Inference routing for European edge locations
NeuralRoad operates a small fleet of GPU-adjacent edge nodes and routes model inference requests to the closest healthy location. Built for teams that need predictable latency without running their own regional footprint.
What we run
Edge routing
Requests are terminated at the nearest node and forwarded to the serving backend over persistent links. Failover is automatic.
Model caching
Frequently requested weights are kept warm on node-local NVMe, which removes cold-start penalties on bursty workloads.
Signed artifacts
Model and dataset downloads are served through expiring signed URLs. Unsigned requests are rejected at the edge.
Locations
| Node | Region | Endpoint | State |
|---|---|---|---|
| de-nue-01 | eu-central | de.neuralroad.online | operational |
| dk-cph-01 | eu-north | dk.neuralroad.online | operational |
Per-node health is exposed at /healthz on each endpoint. Aggregate state is on the status page.
Access
The platform is currently in limited availability. Existing customers use the credentials issued during onboarding; new access requests go through hello@neuralroad.online.