Sedai Kubernetes Optimization
See how Sedai autonomously optimizes K8s — across EKS, AKS, GKE, and more.
The Intelligence Layer Your Autoscalers Are Missing
HPA scales pods, Karpenter provisions nodes, but neither knows whether your workloads are actually sized correctly. Sedai continuously right-sizes pod requests and tunes scaling targets based on real workload behavior, so your autoscalers are always working from accurate inputs.
Autoscalers Scale Load. They Don't Fix Sizing.
Autoscalers like HPA and Karpenter are reactive by design. The inputs they scale from are usually wrong before scaling even begins.
Resource requests are set once and inherited blindly.
Estimates go stale, but HPA still scales from whatever they say.
VPA fights with HPA in live clusters.
Vertical and horizontal autoscaling conflict at runtime, so most teams disable VPA in production or limit it to off-hours, leaving right-sizing unaddressed.
Node provisioning inherits pod-level inefficiency.
Karpenter bins pods onto nodes based on declared requests, so inflated requests mean larger nodes, lower density, and higher spend regardless of actual utilization.
Key Capabilities
Sedai adds an intelligence layer at both the pod and node level, filling the gaps that leave native autoscalers guessing.
Set the Right HPA Target, Not Just a Guess
Sedai analyzes workload behavior and SLOs to automatically set the optimal HPA target, balancing pod size against replica count for the most cost-effective configuration.
Get Vertical and Horizontal Scaling at the Same Time
Kubernetes blocks HPA and VPA from running together. Sedai's vertical scaling works alongside HPA, safely adjusting pod requests and limits in small increments — conflict-free.
Help Karpenter Pack Nodes More Efficiently
Oversized pods waste node capacity. Sedai right-sizes pods first so Karpenter bin-packs onto fewer, cheaper nodes, and recommends instance types based on usage and pricing.
“We’ve gone from what used to be automated, deterministic workflows to autonomous, with Sedai. The human element is indispensable, and it always will be. But more and more engineering toil is being done by AI, so as humans, we can move up the value chain. That’s how we can deliver what our customers expect of us.”
Suresh Sangiah
SVP of Engineering // Palo Alto Networks
Works With Your Stack
Horizontal Pod Autoscaler (HPA)
Vertical Pod Autoscaler (VPA)
Karpenter
Cluster Autoscaler
KEDA
The Sedai Difference
Without Sedai
- HPA utilization targets are set manually and rarely revisited
- Kubernetes prevents using HPA and VPA on the same workload
- Oversized pods force Karpenter to provision larger, more expensive nodes
- Node autoscalers pick from the full instance catalog without guidance
- Scaling decisions are based on infrastructure metrics alone
- Sedai continuously determines and tunes the optimal HPA target based on live workload behavior
- Sedai's vertical scaling works alongside HPA simultaneously, without the native conflict
- Right-sized pods let Karpenter pack workloads onto fewer nodes at lower cost
- Sedai recommends a targeted instance type list so autoscalers make more cost-effective selections
- Sedai factors in application performance and SLOs, ensuring scaling doesn't compromise reliability
Resources
How We Solved the GPU Problem for Kubernetes
We solved GPU optimization. In this episode of 1 IDEA, Suresh Mathew sits down with Pooja Malik, Distinguished Engineer at Sedai, to talk through how Sedai's engineering team built GPU optimization from scratch, why the standard metrics fail, and what's still unsolved.
How Palo Alto Networks Takes Control of Its High-Stakes Cloud
Learn how Palo Alto Networks dramatically reduced its cloud costs with Sedai
How Palo Alto Networks Saved $3.5M with Sedai's AI Agent
See how Sedai's AI agent saved Palo Alto Networks $3.5M in Kubernetes optimization autonomously & safely.