Deploy your first model in a disconnected environment

After you install the Foundry Local extension expansion pack, the phi-4-mini CPU is available as part of the base pack. This article shows you how to create your first deployment from the catalog.

Prerequisites

Make sure you followed the steps in Prepare to deploy Foundry Local on Azure Local in disconnected environments and Deploy Foundry Local on Azure Local in a disconnected environment to set up your environment and deploy the Foundry Local extension.

Generate access token and request headers

Run these commands to generate an access token and create the request headers that authenticate REST API calls to Foundry Local.

$DisplayName = "FoundryOnArc-Disconnected"
$app = az ad app list --display-name $DisplayName --query "[0]" -o json | ConvertFrom-Json
$appId = $app.AppId

$token = az account get-access-token --resource "$appId" --query accessToken -o tsv
$headers = @{ "Authorization" = "Bearer $token" }

# If using gateway api:
$baseUrl = "https://<FOUNDRY_API_BASE_PATH>/inference-api"

# If not using ingress, use the direct API endpoint instead:
# $baseUrl = "https://<FOUNDRY_API_BASE_PATH>"

Create your first deployment (phi-4-mini CPU)

Run this request to create a phi-4-mini CPU deployment from the model catalog in your Foundry Local environment.

$namespace = "foundry-local-operator"
$deploymentName = "phi4-cpu-demo"

# Use 'external' to expose model API outside the Kubernetes cluster. 'internal' is the default value.
$exposure = "external"

$body = @{
  name = $deploymentName
  spec = @{
    displayName = "Phi 4 CPU Demo"
    model = @{
      catalog = @{
        name = "phi-4-mini"
      }
    }
    workloadType = "generative"
    compute = "cpu"
    runtime = "onnx-genai"
    replicas = 1
    port = 5000
    authentication = @{
      enabled = $true
    }
    endpoint = @{
      exposure = $exposure
      path = "/phi4-cpu-demo(/|$)(.*)"
      pathType = "ImplementationSpecific"
    }
  }
} | ConvertTo-Json -Depth 20

Invoke-RestMethod `
  -Uri "$baseUrl/api/v1/namespaces/$namespace/deployments" `
  -Headers ($headers + @{ "Content-Type" = "application/json" }) `
  -Method POST `
  -Body $body

Verify deployment

Run the following command to confirm the deployment exists and check whether it's moving to a ready state.

Invoke-RestMethod `
  -Uri "$baseUrl/api/v1/namespaces/$namespace/deployments/$deploymentName" `
  -Headers $headers `
  -Method GET

Expected result:

• Deployment exists and returns successfully from the GET request.
• Deployment status moves to ready state after model cache and pod startup complete.

Related content

• Troubleshoot Foundry Local on Azure Local in disconnected environments
• Configure authentication and authorization for Foundry Local on Azure Local in disconnected environments
• Expand model catalog for Foundry Local on Azure Local in disconnected environments