kaito-workspace.yaml

# KAITO's Workspace schema is version-sensitive: this manifest targets
# kaito.sh/v1beta1 (top-level `resource:` and `inference:`). Validated against
# the AKS AI toolchain operator add-on in mid-2026; if your add-on serves a
# different Workspace API version, confirm the field layout with
# `kubectl explain workspace.resource` before applying.
apiVersion: kaito.sh/v1beta1
kind: Workspace
metadata:
  name: workspace-phi-4-mini
  namespace: llm
resource:
  instanceType: "Standard_NC24ads_A100_v4"   # KAITO provisions this node pool
  count: 2
  labelSelector:
    matchLabels:
      apps: phi-4-mini                        # node placement — NOT the pod label; KAITO stamps its pods with kaito.sh/workspace, which is what inference-pool.yaml and podmonitors.yaml select on
inference:
  preset:
    name: phi-4-mini-instruct                 # OpenAI-compatible, served via vLLM
# KAITO publishes a Service `workspace-phi-4-mini` on port 80 that maps to the
# vLLM container's port 5000. The InferencePool routes to pod IPs directly
# (bypassing the Service), so its targetPort must be 5000 — see inference-pool.yaml.