Kubex
Kubex provides an autonomous Kubernetes optimization engine that continuously right-sizes containers, nodes, and autoscaling parameters using ML-driven pattern models. The platform operates in recommendation or fully automated mode with policy guardrails, and includes an AI agent accessible via natural language or MCP for teams working across EKS, AKS, GKE, and on-prem environments. Customers report reclaiming hundreds of thousands of dollars in annual cloud costs within weeks of deployment.
- Developer Tools
- Artificial Intelligence
- Software Only
- AI Agents
Funding
Founders
Founder details are not available yet.
Product
Problem
Kubernetes workloads often run with misconfigured resource requests and limits, leading to either overprovisioning that inflates cloud costs or underprovisioning that causes performance issues like OOM kills and CPU throttling. Manual tuning of pod requests, node types, and autoscaling parameters is time-consuming, error-prone, and does not scale across large, dynamic environments—especially for AI/GPU workloads that have highly variable demand patterns.
Solution
Kubex provides an autonomous optimization engine that continuously analyzes Kubernetes resource utilization and scaling behavior across containers, nodes, and cloud instances. The platform applies ML-driven pattern models to right-size pod requests, optimize node selection, predictively scale for cyclical loads, and pre-warm nodes ahead of demand spikes. It operates in either recommendation mode for human-in-the-loop review or fully automated mode with policy-driven guardrails, and integrates with GitOps and CI/CD pipelines for controlled change management. The platform also includes an AI agent accessible via natural language or the Model Context Protocol (MCP) interface, enabling teams to interact with optimization intelligence from their existing workflows.
Target Audience
Primary users are SREs, platform engineers, and FinOps practitioners responsible for managing Kubernetes infrastructure and cloud costs, as well as AI/ML and GPU infrastructure teams operating inference and training workloads.
Features
- Automated Pod Scaler that right-sizes pod requests and limits using ML models of learned utilization, with human-in-the-loop or fully automated execution via mutating webhooks and in-place resize
- Predictive Pod Scaler that drives high efficiency for cyclical loads by generating scaling plans ahead of peak and off-peak periods, reducing throttling and OOM kills associated with reactive approaches
- Node Pre-Warmer that uses ML pattern models to generate pre-warming plans, deploying nodes in anticipation of load to prevent scheduling delays—particularly effective for AI workloads where CUDA pulls and model loading dominate cold starts
- Node Optimizer that determines optimal node types, CPU-to-memory ratios, and scaling parameters based on learned behavior, supporting Karpenter node autoscaling optimization and automated application via CI/CD pipelines
- Bin Packer that intelligently schedules statefulsets and long-running workloads to increase pod density while adapting aggressiveness based on pod-sizing maturity
- HPA Optimizer that recomputes horizontal autoscaler thresholds, scale policies, and replica bounds against observed workload behavior rather than template defaults
- GPU optimization covering GPU fractioning via time-slicing, MIG, or MPS, KV cache tuning for vLLM, and SKU selection across AWS, neoclouds, and on-prem environments
- AI Agent delivering deterministic AI via natural language or MCP interface, with skills for rightsizing, node selection, bin packing, HPA and JVM tuning
- Kubernetes-native stack with support for EKS, AKS, GKE, OpenShift, and OKE, delivering recommendations as Karpenter NodePool, Terraform, or GitOps diffs