Intelligent Troubleshooting (Experimental)

The Intelligent troubleshooting page helps you inspect a KaaS cluster using natural language. The assistant can read cluster information, Namespaces, Pods, machines, events, machine groups, and Kubit Packs and explain its findings.

This feature is experimental and is only displayed to users with sufficient access to KaaS and the AI service.

1. Open the cluster and select Intelligent troubleshooting.
2. Select New chat or one of the suggested questions.
3. Describe the symptom and, if known, include the Namespace, Workload or Pod name, and approximate time.
4. Follow the tools shown in the conversation to understand which information was inspected.

Example questions:

• Why are Pods in the shop Namespace stuck in Pending?
• Check recent Warning events in the default Namespace.
• Check the previous logs of Pod api-7d9... in the production Namespace.
• List the machines and summarize Readiness problems.

The assistant's tools are read-only and do not create, edit, restart, or delete Kubernetes resources. After reviewing the evidence, apply corrections with kubectl.

Each chat has an independent history. Select an earlier chat to continue it, or create a new chat for a separate incident. Deleting a chat cannot be undone.

• Running: the assistant is investigating or preparing a response. Use Stop to cancel the current turn.
• Ready: you can send another message.
• Pending: the request is waiting to run.
• Failed: contact support.

The Stop button preserves the chat, and you can continue the same conversation later.

AI responses may be incomplete or incorrect. Compare important conclusions with tool outputs, the current cluster status, metrics, and expected application behavior. Do not enter passwords, Tokens, private keys, or unrelated personal information in a chat. Pod logs may themselves contain sensitive data; store and share chat content according to your organization's security policy.

When speaking with the AI, provide the cluster and project names, affected Namespace or Workload, time observed, related status or error, and checks already performed, and never send Credentials.