A gate in front of dispatch
Fleet managers dispatch what they are told to dispatch. The runtime sits upstream and refuses work that fails policy — a low-confidence detection or an out-of-bounds payload never becomes an order, and no vehicle moves.
Solution · Fleet operations
The runtime does not have to live on the robot. In fleet mode it runs on a PC or edge server beside the fleet stack you already operate, and every transport request — from a vision system, a WMS, or an MES — is admitted or rejected against your policy before it becomes an order. We are not building a fleet manager. We are the gate in front of the one you have.
How it fits
The runtime is inserted ahead of the fleet manager. Everything downstream — assignment, routing, traffic control, the vehicles themselves — keeps working the way it already does.
A vision result, a WMS order, an MES signal. Probabilistic or unvalidated, and not yet trusted.
Runs on a PC or edge server beside the fleet stack. Admits or rejects the request against your policy.
Receives only approved work, then does what it already does — assignment, routing, traffic.
Vehicles receive standard orders from any vendor, and the trace id ties the movement back to the decision.
One runtime product, three deployment solutions
Every solution is the same deterministic engine, deployed differently. Which one you need depends on what has to be governed — a single machine's actions, a robot's ROS stack, or the work handed to a whole fleet.
What the runtime adds
Fleet managers are good at assignment, routing, and traffic. None of them were built to decide whether an AI-originated request should have been actioned at all — or to prove, afterwards, why it was.
Fleet managers dispatch what they are told to dispatch. The runtime sits upstream and refuses work that fails policy — a low-confidence detection or an out-of-bounds payload never becomes an order, and no vehicle moves.
A single trace id follows the chain from the originating request, through the runtime verdict, into the fleet task and the dispatched order — so any movement on the floor can be walked back to the decision that authorised it.
Your fleet manager, your AGV vendors, your WMS all stay exactly where they are. The runtime speaks VDA 5050 — the standard the industry already runs on — so it drops in beside the stack rather than through it.
Reference integration
We maintain an open reference fleet integration — a VDA 5050 fleet manager, mixed-brand vehicle adapters, and a live dashboard — purely to prove the pattern end to end. Scenarios publish real events and then check the actual message traffic, confirming a blocked request produces no fleet task and no vehicle order. The whole thing is public, so you can read exactly how the gate sits in front of dispatch.
A reference integration for proving the pattern — not a fleet-management product, and not a certified VDA 5050 implementation.
Pilots
If you operate AMRs from more than one vendor — or you're being asked to prove why a robot did what it did — we would love to talk.