Use case · Teams running ML models in production.
MLOps
Standard APMs do not understand per-feature drift / inference latency.
What Unimoni offers
OTLP ingestion + custom metrics via the SDK. Anomaly detection for drift signals.
Concretely
- ✓OTLP receiver
- ✓Custom metrics via REST or OTLP
- ✓Anomaly rules: current vs baseline with a σ threshold
- ✓Multi-region for multi-model deploys
How it works
Arbitrary metrics
The OTLP receiver and REST accept custom model metrics: inference latency, per-feature drift, inference-queue depth.
Anomalies
Anomaly rules compare the current value against a baseline by a σ threshold — a drift signal fires without hand-tuned thresholds.
Scale
Multi-region for multi-model deploys; pricing does not grow with the number of features or their cardinality.