Skip to main content

​
How it works

The Azure AI connector:
  • Discovers Azure OpenAI deployments and AI Foundry projects via the Azure Resource Manager API
  • Streams invocations from Azure Application Insights dependency logs
Elitery deploys and manages the connector. Your developers do nothing.

​
What Elitery needs from you

Azure service principal with read-only access: 
# Create service principal
az ad sp create-for-rbac --name "muster-connector" --role Reader \
  --scopes /subscriptions/<subscription-id>

# Output:
# AZURE_CLIENT_ID
# AZURE_CLIENT_SECRET
# AZURE_TENANT_ID
Additional permissions needed:
  • Microsoft.CognitiveServices/accounts/read — list Azure OpenAI accounts
  • Microsoft.CognitiveServices/accounts/deployments/read — list deployments
  • Reader on Application Insights workspace
Application Insights connection string — for execution streaming: 
AZURE_APPINSIGHTS_CONNECTION_STRING=InstrumentationKey=...;WorkspaceId=...

​
What your developers do

Nothing for automatic monitoring. For execution data to appear in Application Insights (which the connector reads), ensure your Azure OpenAI calls go through the Azure SDK with diagnostics enabled — this is the default when using openai Python SDK with Azure endpoint configuration. For precise checks on critical agents: 
import httpx, threading, uuid
from openai import AzureOpenAI

client = AzureOpenAI(
    azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT"],
    api_key=os.environ["AZURE_OPENAI_KEY"],
    api_version="2024-02-01",
)

def run_agent(prompt: str) -> str:
    job_id = str(uuid.uuid4())
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": prompt}]
    )
    output = response.choices[0].message.content

    threading.Thread(target=lambda: httpx.post(
        f"{MUSTER_URL}/api/v1/jobs/{job_id}/quality",
        json={"agent_id": "azure-gpt4o-agent", "checks": [
            {"check_id": "output_not_empty", "severity": "HIGH", "passed": bool(output)},
        ]},
        timeout=2.0,
    ), daemon=True).start()

    return output