|Index|Creating SUSE AI Factory Blueprints|Blueprint templates and reference architectures
Applies to SUSE AI Factory

A Blueprint templates and reference architectures

A1 Minimal two-component blueprint template

apiVersion: ai-factory.suse.com/v1alpha1
kind: Blueprint
metadata:
  labels:
    ai-factory.suse.com/blueprint-name: my-demo-ollama   # == folder name
    ai-factory.suse.com/blueprint-version: 1.0.0         # dots
  name: my-demo-ollama-1-0-0                             # dashes
# One-paragraph comment: what this demo does, which models it serves, and how it
# differs from its sibling variant.
spec:
  source: Custom  # use Community once every target operator runs 2.3.1 or newer
  displayName: My Demo (Ollama, CPU)
  version: 1.0.0
  description: |
    What the demo does and which component does what.

    What this CR deploys versus what runs locally, plus a pointer to the sibling
    variant.

    Prerequisites: the SUSE AI Factory operator, the application-collection
    ClusterRepo (+ credentials secret), a default StorageClass. After importing,
    create an AIWorkload from this blueprint in the AI Factory UI.
  components:
  # ----------------------------------------------------------------------- #
  # Ollama — local LLM, CPU. Reachable in-cluster at http://ollama:11434.
  # ----------------------------------------------------------------------- #
  - chartName: ollama
    chartRepo: application-collection
    chartVersion: 1.55.0
    values:
      fullnameOverride: ollama          # pin the service name
      image:
        pullPolicy: IfNotPresent
        registry: dp.apps.rancher.io
        repository: containers/ollama
        tag: 0.21.2-11.48
      ollama:
        gpu:
          enabled: false
          draEnabled: false
        models:
          pull:
          - qwen2.5:0.5b                # literal value — never {{MODEL}}
          run:
          - qwen2.5:0.5b
        port: 11434
      persistentVolume:
        enabled: true
        accessModes:
        - ReadWriteOnce
        size: 20Gi
      service:
        port: 11434
        type: ClusterIP
  # ----------------------------------------------------------------------- #
  # Qdrant — vector store. Reachable in-cluster at http://qdrant:6333 (REST).
  # ----------------------------------------------------------------------- #
  - chartName: qdrant
    chartRepo: suse-ai-registry
    chartVersion: 1.17.0
    values:
      fullnameOverride: qdrant
      persistence:
        accessModes: ["ReadWriteOnce"]
        size: 5Gi
      service:
        type: ClusterIP

A2 ClusterRepo template

# Registers an upstream Helm repo as a ClusterRepo, so Blueprint CR components with
# `chartRepo: <NAME>` resolve. Apply BEFORE the Blueprint CR:
#   kubectl apply -f <NAME>-clusterrepo.yaml
apiVersion: catalog.cattle.io/v1
kind: ClusterRepo
metadata:
  name: <NAME>          # MUST equal the chartRepo value used in the CR
spec:
  url: https://charts.example.com          # or: oci://registry-1.docker.io/<ORG>

A3 Reference blueprints

Architecture patternReference blueprint

Minimal single-model CR

visiongpt-ollama, visiongpt-vllm

Full stack (Airflow, PostgreSQL, Qdrant, Ollama)

dora-compliance-ollama

Upstream third-party Helm repositories

data-governance-ollama

Custom Airflow images

fraud-detection-ollama

Multi-model GPU serving

finops-multimodel-vllm

Hybrid architecture (GPU chat and CPU embeddings)

insurance-support-vllm

Comprehensive README and DAG documentation

airflow-genai-rag