refactor: migrate to Azure OpenAI v1 API

Author: sinedied

infra/main.bicep

@@ -11,17 +11,16 @@ param environmentName string
 // Run `az functionapp list-flexconsumption-locations --output table` to get the latest list
 @allowed([
   'northeurope'
-  'southeastasia'
-  'eastasia'
+  'uksouth'
+  'swedencentral'
+  'eastus'
   'eastus2'
   'southcentralus'
-  'australiaeast'
-  'eastus'
   'westus2'
-  'uksouth'
-  'eastus2euap'
   'westus3'
-  'swedencentral'
+  'eastasia'
+  'southeastasia'
+  'australiaeast'
 ])
 param location string
 
@@ -38,7 +37,9 @@ param blobContainerName string = 'blobs'
   'australiaeast'
   'canadaeast'
   'eastus'
+  'westus2'
   'eastus2'
+  'westus3'
   'francecentral'
   'japaneast'
   'northcentralus'
@@ -53,15 +54,14 @@ param blobContainerName string = 'blobs'
   }
 })
 param aiServicesLocation string // Set in main.parameters.json
-param openAiApiVersion string // Set in main.parameters.json
 param defaultModelName string // Set in main.parameters.json
 param defaultModelVersion string // Set in main.parameters.json
 param defaultModelCapacity int // Set in main.parameters.json
 
 // Location is not relevant here as it's only for the built-in api
 // which is not used here. Static Web App is a global service otherwise
 @description('Location for the Static Web App')
-@allowed(['westus2', 'centralus', 'eastus2', 'westeurope', 'eastasia', 'eastasiastage'])
+@allowed(['westus2', 'centralus', 'eastus2', 'westeurope', 'eastasia'])
 @metadata({
   azd: {
     type: 'location'
@@ -87,7 +87,7 @@ var burgerApiResourceName = '${abbrs.webSitesFunctions}burger-api-${resourceToke
 var burgerMcpResourceName = '${abbrs.webSitesFunctions}burger-mcp-${resourceToken}'
 var agentApiResourceName = '${abbrs.webSitesFunctions}agent-api-${resourceToken}'
 var storageAccountName = '${abbrs.storageStorageAccounts}${resourceToken}'
-var openAiUrl = 'https://${openAi.outputs.name}.openai.azure.com'
+var openAiUrl = 'https://${openAi.outputs.name}.openai.azure.com/openai/v1'
 var storageUrl = 'https://${storage.outputs.name}.blob.${environment().suffixes.storage}'
 var burgerApiUrl = 'https://${burgerApiFunction.outputs.defaultHostname}'
 var burgerMcpUrl = 'https://${burgerMcpFunction.outputs.defaultHostname}/mcp'
@@ -265,10 +265,8 @@ module agentApiFunctionSettings 'br/public:avm/res/web/site/config:0.1.0' = {
     appName: agentApiFunction.outputs.name
     properties: {
       AZURE_COSMOSDB_NOSQL_ENDPOINT: cosmosDb.outputs.endpoint
-      AZURE_OPENAI_ENDPOINT: openAiUrl
-      AZURE_OPENAI_CHAT_DEPLOYMENT_NAME: defaultModelName
-      AZURE_OPENAI_INSTANCE_NAME: openAi.outputs.name
-      AZURE_OPENAI_API_VERSION: openAiApiVersion
+      AZURE_OPENAI_API_ENDPOINT: openAiUrl
+      AZURE_OPENAI_MODEL: defaultModelName
       BURGER_MCP_ENDPOINT: burgerMcpUrl
     }
     storageAccountResourceId: storage.outputs.resourceId
@@ -444,7 +442,7 @@ module monitoring 'br/public:avm/ptn/azd/monitoring:0.2.1' = {
   }
 }
 
-// TODO: migrate to Foundry
+// TODO: migrate to AI Foundry
 module openAi 'br/public:avm/res/cognitive-services/account:0.13.2' = {
   name: 'openai'
   scope: resourceGroup
@@ -625,9 +623,6 @@ output AZURE_STORAGE_CONTAINER_NAME string = blobContainerName
 output AZURE_COSMOSDB_NOSQL_ENDPOINT string = cosmosDb.outputs.endpoint
 
 output AZURE_OPENAI_API_ENDPOINT string = openAiUrl
-output AZURE_OPENAI_API_INSTANCE_NAME string = openAi.outputs.name
-output AZURE_OPENAI_API_DEPLOYMENT_NAME string = defaultModelName
-output AZURE_OPENAI_API_VERSION string = openAiApiVersion
 output AZURE_OPENAI_MODEL string = defaultModelName
 
 output GENAISCRIPT_DEFAULT_MODEL string = 'azure:${defaultModelName}'

packages/agent-api/scripts/update-local-settings.mjs

@@ -27,10 +27,8 @@ const settingsFilePath = path.join(__dirname, '../local.settings.json');
 console.log('Setting Azure AI service values...');
 settings = {
   ...settings,
-  AZURE_OPENAI_API_VERSION: process.env.AZURE_OPENAI_API_VERSION,
-  AZURE_OPENAI_API_INSTANCE_NAME: process.env.AZURE_OPENAI_API_INSTANCE_NAME,
-  AZURE_OPENAI_API_DEPLOYMENT_NAME: process.env.AZURE_OPENAI_API_DEPLOYMENT_NAME,
   AZURE_OPENAI_API_ENDPOINT: process.env.AZURE_OPENAI_API_ENDPOINT,
+  AZURE_OPENAI_MODEL: process.env.AZURE_OPENAI_API_DEPLOYMENT_NAME,
 };
 
 console.log('Setting Cosmos DB service values...');

packages/agent-api/src/functions/chats-post.ts

@@ -2,7 +2,7 @@ import { Readable } from 'node:stream';
 import { randomUUID } from 'node:crypto';
 import { HttpRequest, InvocationContext, HttpResponseInit, app } from '@azure/functions';
 import { AIChatCompletionRequest, AIChatCompletionDelta } from '@microsoft/ai-chat-protocol';
-import { AzureChatOpenAI } from '@langchain/openai';
+import { ChatOpenAI } from '@langchain/openai';
 import { AzureCosmsosDBNoSQLChatMessageHistory } from '@langchain/azure-cosmosdb';
 import { createReactAgent } from "@langchain/langgraph/prebuilt";
 import { AIMessage, HumanMessage } from '@langchain/core/messages';
@@ -40,7 +40,7 @@ Make sure the last question ends with ">>".
 - When using images in answers, use tables if you are showing multiple images in a list, to make the layout cleaner. Otherwise, try using a single image at the bottom of your answer.
 `;
 
-const titleSystemPrompt = `Create a title for this chat session, based on the user question. The title should be less than 32 characters. Do NOT use double-quotes.`;
+const titleSystemPrompt = `Create a title for this chat session, based on the user question. The title should be less than 32 characters. Do NOT use double-quotes. The title should be concise, descriptive, and catchy. Respond with only the title, no other text.`;
 
 export async function postChats(request: HttpRequest, context: InvocationContext): Promise<HttpResponseInit> {
   const azureOpenAiEndpoint = process.env.AZURE_OPENAI_API_ENDPOINT;
@@ -83,14 +83,25 @@ export async function postChats(request: HttpRequest, context: InvocationContext
       };
     }
 
-    const credentials = getCredentials();
-    const azureADTokenProvider = getAzureOpenAiTokenProvider();
-
-    const model = new AzureChatOpenAI({ azureADTokenProvider, streaming: true });
+    const model = new ChatOpenAI({
+      configuration: {
+        baseURL: azureOpenAiEndpoint,
+        async fetch(url, init = {}) {
+          const token = await getAzureOpenAiTokenProvider()();
+          const headers = new Headers(init.headers);
+          headers.set('Authorization', `Bearer ${token}`);
+          return fetch(url, { ...init, headers });
+        },
+      },
+      modelName: process.env.AZURE_OPENAI_MODEL ?? 'gpt-5-mini',
+      streaming: true,
+      useResponsesApi: true,
+      apiKey: 'not_used'
+    });
     const chatHistory = new AzureCosmsosDBNoSQLChatMessageHistory({
       sessionId,
       userId,
-      credentials,
+      credentials: getCredentials(),
       containerName: 'history',
       databaseName: 'historyDB',
     });
@@ -156,8 +167,8 @@ export async function postChats(request: HttpRequest, context: InvocationContext
         ['system', titleSystemPrompt],
         ['human', question],
       ]);
-      context.log(`Title for session: ${response.content as string}`);
-      chatHistory.setContext({ title: response.content });
+      context.log(`Title for session: ${response.text}`);
+      chatHistory.setContext({ title: response.text });
     }
 
     return {
@@ -186,16 +197,16 @@ async function* createJsonStream(chunks: AsyncIterable<StreamEvent>, sessionId:
     const data = chunk.data;
     let responseChunk: AIChatCompletionDelta | undefined;
 
-    if (chunk.event === 'on_chain_end' && chunk.name === 'RunnableSequence') {
+    if (chunk.event === 'on_chain_end' && chunk.name === 'RunnableSequence' && data.output?.content.length > 0) {
       // End of our agentic chain
-      const content = data?.output?.content ?? '';
+      const content = data?.output.content[0].text ?? '';
       await onComplete(content);
 
     } else if (chunk.event === 'on_chat_model_stream' && data.chunk.content.length > 0) {
       // Streaming response from the LLM
       responseChunk = {
         delta: {
-          content: data.chunk.content,
+          content: data.chunk.content[0].text,
           role: 'assistant',
         },
         context: {