moving to your private cluster when the task demands it.
LIVE ROUTING SIMULATIONDATA BOUNDARY: PRIVATE
Request
Summarise my meeting notes and list the next actions.
Model library
- 1B
- 2B
- 3B
- 8B
- 14B
- 70B
- 671B MoE
Your fleet
- SatellitesEdge
- RobotEdge
- SmartphoneEdge
- LaptopEdge
- WorkstationEdge
- ServersCluster
DecisionLAPTOP · ON DEVICE
Selected model8B · THIS DEVICE
WhyLow complexity · private context · local capacity available
Compute required
28%
How it works
From idle hardware
to useful intelligence.
Fit the local model
A model is selected and provisioned to fit the available CPU, GPU and memory without unnecessarily competing with everyday work.
Decide per prompt
Every request is evaluated at runtime. Straightforward work stays local; tasks that need more capability move to the private cluster.
The outcome
Local efficiency.
Central capability.
Your users get a responsive model close to their work. Your cluster stays available for the difficult requests. Your data remains inside the environment you control.
Explore the full stack →1PRIVATE
FLEET
FLEET
LocalFast + efficient
ClusterDeep + capable