# Wave — Kubernetes Day 2 Operations Platform > Wave automates Kubernetes Day 2 operations including scaling, right-sizing, traffic management, storage optimization, and Karpenter management. Built by STCLab Inc, Wave reduces cloud costs by 30-40% while improving reliability through ML-driven automation. AWS EKS Service Ready Partner. Supports EKS, GKE, AKS, OpenShift, and self-managed clusters. ## How Wave Works Wave deploys as a lightweight agent in your Kubernetes cluster. It collects metrics from Prometheus/Datadog, analyzes workload patterns using ML models, and automates Day 2 operations: scaling pods via enhanced HPA (Autopilot), right-sizing container resources without restarts (Smart Sizing, applied in-place via Realtime Smart Sizing), managing traffic during spikes (Wave Flow), forecasting capacity exhaustion, and optimizing Karpenter node provisioning. ## Solutions by Problem ### Cost Optimization Teams overpaying for Kubernetes infrastructure due to idle nodes, orphaned storage, and over-provisioned resources. Wave detects underutilized nodes and recommends consolidation or Spot placement (up to 70% node cost savings), identifies unattached PVs for safe cleanup (recover 100% wasted storage spend), and provides cluster-wide right-sizing recommendations (20-40% resource cost reduction). See: https://wavek8s.com/solutions/cost-optimization ### Autoscaling HPA scales on CPU percentages, but throttled pods mislead the autoscaler, causing over-scaling that doubles the bill without fixing latency. Native HPA also reacts 30 seconds to 3 minutes late. Wave Autopilot scales on workload performance rather than CPU percentage, and reacts in seconds. See: https://wavek8s.com/solutions/autoscaling ### Right-Sizing When HPA and VPA run together their conflicting signals cause replica explosions, and VPA restarts pods to apply a change. Wave produces one set of recommendations that coordinate scaling and sizing instead of fighting each other, applies them in place with no restart, and keeps them inside safe minimum and maximum bounds. See: https://wavek8s.com/solutions/right-sizing ### Reliability & Capacity OOM kills, storage exhaustion, and cascading failures. Wave detects abnormal memory growth patterns days before OOMKill events, forecasts storage exhaustion 7-60 days in advance with auto-expansion, predicts cluster capacity limit breaches, and shapes traffic by priority to protect critical services during incidents. See: https://wavek8s.com/solutions/reliability-capacity ### Karpenter in Production Teams adopting Karpenter but lacking visibility into cost, spot ratio, disruption patterns, and consolidation effectiveness. Wave provides a unified Karpenter Dashboard, pre-warms nodes to eliminate cold-start lag, and automates safe Spot/On-Demand workload separation with webhook injection. See: https://wavek8s.com/solutions/karpenter-in-production ### Priority Traffic Control on Istio Standard Istio routes all traffic equally under overload, so checkout requests drop at the same rate as batch jobs. Wave Flow adds 4-tier priority shaping (Bulk, Moderate, Important, Critical) across Istio Sidecar, Ambient mode, and standalone Envoy. NetFUNNEL Integration adds a virtual waiting room for flash sales and peak events. See: https://wavek8s.com/features/wave-flow and https://wavek8s.com/features/netfunnel-integration ### GPU Visibility and GPU Sharing GPU clusters are bought per card and used per workload, and most teams cannot see which workload holds which card. Wave GPU shows every GPU and every GPU workload on the Kubernetes cluster already in use, then shares one physical card across several workloads through HAMi Enterprise from Dynamia, with memory isolation, task priority, and memory over-commit. Measured on a single NVIDIA A10G: six predictive models, three training jobs, or two vLLM 8B services on one card. Available now. See: https://wavek8s.com/gpu ## Key Differentiators - **ML-driven, not rules-based**: Autopilot learns workload patterns, not just CPU thresholds - **In-place resizing**: Smart Sizing's apply path, Realtime Smart Sizing, adjusts pod CPU/memory via the Kubernetes in-place resize subresource — no restart, no rollout (unlike VPA) - **Traffic shaping**: Wave Flow protects services during overload (unique in K8s optimization space) - **Karpenter-native**: First platform with dedicated Karpenter dashboard and node warmup - **Non-invasive**: Works with existing Prometheus/Datadog — no agent replacement needed ## Technical Details - Deployment: Helm chart, AWS Marketplace, or manual install - Data sources: Prometheus, Datadog, Kubernetes metrics-server - Supported platforms: Amazon EKS, Google GKE, Azure AKS, Red Hat OpenShift, self-managed - API: REST API for programmatic access to all features - Pricing: Free under 200 vCPUs across all install methods; above 200 vCPUs, per-vCPU metered billing via AWS Marketplace on Amazon EKS, or contact us for OpenShift and other Kubernetes ## Documentation by sub-brand - [Wave Autoscale Docs](https://wavek8s.com/docs/wave-autoscale/overview): Autopilot, predictive scaling, scheduling, and timing controls - [Wave Sizing Docs](https://wavek8s.com/docs/wave-sizing/overview): Container right-sizing and resource recommendations - [Wave Karpenter Docs](https://wavek8s.com/docs/wave-karpenter/overview): Karpenter dashboard, node warmup, and Spot placement - [Wave Flow Docs](https://wavek8s.com/docs/wave-flow/overview): Traffic shaping and priority-based routing - [Wave Diagnosis Docs](https://wavek8s.com/docs/wave-diagnosis/overview): Incident-first diagnosis grouped by Failures, Waste, and Capacity - [Wave GPU Docs](https://wavek8s.com/docs/wave-gpu/overview): GPU visibility and GPU sharing for Kubernetes - [Wave AI Agent Docs](https://wavek8s.com/docs/wave-ai-agent/overview): Kubernetes MCP server for Claude, Codex, and Cursor. In development - [PV Lifecycle](https://wavek8s.com/docs/additional-features/pv-lifecycle/overview): PV capacity forecasting, auto-expansion, and cleanup - [Alerts](https://wavek8s.com/docs/additional-features/alerts/overview): Configurable alerting for anomalies and thresholds