Appendix A: Bring Your Own Model

Your environment comes pre-configured with a default LLM. To use your own model provider, follow the steps below for your provider.

The following models have been tested with this environment (as of August 2026). The agentic workflows require a capable model with strong tool-calling support. This is not an endorsement or recommendation of any specific model or provider.

Model Provider Tested Use Cases

GPT-5.4

OpenAI

Agentic troubleshooting, Lightspeed chat with Perses, Agentic updates

GPT-5.5

OpenAI

Agentic troubleshooting, Lightspeed chat with Perses, MCP server for OpenShift

gpt-5-mini-2025-08-07

OpenAI

ACS Lightspeed integration, MCP server for ACS

GPT-5.4-nano

OpenAI

ACS Lightspeed integration, MCP server for ACS

Gemini 2.5 Pro

Google

Lightspeed chat with Perses, Autonomous troubleshooting

Gemini 2.5 Flash

Google

Lightspeed chat with Perses, Autonomous troubleshooting

Gemini 3.1 Pro

Google

MCP server for OpenShift

Claude Opus 4.6

Anthropic

Agentic troubleshooting, Autonomous troubleshooting, Agentic updates, MCP server for OpenShift

Claude Opus 4.8

Anthropic

Testing in progress

Gemma 4 31B

Google

Planned for offline/disconnected deployments

ArgoCD is configured to ignore LLM configuration changes, so your custom model settings will persist without needing to pause auto-sync.

Select your LLM provider below:

OpenAI

Applies to both Lightspeed and Agentic Troubleshooting. Set the variables below with your values, then run the commands.

# Set your OpenAI values
API_KEY="sk-abc123"
ENDPOINT_URL="https://api.openai.com/v1"
MODEL_NAME="gpt-5.4"

# Update Lightspeed credentials and config
oc create secret generic llm-credentials-custom -n openshift-lightspeed \
  --from-literal=apitoken="${API_KEY}" \
  --dry-run=client -o yaml | oc apply -f -

oc patch olsconfig cluster --type=merge -p '{
  "spec": {
    "llm": {
      "providers": [{
        "name": "openai-custom",
        "type": "openai",
        "url": "'"${ENDPOINT_URL}"'",
        "credentialsSecretRef": {"name": "llm-credentials-custom"},
        "models": [{"name": "'"${MODEL_NAME}"'"}]
      }]
    },
    "ols": {
      "defaultProvider": "openai-custom",
      "defaultModel": "'"${MODEL_NAME}"'"
    }
  }
}'

# Update Agentic Troubleshooting credentials and config
oc create secret generic llm-creds-openai-custom -n openshift-lightspeed \
  --from-literal=OPENAI_API_KEY="${API_KEY}" \
  --dry-run=client -o yaml | oc apply -f -

oc apply -f - <<EOF
apiVersion: agentic.openshift.io/v1alpha1
kind: LLMProvider
metadata:
  name: openai-custom
spec:
  type: OpenAI
  openAI:
    url: "${ENDPOINT_URL}"
    credentialsSecret:
      name: llm-creds-openai-custom
EOF

oc patch agent default --type=merge \
  -p '{"spec":{"llmProvider":{"name":"openai-custom"},"model":"'"${MODEL_NAME}"'"}}'
oc patch agent smart --type=merge \
  -p '{"spec":{"llmProvider":{"name":"openai-custom"},"model":"'"${MODEL_NAME}"'"}}'

# Restart Lightspeed to pick up changes
oc rollout restart deployment/lightspeed-app-server -n openshift-lightspeed
Azure OpenAI

Applies to both Lightspeed and Agentic Troubleshooting. Set the variables below with your Azure OpenAI values, then run the commands.

# Set your Azure OpenAI values
API_KEY="your-azure-api-key"
AZURE_ENDPOINT="https://myendpoint.openai.azure.com/openai/v1"
DEPLOYMENT_NAME="gpt-5.4"

# Update Lightspeed credentials and config
oc create secret generic llm-credentials-custom -n openshift-lightspeed \
  --from-literal=apitoken="${API_KEY}" \
  --dry-run=client -o yaml | oc apply -f -

oc patch olsconfig cluster --type=merge -p '{
  "spec": {
    "llm": {
      "providers": [{
        "name": "azure-custom",
        "type": "azure_openai",
        "url": "'"${AZURE_ENDPOINT}"'",
        "deploymentName": "'"${DEPLOYMENT_NAME}"'",
        "credentialsSecretRef": {"name": "llm-credentials-custom"},
        "models": [{"name": "'"${DEPLOYMENT_NAME}"'"}]
      }]
    },
    "ols": {
      "defaultProvider": "azure-custom",
      "defaultModel": "'"${DEPLOYMENT_NAME}"'"
    }
  }
}'

# Update Agentic Troubleshooting credentials and config
oc create secret generic llm-creds-openai-custom -n openshift-lightspeed \
  --from-literal=OPENAI_API_KEY="${API_KEY}" \
  --dry-run=client -o yaml | oc apply -f -

oc apply -f - <<EOF
apiVersion: agentic.openshift.io/v1alpha1
kind: LLMProvider
metadata:
  name: openai-custom
spec:
  type: OpenAI
  openAI:
    url: "${AZURE_ENDPOINT}"
    credentialsSecret:
      name: llm-creds-openai-custom
EOF

oc patch agent default --type=merge \
  -p '{"spec":{"llmProvider":{"name":"openai-custom"},"model":"'"${DEPLOYMENT_NAME}"'"}}'
oc patch agent smart --type=merge \
  -p '{"spec":{"llmProvider":{"name":"openai-custom"},"model":"'"${DEPLOYMENT_NAME}"'"}}'

# Restart Lightspeed to pick up changes
oc rollout restart deployment/lightspeed-app-server -n openshift-lightspeed
Google Cloud Vertex AI (Agentic only)
Vertex AI is supported for Agentic Troubleshooting only. Lightspeed core will continue using the pre-configured model.

Set the variables below with your values, then run the commands. The example uses Anthropic Claude. For Google Gemini, change MODEL_PROVIDER to Google and update the model name.

# Set your Vertex AI values
GCP_PROJECT_ID="my-project"
GCP_REGION="us-east5"
MODEL_PROVIDER="Anthropic"
MODEL_NAME="claude-sonnet-4@20250514"
GCP_CREDENTIALS_FILE="/path/to/your/adc.json"

# Create Vertex AI credentials from your GCP service account JSON
oc create secret generic llm-creds-vertex-custom -n openshift-lightspeed \
  --from-file=GOOGLE_APPLICATION_CREDENTIALS="${GCP_CREDENTIALS_FILE}" \
  --dry-run=client -o yaml | oc apply -f -

# Update the Agentic LLMProvider
oc apply -f - <<EOF
apiVersion: agentic.openshift.io/v1alpha1
kind: LLMProvider
metadata:
  name: vertex-custom
spec:
  type: GoogleCloudVertex
  googleCloudVertex:
    projectID: "${GCP_PROJECT_ID}"
    region: "${GCP_REGION}"
    modelProvider: ${MODEL_PROVIDER}
    credentialsSecret:
      name: llm-creds-vertex-custom
EOF

# Update the Agentic Agent
oc patch agent default --type=merge \
  -p '{"spec":{"llmProvider":{"name":"vertex-custom"},"model":"'"${MODEL_NAME}"'"}}'
oc patch agent smart --type=merge \
  -p '{"spec":{"llmProvider":{"name":"vertex-custom"},"model":"'"${MODEL_NAME}"'"}}'
RHEL AI / vLLM (Lightspeed only)
RHEL AI is supported for Lightspeed core only. Agentic Troubleshooting will continue using the pre-configured model.

Set the variables below with your RHEL AI values, then run the commands.

# Set your RHEL AI values
API_KEY="your-api-key"
RHELAI_ENDPOINT="https://my-rhelai-host:8443/v1"
MODEL_NAME="granite-3-8b-instruct"

# Update Lightspeed credentials
oc create secret generic llm-credentials-custom -n openshift-lightspeed \
  --from-literal=apitoken="${API_KEY}" \
  --dry-run=client -o yaml | oc apply -f -

# Update Lightspeed OLSConfig for RHEL AI
oc patch olsconfig cluster --type=merge -p '{
  "spec": {
    "llm": {
      "providers": [{
        "name": "rhelai-custom",
        "type": "rhelai_vllm",
        "url": "'"${RHELAI_ENDPOINT}"'",
        "credentialsSecretRef": {"name": "llm-credentials-custom"},
        "models": [{"name": "'"${MODEL_NAME}"'"}]
      }]
    },
    "ols": {
      "defaultProvider": "rhelai-custom",
      "defaultModel": "'"${MODEL_NAME}"'"
    }
  }
}'

# Restart Lightspeed to pick up changes
oc rollout restart deployment/lightspeed-app-server -n openshift-lightspeed