✨ feat(ai): Enhance multimodal support by allowing smart selection of model based on message type.

📄 config: Add smart multimodal option for flexible AI model usage.
🛠️ fix(aiCaller): Update AI caller logic to utilize smart multimodal settings effectively.

apps/ai.js

@@ -352,10 +352,23 @@ async function callAiForResponse(messageData, e, aiConfig) {
       e.user_id
     );
     //更新session
+    let userMessageContent, assistantMessageContent;
+    const usedMultimodal = aiConfig.multimodalEnabled && 
+      (!aiConfig.smartMultimodal || messageData.originalMessages?.some(msg => msg.type === 'image_url'));
+    
+    if (usedMultimodal && messageData.originalMessages) {
+      userMessageContent = messageData.originalMessages.map(msg => {
+        if (msg.type === 'text') return msg.content;
+        if (msg.type === 'image_url') return `[图片消息]`;
+      }).filter(Boolean).join('');
+    } else {
+      userMessageContent = messageData.text;
+    }
+    assistantMessageContent = aiResult.response;
     const newChatHistory = [
       ...chatHistory,
-      { role: 'user', content: messageData.text },
-      { role: 'assistant', content: aiResult.response },
+      { role: 'user', content: userMessageContent },
+      { role: 'assistant', content: assistantMessageContent },
     ];
     SessionManager.updateChatHistory(e.group_id, newChatHistory);
     SessionManager.deactivateSession(e.group_id, e.user_id);

config/ai.json

@@ -10,6 +10,8 @@
   "modelType": "deepseek-ai/DeepSeek-V3.2-Exp",
   "?multimodalEnabled": "是否启用多模态模型模式,启用后将忽略文本模型",
   "multimodalEnabled": false,
+  "?smartMultimodal": "智能多模态模式,开启时只有文字用文本模型,有图片才用多模态模型",
+  "smartMultimodal": false,
   "?multimodalModel": "多模态模型名称",
   "multimodalModel": "Qwen/Qwen2.5-VL-72B-Instruct",
   "?temperature": "聊天温度,可选0-2.0,温度越高创造性越高",

lib/ai/aiCaller.js

@@ -45,7 +45,16 @@ class AiCaller {
       return { success: false, error: 'AI调用器未初始化' };
     }
     try {
-      if (this.config.multimodalEnabled) {
+      if (this.config.smartMultimodal && this.config.multimodalEnabled) {
+        const hasImage = originalMessages.some(msg => msg.type === 'image_url');
+        if (hasImage) {
+          logger.info('[crystelf-ai] 检测到图片，使用多模态模型');
+          return await this.callMultimodalAi(originalMessages, chatHistory, memories, e);
+        } else {
+          logger.info('[crystelf-ai] 纯文本消息，使用文本模型');
+          return await this.callTextAi(prompt, chatHistory, memories, e);
+        }
+      } else if (this.config.multimodalEnabled) {
         return await this.callMultimodalAi(originalMessages, chatHistory, memories, e);
       } else {
         return await this.callTextAi(prompt, chatHistory, memories, e);
@@ -72,9 +81,15 @@ class AiCaller {
     try {
       const fullPrompt = this.buildPrompt(prompt);
       const apiCaller = this.openaiChat;
+    
+      const formattedChatHistory = chatHistory.map(msg => ({
+        role: msg.role,
+        content: msg.content
+      }));
+      
       const result = await apiCaller.callAi({
         prompt: fullPrompt,
-        chatHistory: chatHistory,
+        chatHistory: formattedChatHistory,
         model: this.config.modelType,
         temperature: this.config.temperature,
         customPrompt: await this.getSystemPrompt(e, memories),