Ansible Lightspeed Enterprise

The enterprise version of Ansible Lightspeed — IBM watsonx Code Assistant for Red Hat Ansible Automation Platform — is where things get serious. Custom model tuning on your playbooks, content attribution for compliance, and on-premises deployment for air-gapped environments.

I have deployed this for enterprise clients and here is what you need to know.

What Changes in Enterprise

Architecture

┌─────────────────────────────────────────────────┐
│              Your Environment                     │
│                                                   │
│  ┌─────────┐    ┌──────────────────────────────┐ │
│  │ VS Code  │───▶│  Ansible Automation Platform  │ │
│  │ Extension│    │  ┌────────────────────────┐  │ │
│  └─────────┘    │  │  watsonx Code Assistant │  │ │
│                  │  │  ┌──────────────────┐  │  │ │
│                  │  │  │  Granite Model   │  │  │ │
│                  │  │  │  (Fine-tuned)    │  │  │ │
│                  │  │  └──────────────────┘  │  │ │
│                  │  └────────────────────────┘  │ │
│                  └──────────────────────────────┘ │
└─────────────────────────────────────────────────┘

The model runs inside your Ansible Automation Platform on OpenShift. No data leaves your network.

Deployment on OpenShift

Prerequisites

• OpenShift 4.14+
• Red Hat Ansible Automation Platform 2.4+
• NVIDIA GPU Operator installed (for GPU inference)
• watsonx Code Assistant subscription

Install the Operator

# Create namespace
oc new-project watsonx-code-assistant

# Install the operator via OperatorHub
# Navigate to: Operators → OperatorHub → search "watsonx Code Assistant"
# Click Install → select watsonx-code-assistant namespace

Deploy the Model Service

apiVersion: wca.ibm.com/v1
kind: WatsonxCodeAssistant
metadata:
  name: ansible-lightspeed
  namespace: watsonx-code-assistant
spec:
  license:
    accept: true
  model:
    name: granite-8b-code-instruct-ansible
    resources:
      limits:
        nvidia.com/gpu: "1"
        memory: "32Gi"
      requests:
        nvidia.com/gpu: "1"
        memory: "24Gi"
  storage:
    size: 50Gi
    storageClassName: gp3-csi
  replicas: 2

Connect to Ansible Automation Platform

# In the AAP configuration
ansible_lightspeed:
  enabled: true
  url: https://watsonx-code-assistant.apps.ocp.internal.com
  model_name: granite-8b-code-instruct-ansible
  auth:
    type: service-account
    secret: watsonx-sa-token

Custom Model Tuning

The killer feature: train the model on YOUR organization’s playbooks.

Prepare Training Data

# Export your best playbooks
find /path/to/ansible-repos -name "*.yml" -o -name "*.yaml" | \
  grep -v "molecule\|tests\|.github" | \
  head -5000 > training_files.txt

# The tuning pipeline expects:
# - Well-structured YAML
# - FQCN module names
# - Proper task naming
# - Consistent coding style

Run Tuning Job

apiVersion: wca.ibm.com/v1
kind: ModelTuningJob
metadata:
  name: custom-ansible-model-v1
spec:
  baseModel: granite-8b-code-instruct-ansible
  trainingData:
    source: pvc
    pvcName: ansible-training-data
    path: /data/playbooks/
  hyperparameters:
    epochs: 3
    learningRate: 1e-5
    batchSize: 8
  output:
    modelName: granite-8b-ansible-custom-v1
    pvcName: custom-models

Results I Have Seen

After tuning on 3,000 enterprise playbooks:

• Module accuracy: 92% → 97% (knows your custom collections)
• Variable naming: follows your conventions (company prefix, etc.)
• Role patterns: generates roles matching your directory structure
• Error handling: includes your standard retry/rescue patterns

Content Attribution

Enterprise compliance teams love this feature. Every suggestion shows which Ansible Galaxy content or documentation influenced it:

{
  "suggestion": "ansible.builtin.dnf: name: nginx state: present",
  "attributions": [
    {
      "source": "ansible.builtin",
      "version": "2.16.0",
      "license": "GPL-3.0",
      "url": "https://galaxy.ansible.com/ansible/builtin"
    }
  ],
  "confidence": 0.94
}

This helps with:

• License compliance — know what licenses your generated code carries
• Audit requirements — trace every suggestion to its source
• IP protection — ensure generated code does not contain proprietary content

Post-Processing Rules

Enforce your organization’s standards automatically:

# watsonx-rules.yml
rules:
  - name: enforce-fqcn
    description: All modules must use FQCN
    action: transform
    
  - name: ban-command-module
    description: Prefer specific modules over command/shell
    pattern: "ansible.builtin.command"
    action: warn
    message: "Consider using a specific module instead of command"
    
  - name: require-tags
    description: All tasks must have tags
    pattern: "missing:tags"
    action: add
    default_tag: "untagged"
    
  - name: naming-convention
    description: Task names must start with uppercase
    pattern: "^- name: [a-z]"
    action: transform

Governance Dashboard

The admin console provides:

• Usage metrics — suggestions generated, accepted, rejected per user/team
• Quality metrics — acceptance rate, edit distance after acceptance
• Compliance status — rule violations, attribution coverage
• Model performance — latency, accuracy trends over time

ROI Calculation

Based on enterprise deployments:

For a 10-person platform team: ~$200K/year in productivity gains vs ~$50K subscription cost.

Related Resources

• Ansible Lightspeed tutorial
• Ansible Lightspeed vs GitHub Copilot
• Ansible Lightspeed AI automation 2026
• Red Hat Ansible Automation Platform
• Ansible tutorial for beginners
• Book a consultation

About the Author

I am Luca Berton, AI and Cloud Advisor. I deploy Ansible Automation Platform and watsonx Code Assistant for enterprise clients across Europe. Contact me to discuss your Ansible AI strategy.