First Agent and Run

Before you begin

Install the operator using images available to your cluster and configure its runner and Temporal adapter first. You also need an OpenAI-compatible model endpoint reachable from the runner. This example is a schema-valid integration exercise, not a promise that the public chart works without supplying its private-image replacements and service configuration.

Define a model endpoint and agent

Replace the example URL and model ID with your own reachable model. All resources are created in the same namespace. Save the next two examples together in a file named first-run.yaml, separated by YAML document markers.

apiVersion: core.kaigents.io/v1alpha1
kind: ModelEndpoint
metadata:
  name: local-model
  namespace: kaigents
spec:
  url: http://model-server.kaigents.svc.cluster.local:8000/v1
  model: example-model
  provider: local
  capabilities:
    chat: true
---
apiVersion: core.kaigents.io/v1alpha1
kind: Agent
metadata:
  name: summarizer
  namespace: kaigents
spec:
  systemPrompt: "Summarize the input in three sentences; do not call tools."
  modelEndpointRef: local-model
  modelName: example-model
  allowedTools: []

Add and submit a Run

Append this resource to first-run.yaml after a document separator, then apply it. Run.spec.target accepts Agent and Team, not Process.

---
apiVersion: core.kaigents.io/v1alpha1
kind: Run
metadata:
  name: first-run
  namespace: kaigents
spec:
  target:
    kind: Agent
    name: summarizer
  input: "Kaigents coordinates AI work as Kubernetes resources."

Inspect the outcome

The API server validates the resource shapes; it cannot prove that the model service, credentials, or execution workers are healthy. Check status.phase, status.message, and conditions. If execution does not begin, inspect the operator and adapter logs and follow the troubleshooting guide.

kubectl apply --dry-run=server -f first-run.yaml
kubectl apply -f first-run.yaml
kubectl get run first-run -n kaigents -w
kubectl get run first-run -n kaigents -o yaml

Clean up

Delete the sample resources when finished. Deleting a Run object does not automatically establish a data-retention policy for artifacts or external model logs.

kubectl delete -f first-run.yaml