AI Agents

AI Agents let you operate your Kubernetes clusters with natural language. Powered by Anthropic's Claude, an agent takes a goal you describe, plans the steps, runs cluster skills (list pods, deploy an app, scale, read logs, …), and reports back — pausing for your approval before anything risky.

Prerequisites

• A running Kubernetes cluster
• An Anthropic (Claude) API key — agents use your own key

Setup: register your Claude API key

Agents will not run until you add a Claude API key.

1. Open Settings → Integrations
2. Under AI Agent API Key, paste your Anthropic API key (starts with sk-ant-)
3. Choose the model the agent should use:
   • Claude Sonnet 4.5 (default) — balanced speed and capability
   • Claude Opus 4.1 — most capable, for complex tasks
   • Claude Haiku 4.5 — fastest and cheapest, for simple tasks
4. Save

You can change the model at any time from the same screen.

Overview

Open AI Agents from the sidebar. The feature is organized into:

Agent Chat

1. Open Agent Chat and select the target cluster
2. Type what you want, for example:
   • "List all pods in the default namespace that are not Running."
   • "Scale the web deployment to 3 replicas."
   • "Show the last 100 lines of logs for the failing pod."
3. The agent streams its progress — each thought, action (skill), and result appears live
4. If an action is risky, the run pauses and asks for your approval before continuing

Deploy from Git

The Deploy agent builds and deploys an app straight from a GitHub or GitLab repository.

Before you start: connect GitHub and/or GitLab under Settings → Integrations.

The deploy agent currently supports Nuxt and Next.js projects. When you pick a repository, Kubo checks its package.json and blocks unsupported frameworks.

1. Open Deploy and choose the source provider (GitHub / GitLab)
2. Select a repository and branch
3. Configure the deployment:
   • Cluster and Namespace
   • Replicas
   • Expose (port, and optionally a domain)
   • Environment variables
4. Start the deployment and watch the agent build the image, push it, and create the Deployment / Service / Ingress in real time

→ See Deploy Applications for more.

Templates and schedules

Templates are saved agent goals you can reuse.

• Save any goal as a template, or create one from Templates
• Run a template on demand against a chosen cluster
• Optionally attach a schedule so the template runs periodically

Managing deployed apps

The Apps section lists everything agents have deployed. For each app you can:

• Update — pull the latest code and redeploy
• Scale — change the replica count
• Delete — remove the deployment and its resources

Approvals and safety

Agents are designed to be safe by default:

• Approvals — actions that change or delete resources require your explicit approval before they run. Low-risk read operations run without interruption.
• Limits — each run is bounded by a maximum number of iterations and write actions, and by a timeout, so an agent cannot loop or run indefinitely.
• Scope — an agent only acts on the cluster you selected.

Activities and history

Every run is recorded under Activities. Open a run to review its full step-by-step trace, the skills it used, token usage, and the final result. This is the place to look when a task didn't do what you expected.

Troubleshooting

"Claude API Key Required"

No key is registered. Add your Anthropic API key under Settings → Integrations.

The chat replies with an error on the first message

Your API key may be invalid, out of quota, or set to a retired model. Re-check the key and pick a current model in Settings → Integrations.

A deploy is rejected as an unsupported framework

The deploy agent supports Nuxt and Next.js only. Confirm the repository's package.json declares one of them.

A run "timed out"

Long deploys (image build + rollout) can exceed the default limits. Re-run the task, or split it into smaller goals.