What shipped
Every alert incident has an Explain this incident action. Kubeadapt writes up what happened, why it matters and what to do next, including the cost drivers behind a spike or a budget crossing. The text streams in as it is written, and you can regenerate it at any time. See Smart Alerting for the alerting side.
Rightsizing recommendations have a similar Explain with AI action that explains why a recommendation was made.
What data the model sees
Kubeadapt runs the analysis itself: it finds the contributors, classifies the change and computes the numbers. It then sends a prompt to a hosted large language model (Anthropic models, served through the Anthropic API or AWS Bedrock), which writes the explanation. Depending on the incident or recommendation, the prompt includes:
- the workload's namespace, name and kind
- its current and recommended CPU and memory requests and limits
- usage statistics: percentiles, maximums, trends, OOM kills and restart counts
- cost figures: current monthly cost, the cost change behind the incident and estimated savings
- the incident's findings, such as the contributing workloads and their likely causes
Turning it on
Kubeadapt AI is off by default. To turn it on for your workspace, contact support@kubeadapt.io. If your data-handling rules restrict where this data may go, check the list above with your team first.
Usage is capped per workspace for each 30-day period. When the cap is reached, the incident page says so instead of hiding the action.
Getting started
Once it is on, open any incident under Alerts > Incidents and select Explain this incident.