Custom AI Models
Inference health for the models your organization deploys — latency, failures, and the thresholds that decide what counts as broken
Models
12
Inferences (24h)
4.8M
Failure rate
0.42%
p95 latency
318ms
Registered models
Every model an organization has registered, scored against its own thresholds rather than a global default
No matching results
Thresholds
A model is marked degraded or down by its own limits. Raise them for a batch model that is allowed to be slow; tighten them for anything in a request path.
ms
%
%
