Autoscaling config should not be a guessing game

Our approach

• In your cluster. stepscale is a self-hosted operator. Metrics, analysis, and any applied change stay in your account. There is no cross-account access and no phone-home.
• Recommend first. The default is pure advice: it writes recommendations you read with kubectl and never mutates a workload until you approve one.
• Bring your own model. The LLM judgement runs on your own OpenAI or Anthropic key, or not at all in rules-only mode. No model spend flows through us.
• Closed source, public artifacts. The code is private; the image and chart are public and cosign-signed, so you can verify what you run.
• Built by platform engineers. The product solves an autoscaling problem we have hit ourselves, on real production clusters.

Our team

We are a small, distributed team that cares about the cost and shape of production systems, not just shipping features. We work in the open where it counts: public docs, public artifacts, and recommendations you can read and reason about line by line.