refactor: Replace OpenAI API call with Foundry SDK

Author: Avijit-Microsoft

infra/main.bicep

@@ -33,8 +33,8 @@ param gptModelVersion string = '2024-07-18'
 @description('Optional. Version of the GPT model to deploy.')
 param embeddingModelVersion string = '2'
 
-@description('Optional. API version for the Azure OpenAI service.')
-param azureOpenaiAPIVersion string = '2025-04-01-preview'
+@description('Optional. API version for the Azure AI Services.')
+param azureAIServicesAPIVersion string = '2025-04-01-preview'
 
 @minValue(10)
 @description('Optional. Capacity of the GPT deployment:')
@@ -494,7 +494,6 @@ module jumpboxVM 'br/public:avm/res/compute/virtual-machine:0.15.0' = if (enable
 // ========== Private DNS Zones ========== //
 var privateDnsZones = [
   'privatelink.cognitiveservices.azure.com'
-  'privatelink.openai.azure.com'
   'privatelink.services.ai.azure.com'
   'privatelink.azurewebsites.net'
   'privatelink.blob.${environment().suffixes.storage}'
@@ -509,22 +508,20 @@ var privateDnsZones = [
 // DNS Zone Index Constants
 var dnsZoneIndex = {
   cognitiveServices: 0
-  openAI: 1
-  aiServices: 2
-  appService: 3
-  storageBlob: 4
-  storageQueue: 5
-  storageFile: 6
-  cosmosDB: 7
-  keyVault: 8
-  sqlServer: 9
-  searchService: 10
+  aiServices: 1
+  appService: 2
+  storageBlob: 3
+  storageQueue: 4
+  storageFile: 5
+  cosmosDB: 6
+  keyVault: 7
+  sqlServer: 8
+  searchService: 9
 }
 
 // List of DNS zone indices that correspond to AI-related services.
 var aiRelatedDnsZoneIndices = [
   dnsZoneIndex.cognitiveServices
-  dnsZoneIndex.openAI
   dnsZoneIndex.aiServices
 ]
 
@@ -613,11 +610,11 @@ module keyvault 'br/public:avm/res/key-vault/vault:0.12.1' = {
         }
         {
           name: 'AZURE-OPENAI-PREVIEW-API-VERSION'
-          value: azureOpenaiAPIVersion
+          value: azureAIServicesAPIVersion
         }
         {
           name: 'AZURE-OPENAI-ENDPOINT'
-          value: aiFoundryAiServices.outputs.endpoints['OpenAI Language Model Instance API']
+          value: aiFoundryAiServices.outputs.endpoint
         }
         {
           name: 'AZURE-OPENAI-EMBEDDING-MODEL'
@@ -631,6 +628,10 @@ module keyvault 'br/public:avm/res/key-vault/vault:0.12.1' = {
           name: 'AZURE-SEARCH-ENDPOINT'
           value: 'https://${aiSearchName}.search.windows.net'
         }
+        {
+          name: 'AZURE-AI-AGENT-ENDPOINT'
+          value: aiFoundryAiServices.outputs.aiProjectInfo.apiEndpoint
+        }
     ]
     enableTelemetry: enableTelemetry
   }
@@ -733,10 +734,6 @@ module aiFoundryAiServices 'modules/ai-services.bicep' = if (aiFoundryAIservices
                   name: 'ai-services-dns-zone-cognitiveservices'
                   privateDnsZoneResourceId: avmPrivateDnsZones[dnsZoneIndex.cognitiveServices]!.outputs.resourceId
                 }
-                {
-                  name: 'ai-services-dns-zone-openai'
-                  privateDnsZoneResourceId: avmPrivateDnsZones[dnsZoneIndex.openAI]!.outputs.resourceId
-                }
                 {
                   name: 'ai-services-dns-zone-aiservices'
                   privateDnsZoneResourceId: avmPrivateDnsZones[dnsZoneIndex.aiServices]!.outputs.resourceId
@@ -1111,20 +1108,20 @@ module webSite 'modules/web-sites.bicep' = {
           AZURE_SEARCH_URL_COLUMN: azureSearchUrlColumn
           AZURE_OPENAI_RESOURCE: aiFoundryAiServices.outputs.name
           AZURE_OPENAI_MODEL: gptModelName
-          AZURE_OPENAI_ENDPOINT: aiFoundryAiServices.outputs.endpoints['OpenAI Language Model Instance API']
+          AZURE_OPENAI_ENDPOINT: aiFoundryAiServices.outputs.endpoint
           AZURE_OPENAI_TEMPERATURE: azureOpenAITemperature
           AZURE_OPENAI_TOP_P: azureOpenAITopP
           AZURE_OPENAI_MAX_TOKENS: azureOpenAIMaxTokens
           AZURE_OPENAI_STOP_SEQUENCE: azureOpenAIStopSequence
           AZURE_OPENAI_SYSTEM_MESSAGE: azureOpenAISystemMessage
-          AZURE_OPENAI_PREVIEW_API_VERSION: azureOpenaiAPIVersion
+          AZURE_OPENAI_PREVIEW_API_VERSION: azureAIServicesAPIVersion
           AZURE_OPENAI_STREAM: azureOpenAIStream
           AZURE_SEARCH_QUERY_TYPE: azureSearchQueryType
           AZURE_SEARCH_VECTOR_COLUMNS: azureSearchVectorFields
           AZURE_SEARCH_PERMITTED_GROUPS_COLUMN: azureSearchPermittedGroupsField
           AZURE_SEARCH_STRICTNESS: azureSearchStrictness
           AZURE_OPENAI_EMBEDDING_NAME: embeddingModel
-          AZURE_OPENAI_EMBEDDING_ENDPOINT : aiFoundryAiServices.outputs.endpoints['OpenAI Language Model Instance API']
+          AZURE_OPENAI_EMBEDDING_ENDPOINT : aiFoundryAiServices.outputs.endpoint
           SQLDB_SERVER: sqlServerFqdn
           SQLDB_DATABASE: sqlDbName
           USE_INTERNAL_STREAM: useInternalStream
@@ -1140,7 +1137,7 @@ module webSite 'modules/web-sites.bicep' = {
           USE_AI_PROJECT_CLIENT: useAIProjectClientFlag
           AZURE_AI_AGENT_ENDPOINT: aiFoundryAiServices.outputs.aiProjectInfo.apiEndpoint
           AZURE_AI_AGENT_MODEL_DEPLOYMENT_NAME: gptModelName
-          AZURE_AI_AGENT_API_VERSION: azureOpenaiAPIVersion
+          AZURE_AI_AGENT_API_VERSION: azureAIServicesAPIVersion
           AZURE_SEARCH_CONNECTION_NAME: aiSearchName
           AZURE_CLIENT_ID: userAssignedIdentity.outputs.clientId
         }
@@ -1345,8 +1342,8 @@ output APPLICATIONINSIGHTS_CONNECTION_STRING string = enableMonitoring
   ? applicationInsights!.outputs.connectionString
   : ''
 
-@description('The API version used for the Azure AI Agent service.')
-output AZURE_AI_AGENT_API_VERSION string = azureOpenaiAPIVersion
+  @description('The API version used for the Azure AI Agent service.')
+output AZURE_AI_AGENT_API_VERSION string = azureAIServicesAPIVersion
 
 @description('The endpoint URL of the Azure AI Agent project.')
 output AZURE_AI_AGENT_ENDPOINT string = aiFoundryAiServices.outputs.aiProjectInfo.apiEndpoint
@@ -1373,13 +1370,13 @@ output AZURE_COSMOSDB_DATABASE string = cosmosDbDatabaseName
 output AZURE_COSMOSDB_ENABLE_FEEDBACK string = azureCosmosDbEnableFeedback
 
 @description('The endpoint URL for the Azure OpenAI Embedding model.')
-output AZURE_OPENAI_EMBEDDING_ENDPOINT string = aiFoundryAiServices.outputs.endpoints['OpenAI Language Model Instance API']
+output AZURE_OPENAI_EMBEDDING_ENDPOINT string = aiFoundryAiServices.outputs.endpoint
 
 @description('The name of the Azure OpenAI Embedding model.')
 output AZURE_OPENAI_EMBEDDING_NAME string = embeddingModel
 
 @description('The endpoint URL for the Azure OpenAI service.')
-output AZURE_OPENAI_ENDPOINT string = aiFoundryAiServices.outputs.endpoints['OpenAI Language Model Instance API']
+output AZURE_OPENAI_ENDPOINT string = aiFoundryAiServices.outputs.endpoint
 
 @description('The maximum number of tokens for Azure OpenAI responses.')
 output AZURE_OPENAI_MAX_TOKENS string = azureOpenAIMaxTokens
@@ -1388,7 +1385,7 @@ output AZURE_OPENAI_MAX_TOKENS string = azureOpenAIMaxTokens
 output AZURE_OPENAI_MODEL string = gptModelName
 
 @description('The preview API version for Azure OpenAI.')
-output AZURE_OPENAI_PREVIEW_API_VERSION string = azureOpenaiAPIVersion
+output AZURE_OPENAI_PREVIEW_API_VERSION string = azureAIServicesAPIVersion
 
 @description('The Azure OpenAI resource name.')
 output AZURE_OPENAI_RESOURCE string = aiFoundryAiServices.outputs.name

infra/main.parameters.json

@@ -14,7 +14,7 @@
     "gptModelName": {
       "value": "${AZURE_ENV_MODEL_NAME}"
     },
-    "azureOpenaiAPIVersion": {
+    "azureAIServicesAPIVersion": {
       "value": "${AZURE_ENV_MODEL_VERSION}"
     },
     "gptDeploymentCapacity": {

infra/main.waf.parameters.json

@@ -14,7 +14,7 @@
     "gptModelName": {
       "value": "${AZURE_ENV_MODEL_NAME}"
     },
-    "azureOpenaiAPIVersion": {
+    "azureAIServicesAPIVersion": {
       "value": "${AZURE_ENV_MODEL_VERSION}"
     },
     "gptDeploymentCapacity": {

infra/modules/dependencies.bicep

@@ -129,14 +129,6 @@ var builtInRoleNames = {
     'Microsoft.Authorization/roleDefinitions',
     '3b20f47b-3825-43cb-8114-4bd2201156a8'
   )
-  'Cognitive Services OpenAI Contributor': subscriptionResourceId(
-    'Microsoft.Authorization/roleDefinitions',
-    'a001fd3d-188f-4b5d-821b-7da978bf7442'
-  )
-  'Cognitive Services OpenAI User': subscriptionResourceId(
-    'Microsoft.Authorization/roleDefinitions',
-    '5e0bd9bd-7b93-4f28-af87-19fc36ad61bd'
-  )
   'Cognitive Services QnA Maker Editor': subscriptionResourceId(
     'Microsoft.Authorization/roleDefinitions',
     'f4cc2bf9-21be-47a1-bdf1-5c5804381025'

infra/scripts/index_scripts/create_search_index.py

@@ -31,7 +31,7 @@
     DataLakeServiceClient,
     FileSystemClient,
 )
-from openai import AzureOpenAI
+from azure.ai.projects import AIProjectClient
 
 # Get Azure Key Vault Client
 key_vault_name = "kv_to-be-replaced"  #'nc6262-kv-2fpeafsylfd2e'
@@ -61,6 +61,7 @@
 openai_api_version = secret_client.get_secret("AZURE-OPENAI-PREVIEW-API-VERSION").value
 openai_embedding_model = secret_client.get_secret("AZURE-OPENAI-EMBEDDING-MODEL").value
 account_name = secret_client.get_secret("ADLS-ACCOUNT-NAME").value
+ai_project_endpoint = secret_client.get_secret("AZURE-AI-AGENT-ENDPOINT").value
 
 # Create a search index
 index_client = SearchIndexClient(endpoint=search_endpoint, credential=credential)
@@ -132,15 +133,22 @@
 
 
 # Function: Get Embeddings
-def get_embeddings(text: str, openai_api_base, openai_api_version, azure_token_provider):
+def get_embeddings(text: str, ai_project_endpoint, openai_api_version, credential):
     model_id = openai_embedding_model or "text-embedding-ada-002"
-    client = AzureOpenAI(
+    
+    # Create AI Projects client
+    project_client = AIProjectClient(
+        endpoint=ai_project_endpoint,
+        credential=credential,
         api_version=openai_api_version,
-        azure_endpoint=openai_api_base,
-        azure_ad_token_provider=azure_token_provider,
+    )
+    
+    # Get the OpenAI client from the AI Projects client
+    openai_client = project_client.get_openai_client(
+        api_version=openai_api_version
     )
 
-    embedding = client.embeddings.create(input=text, model=model_id).data[0].embedding
+    embedding = openai_client.embeddings.create(input=text, model=model_id).data[0].embedding
 
     return embedding
 
@@ -260,12 +268,12 @@ def chunk_data(text):
 
         try:
             v_contentVector = get_embeddings(
-                d["content"], openai_api_base, openai_api_version, token_provider
+                d["content"], ai_project_endpoint, openai_api_version, credential
             )
         except:
             time.sleep(30)
             v_contentVector = get_embeddings(
-                d["content"], openai_api_base, openai_api_version, token_provider
+                d["content"], ai_project_endpoint, openai_api_version, credential
             )
 
         docs.append(

src/App/app.py

@@ -6,12 +6,11 @@
 import uuid
 from types import SimpleNamespace
 
-from azure.identity import get_bearer_token_provider
 from backend.helpers.azure_credential_utils import get_azure_credential
 from azure.monitor.opentelemetry import configure_azure_monitor
+from azure.ai.projects import AIProjectClient
 
 # from quart.sessions import SecureCookieSessionInterface
-from openai import AsyncAzureOpenAI
 from opentelemetry import trace
 from opentelemetry.trace import Status, StatusCode
 from quart import (
@@ -129,7 +128,7 @@ async def assets(path):
 if DEBUG.lower() == "true":
     logging.basicConfig(level=logging.DEBUG)
 
-USER_AGENT = "GitHubSampleWebApp/AsyncAzureOpenAI/1.0.0"
+USER_AGENT = "GitHubSampleWebApp/AzureAIProjects/1.0.0"
 
 frontend_settings = {
     "auth_enabled": config.AUTH_ENABLED,
@@ -163,9 +162,9 @@ def should_use_data():
 SHOULD_USE_DATA = should_use_data()
 
 
-# Initialize Azure OpenAI Client
-def init_openai_client(use_data=SHOULD_USE_DATA):
-    azure_openai_client = None
+# Initialize Azure AI Projects Client
+def init_ai_projects_client(use_data=SHOULD_USE_DATA):
+    ai_projects_client = None
     try:
         # API version check
         if (
@@ -177,59 +176,42 @@ def init_openai_client(use_data=SHOULD_USE_DATA):
             )
 
         # Endpoint
-        if not config.AZURE_OPENAI_ENDPOINT and not config.AZURE_OPENAI_RESOURCE:
+        if not config.AI_PROJECT_ENDPOINT:
             raise Exception(
-                "AZURE_OPENAI_ENDPOINT or AZURE_OPENAI_RESOURCE is required"
+                "AI_PROJECT_ENDPOINT is required for Azure AI Projects client"
             )
 
-        endpoint = (
-            config.AZURE_OPENAI_ENDPOINT
-            if config.AZURE_OPENAI_ENDPOINT
-            else f"https://{config.AZURE_OPENAI_RESOURCE}.openai.azure.com/"
+        # Authentication using managed identity
+        credential = get_azure_credential(config.MID_ID)
+        # Create AI Projects client
+        ai_projects_client = AIProjectClient(
+            endpoint=config.AI_PROJECT_ENDPOINT,
+            credential=credential,
+            api_version=config.AZURE_OPENAI_PREVIEW_API_VERSION,
         )
 
-        # Authentication
-        aoai_api_key = config.AZURE_OPENAI_KEY
-        ad_token_provider = None
-        if not aoai_api_key:
-            logging.debug("No AZURE_OPENAI_KEY found, using Azure AD auth")
-            ad_token_provider = get_bearer_token_provider(
-                get_azure_credential(config.MID_ID), "https://cognitiveservices.azure.com/.default"
-            )
-
-        # Deployment
-        deployment = config.AZURE_OPENAI_MODEL
-        if not deployment:
-            raise Exception("AZURE_OPENAI_MODEL is required")
-
-        # Default Headers
-        default_headers = {"x-ms-useragent": USER_AGENT}
-
-        azure_openai_client = AsyncAzureOpenAI(
-            api_version=config.AZURE_OPENAI_PREVIEW_API_VERSION,
-            api_key=aoai_api_key,
-            azure_ad_token_provider=ad_token_provider,
-            default_headers=default_headers,
-            azure_endpoint=endpoint,
+        # Get the OpenAI client from the AI Projects client
+        openai_client = ai_projects_client.get_openai_client(
+            api_version=config.AZURE_OPENAI_PREVIEW_API_VERSION
         )
 
         track_event_if_configured(
-            "AzureOpenAIClientInitialized",
+            "AzureAIProjectsClientInitialized",
             {
                 "status": "success",
-                "endpoint": endpoint,
-                "use_api_key": bool(aoai_api_key),
+                "endpoint": config.AI_PROJECT_ENDPOINT,
+                "use_managed_identity": True,
             },
         )
 
-        return azure_openai_client
+        return openai_client
     except Exception as e:
-        logging.exception("Exception in Azure OpenAI initialization", e)
+        logging.exception("Exception in Azure AI Projects initialization")
         span = trace.get_current_span()
         if span is not None:
             span.record_exception(e)
             span.set_status(Status(StatusCode.ERROR, str(e)))
-        azure_openai_client = None
+        ai_projects_client = None
         raise e
 
 
@@ -265,7 +247,7 @@ def init_cosmosdb_client():
                 },
             )
         except Exception as e:
-            logging.exception("Exception in CosmosDB initialization", e)
+            logging.exception("Exception in CosmosDB initialization")
             span = trace.get_current_span()
             if span is not None:
                 span.record_exception(e)
@@ -521,7 +503,7 @@ async def send_chat_request(request_body, request_headers):
     model_args = prepare_model_args(request_body, request_headers)
 
     try:
-        azure_openai_client = init_openai_client()
+        azure_openai_client = init_ai_projects_client()
         raw_response = (
             await azure_openai_client.chat.completions.with_raw_response.create(
                 **model_args
@@ -1324,7 +1306,7 @@ async def generate_title(conversation_messages):
     messages.append({"role": "user", "content": title_prompt})
 
     try:
-        azure_openai_client = init_openai_client(use_data=False)
+        azure_openai_client = init_ai_projects_client(use_data=False)
         response = await azure_openai_client.chat.completions.create(
             model=config.AZURE_OPENAI_MODEL,
             messages=messages,
@@ -1333,6 +1315,7 @@ async def generate_title(conversation_messages):
         )
 
         title = json.loads(response.choices[0].message.content)["title"]
+
         return title
     except Exception:
         return messages[-2]["content"]