stepscale AI sits above your reactive autoscaler as an intelligent tuning layer. Your existing scaler keeps executing every decision in real time; stepscale AI keeps the configuration current, based on actual workload history.
stepscale AI ingests metrics from your ECS or Kubernetes workloads - queue depth, task count, processing rates - and builds a rich history of how your traffic actually behaves.
Statistical models identify daily cycles, weekly patterns, peak-hour signatures, and anomalies. Optimal scaling parameters are calculated from your real traffic, not a generic template.
Optimized thresholds, min/max task counts, and scaling ratios are written back to your existing autoscaler. Your infrastructure adapts continuously, with before/after cost insights.
Thresholds, min/max bounds, tasks-per-message ratios. Adjusted from observed workload, not guessed.
Daily 9am rush versus 3am incident - different signals, different scaling strategies. Alerting integration included.
Reports on over-provisioning, before/after comparisons, and dollar-quantified monthly savings.
Native support for AWS ECS (target tracking, step scaling) and Kubernetes (HPA, KEDA). Same product, both runtimes.
Tuning runs periodically - a few times a day. Zero runtime dependency in your traffic path.
REST API + Terraform provider. Drop into your existing CI/CD and IaC pipelines.
| Capability | Native ECS Autoscaling | K8s HPA / KEDA | stepscale AI |
|---|---|---|---|
| Reactive scaling | Yes | Yes | Delegates to your scaler |
| Manual configuration required | Yes | Yes | Auto-tuned |
| Workload pattern analysis | No | No | Yes |
| Anomaly detection | No | No | Yes |
| Cost insights | No | No | Yes |
| Multi-platform (ECS + K8s) | ECS only | K8s only | Both |
Bring a real ECS or Kubernetes service. Leave with a concrete cost-savings estimate.