Analysis where you need it, not another chatbot
We started adding AI analysis to the places in Kubeadapt where a little extra context saves you real time — the moment an incident fires, the spots where spend looks off, the recommendations that need a "why." Instead of making you go read a dashboard, Kubeadapt writes the explanation for you: what happened, why it matters, and what to do next.
This is handled entirely by Kubeadapt. It's free, and it's on by default.
Your metrics never go to the LLM
This is the part that matters most: we never send your important information into the model. Utilization metrics, cost figures, workload details, and cluster internals stay inside your workspace. Kubeadapt does the number-crunching itself and only uses the LLM to phrase the result in plain language.
- No raw utilization or cost data leaves Kubeadapt. The analysis is computed locally against your data; the model sees a shaped, minimal prompt — never your metrics dumps.
- No training on your data. Nothing you run through Kubeadapt is used to train a model.
- You stay in control. The feature is scoped per workspace, so you decide where it's active.
Where you'll see it first
Incident analysis. Every alert incident has an "Explain this incident" action. It lays out the causal chain, the cost drivers behind the spike or budget crossing, and concrete next steps — with a confidence tier and the data window it looked at. Regenerate any time. See Smart Alerting for the alerting side.
More surfaces are coming as we connect additional context into the analysis to make it more useful.
Under the hood
Kubeadapt AI runs on LiteLLM with AWS Bedrock and the Anthropic API Console as providers today. That gives us a clean abstraction over the model layer: if you have special requirements — a specific provider, a self-hosted model, tighter data-residency needs, or AI in places we haven't reached yet — we can adapt the flows to fit. Tell us what you need and we'll optimize around it.
Getting started
Nothing to install. Open any incident under Alerts > Incidents and hit Explain this incident. For how the privacy model works in detail, see the Kubeadapt AI section above (dedicated docs page coming soon).