Merge pull request #1367 from adenchen123/GTR-13072

Supports json response format for OpenAI

Author: iceljc

src/Infrastructure/BotSharp.Abstraction/Agents/Models/AgentLlmConfig.cs

@@ -4,12 +4,13 @@ public class AgentLlmConfig
 {
     public AgentLlmConfig() { }
 
-    public AgentLlmConfig(AgentTemplateLlmConfig templateLlmConfig) 
+    public AgentLlmConfig(AgentTemplateConfig templateConfig)
     {
-        Provider = templateLlmConfig.Provider;
-        Model = templateLlmConfig.Model;
-        MaxOutputTokens = templateLlmConfig.MaxOutputTokens;
-        ReasoningEffortLevel = templateLlmConfig.ReasoningEffortLevel;
+        Provider = templateConfig.LlmConfig?.Provider;
+        Model = templateConfig.LlmConfig?.Model;
+        MaxOutputTokens = templateConfig.LlmConfig?.MaxOutputTokens;
+        ReasoningEffortLevel = templateConfig.LlmConfig?.ReasoningEffortLevel;
+        ResponseFormat = templateConfig.ResponseFormat;
     }
 
     /// <summary>
@@ -72,6 +73,13 @@ public AgentLlmConfig(AgentTemplateLlmConfig templateLlmConfig)
     [JsonPropertyName("realtime")]
     [JsonIgnore(Condition = JsonIgnoreCondition.WhenWritingNull)]
     public LlmRealtimeConfig? Realtime { get; set; }
+
+    /// <summary>
+    /// Response format: json, xml, markdown, yaml, etc.
+    /// </summary>
+    [JsonPropertyName("response_format")]
+    [JsonIgnore(Condition = JsonIgnoreCondition.WhenWritingNull)]
+    public string? ResponseFormat { get; set; }
 }
 
 public class LlmImageCompositionConfig : LlmProviderModel

src/Infrastructure/BotSharp.Core/Instructs/Services/InstructService.Execute.cs

@@ -248,10 +248,10 @@ private async Task<InstructResult> RunLlm(
         if (!string.IsNullOrEmpty(templateName))
         {
             prompt = agentService.RenderTemplate(agent, templateName);
-            var templateLlmConfig = agent.Templates?.FirstOrDefault(x => x.Name.IsEqualTo(templateName))?.LlmConfig;
-            if (templateLlmConfig?.IsValid == true)
+            var template = agent.Templates?.FirstOrDefault(x => x.Name.IsEqualTo(templateName));
+            if (template?.LlmConfig?.IsValid == true)
             {
-                llmConfig = new AgentLlmConfig(templateLlmConfig);
+                llmConfig = new AgentLlmConfig(template);
             }
         }
         else

src/Infrastructure/BotSharp.Core/Instructs/Services/InstructService.Instruct.cs

@@ -68,10 +68,9 @@ private async Task<Agent> BuildInnerAgent(InstructOptions? options)
             {
                 var template = agent?.Templates?.FirstOrDefault(x => x.Name.IsEqualTo(options.TemplateName));
                 instruction = BuildInstruction(template?.Content ?? string.Empty, options?.Data ?? []);
-                var templateLlmConfig = template?.LlmConfig;
-                if (templateLlmConfig?.IsValid == true)
+                if (template?.LlmConfig?.IsValid == true)
                 {
-                    llmConfig = new AgentLlmConfig(templateLlmConfig);
+                    llmConfig = new AgentLlmConfig(template);
                 }
             }
         }

src/Plugins/BotSharp.Plugin.OpenAI/Providers/Chat/ChatCompletionProvider.Chat.cs

@@ -618,13 +618,30 @@ private ChatCompletionOptions InitChatCompletionOption(Agent agent)
                         ? tokens
                         : agent.LlmConfig?.MaxOutputTokens ?? LlmConstant.DEFAULT_MAX_OUTPUT_TOKEN;
 
-        return new ChatCompletionOptions()
+        var options = new ChatCompletionOptions()
         {
             Temperature = temperature,
             MaxOutputTokenCount = maxTokens,
             ReasoningEffortLevel = reasoningEffortLevel,
             WebSearchOptions = webSearchOptions
         };
+
+        if (webSearchOptions == null)
+        {
+            options.ResponseFormat = GetChatResponseFormat(agent.LlmConfig?.ResponseFormat);
+        }
+
+        return options;
+    }
+
+    private static ChatResponseFormat? GetChatResponseFormat(string? format)
+    {
+        return format?.ToLower() switch
+        {
+            "json" or "json_object" => ChatResponseFormat.CreateJsonObjectFormat(),
+            "text" => ChatResponseFormat.CreateTextFormat(),
+            _ => null
+        };
     }
 
     /// <summary>

src/Plugins/BotSharp.Plugin.OpenAI/Providers/Chat/ChatCompletionProvider.Response.cs

@@ -432,7 +432,7 @@ private async Task<RoleDialogModel> InnerCreateResponseStreamingAsync(Agent agen
         var options = new CreateResponseOptions(_model, [])
         {
             Temperature = temperature,
-            MaxOutputTokenCount = maxTokens
+            MaxOutputTokenCount = maxTokens,
         };
 
         var reasoningEffortLevel = ParseResponseReasoning(settings?.Reasoning, agent);
@@ -450,6 +450,16 @@ private async Task<RoleDialogModel> InnerCreateResponseStreamingAsync(Agent agen
             }
         }
 
+        // Response format
+        var textFormat = GetResponseTextFormat(agent.LlmConfig?.ResponseFormat);
+        if (textFormat != null)
+        {
+            options.TextOptions = new ResponseTextOptions
+            {
+                TextFormat = textFormat
+            };
+        }
+
         // Prepare instruction and functions
         var renderData = agentService.CollectRenderData(agent);
         var (instruction, functions) = agentService.PrepareInstructionAndFunctions(agent, renderData);
@@ -580,6 +590,16 @@ private string GetResponseApiPrompt(CreateResponseOptions options)
         return sb.ToString();
     }
 
+    private static ResponseTextFormat? GetResponseTextFormat(string? format)
+    {
+        return format?.ToLower() switch
+        {
+            "json" or "json_object" => ResponseTextFormat.CreateJsonObjectFormat(),
+            "text" => ResponseTextFormat.CreateTextFormat(),
+            _ => null
+        };
+    }
+
     private ResponseReasoningEffortLevel? ParseResponseReasoning(ReasoningSetting? settings, Agent agent)
     {
         ResponseReasoningEffortLevel? reasoningEffortLevel = null;