Add JSON schema support to Azure OpenAI models

- Add supportedCapabilities() method to both AzureOpenAiChatModel and AzureOpenAiStreamingChatModel
- Return RESPONSE_FORMAT_JSON_SCHEMA capability when responseFormat type is JSON
- Support request-level ResponseFormat override from ChatRequest
- Enables structured JSON output with schema validation for Azure OpenAI models (e.g., gpt-4o-2024-08-06)

Fixes #1953

Author: geoand

model-providers/openai/azure-openai/runtime/src/main/java/io/quarkiverse/langchain4j/azure/openai/AzureOpenAiChatModel.java

@@ -3,6 +3,7 @@
 import static dev.langchain4j.internal.RetryUtils.withRetry;
 import static dev.langchain4j.internal.Utils.getOrDefault;
 import static dev.langchain4j.internal.ValidationUtils.ensureNotBlank;
+import static dev.langchain4j.model.chat.Capability.RESPONSE_FORMAT_JSON_SCHEMA;
 import static dev.langchain4j.model.openai.internal.OpenAiUtils.aiMessageFrom;
 import static dev.langchain4j.model.openai.internal.OpenAiUtils.finishReasonFrom;
 import static dev.langchain4j.model.openai.internal.OpenAiUtils.toFunctions;
@@ -13,15 +14,18 @@
 import java.net.Proxy;
 import java.time.Duration;
 import java.util.Collections;
+import java.util.HashSet;
 import java.util.List;
 import java.util.Locale;
 import java.util.Map;
+import java.util.Set;
 import java.util.concurrent.ConcurrentHashMap;
 
 import org.jboss.logging.Logger;
 
 import dev.langchain4j.agent.tool.ToolSpecification;
 import dev.langchain4j.data.message.ChatMessage;
+import dev.langchain4j.model.chat.Capability;
 import dev.langchain4j.model.ModelProvider;
 import dev.langchain4j.model.TokenCountEstimator;
 import dev.langchain4j.model.chat.ChatModel;
@@ -74,6 +78,7 @@ public class AzureOpenAiChatModel implements ChatModel {
     private final TokenCountEstimator tokenizer;
     private final ResponseFormat responseFormat;
     private final List<ChatModelListener> listeners;
+    private final Set<Capability> supportedCapabilities;
 
     public AzureOpenAiChatModel(String endpoint,
             String apiVersion,
@@ -128,10 +133,22 @@ public AzureOpenAiChatModel(String endpoint,
                 : ResponseFormat.builder()
                         .type(ResponseFormatType.valueOf(responseFormat.toUpperCase(Locale.ROOT)))
                         .build();
+
+        // Azure OpenAI supports JSON schema for models like gpt-4o-2024-08-06+
+        this.supportedCapabilities = new HashSet<>();
+        if (this.responseFormat != null && ResponseFormatType.JSON.equals(this.responseFormat.type())) {
+            this.supportedCapabilities.add(RESPONSE_FORMAT_JSON_SCHEMA);
+        }
     }
 
-    @Override
     public ChatResponse doChat(ChatRequest chatRequest) {
+        ResponseFormat requestResponseFormat = this.responseFormat;
+
+        // Handle ChatRequest-level ResponseFormat if provided
+        if (chatRequest.responseFormat() != null) {
+            requestResponseFormat = chatRequest.responseFormat();
+        }
+
         List<ChatMessage> messages = chatRequest.messages();
         List<ToolSpecification> toolSpecifications = chatRequest.toolSpecifications();
 
@@ -143,7 +160,7 @@ public ChatResponse doChat(ChatRequest chatRequest) {
                 .maxTokens(maxTokens)
                 .presencePenalty(presencePenalty)
                 .frequencyPenalty(frequencyPenalty)
-                .responseFormat(responseFormat);
+                .responseFormat(requestResponseFormat);
 
         if (toolSpecifications != null && !toolSpecifications.isEmpty()) {
             requestBuilder.functions(toFunctions(toolSpecifications));
@@ -239,6 +256,11 @@ private ChatResponse createModelListenerResponse(String responseId,
                 .build();
     }
 
+    @Override
+    public Set<Capability> supportedCapabilities() {
+        return supportedCapabilities;
+    }
+
     public static Builder builder() {
         return new Builder();
     }

model-providers/openai/azure-openai/runtime/src/main/java/io/quarkiverse/langchain4j/azure/openai/AzureOpenAiStreamingChatModel.java

@@ -4,6 +4,7 @@
 import static dev.langchain4j.internal.Utils.isNullOrBlank;
 import static dev.langchain4j.internal.Utils.isNullOrEmpty;
 import static dev.langchain4j.internal.ValidationUtils.ensureNotBlank;
+import static dev.langchain4j.model.chat.Capability.RESPONSE_FORMAT_JSON_SCHEMA;
 import static dev.langchain4j.model.openai.internal.OpenAiUtils.toFunctions;
 import static dev.langchain4j.model.openai.internal.OpenAiUtils.toOpenAiMessages;
 import static io.quarkiverse.langchain4j.azure.openai.Consts.DEFAULT_USER_AGENT;
@@ -12,16 +13,19 @@
 import java.net.Proxy;
 import java.time.Duration;
 import java.util.Collections;
+import java.util.HashSet;
 import java.util.List;
 import java.util.Locale;
 import java.util.Map;
+import java.util.Set;
 import java.util.concurrent.ConcurrentHashMap;
 import java.util.concurrent.atomic.AtomicReference;
 
 import org.jboss.logging.Logger;
 
 import dev.langchain4j.agent.tool.ToolSpecification;
 import dev.langchain4j.data.message.ChatMessage;
+import dev.langchain4j.model.chat.Capability;
 import dev.langchain4j.model.ModelProvider;
 import dev.langchain4j.model.StreamingResponseHandler;
 import dev.langchain4j.model.TokenCountEstimator;
@@ -78,6 +82,7 @@ public class AzureOpenAiStreamingChatModel implements StreamingChatModel {
     private final TokenCountEstimator tokenizer;
     private final ResponseFormat responseFormat;
     private final List<ChatModelListener> listeners;
+    private final Set<Capability> supportedCapabilities;
 
     public AzureOpenAiStreamingChatModel(String endpoint,
             String apiVersion,
@@ -124,10 +129,22 @@ public AzureOpenAiStreamingChatModel(String endpoint,
                 : ResponseFormat.builder()
                         .type(ResponseFormatType.valueOf(responseFormat.toUpperCase(Locale.ROOT)))
                         .build();
+
+        // Azure OpenAI supports JSON schema for models like gpt-4o-2024-08-06+
+        this.supportedCapabilities = new HashSet<>();
+        if (this.responseFormat != null && ResponseFormatType.JSON.equals(this.responseFormat.type())) {
+            this.supportedCapabilities.add(RESPONSE_FORMAT_JSON_SCHEMA);
+        }
     }
 
     @Override
     public void doChat(ChatRequest chatRequest, StreamingChatResponseHandler handler) {
+        ResponseFormat requestResponseFormat = this.responseFormat;
+
+        // Handle ChatRequest-level ResponseFormat if provided
+        if (chatRequest.responseFormat() != null) {
+            requestResponseFormat = chatRequest.responseFormat();
+        }
 
         List<ChatMessage> messages = chatRequest.messages();
         List<ToolSpecification> toolSpecifications = chatRequest.toolSpecifications();
@@ -140,7 +157,7 @@ public void doChat(ChatRequest chatRequest, StreamingChatResponseHandler handler
                 .maxTokens(maxTokens)
                 .presencePenalty(presencePenalty)
                 .frequencyPenalty(frequencyPenalty)
-                .responseFormat(responseFormat);
+                .responseFormat(requestResponseFormat);
 
         Integer inputTokenCount = tokenizer == null ? null : tokenizer.estimateTokenCountInMessages(messages);
 
@@ -272,6 +289,11 @@ private ChatResponse createModelListenerResponse(String responseId,
                 .build();
     }
 
+    @Override
+    public Set<Capability> supportedCapabilities() {
+        return supportedCapabilities;
+    }
+
     public static Builder builder() {
         return new Builder();
     }