Update main.ts

main.ts

@@ -1,4 +1,3 @@
-// main.ts
 import { serve } from "https://deno.land/std@0.208.0/http/server.ts";
 
 // --- 常量定义 ---
@@ -13,7 +12,7 @@ interface OpenAIMessage {
     type: string;
     text?: string;
     image_url?: { url: string };
-    document?: { url: string; type: string };
+    document?: { url: string; type: string }; // 支持多种文档类型
   }>;
 }
 
@@ -25,12 +24,13 @@ interface OpenAIRequest {
   stream?: boolean;
 }
 
+// [新增] OpenAI TTS 请求接口定义
 interface OpenAITTSRequest {
-    model: string;
-    input: string;
+    model: string; // e.g., 'tts-1', 'tts-1-hd'
+    input: string; // The text to synthesize
     voice: 'alloy' | 'echo' | 'fable' | 'onyx' | 'nova' | 'shimmer';
-    response_format?: 'mp3' | 'opus' | 'aac' | 'flac';
-    speed?: number;
+    response_format?: 'mp3' | 'opus' | 'aac' | 'flac'; // Google TTS returns MP3, so we'll ignore others for now
+    speed?: number; // Not directly supported by Gemini TTS, will be ignored
 }
 
 
@@ -62,57 +62,121 @@ class GoogleAIService {
     return key;
   }
   
+  // --- [新增] TTS 功能 ---
+
+  /**
+   * 映射 OpenAI 的语音名称到 Google Gemini TTS 的预置语音名称。
+   * 参考: https://ai.google.dev/gemini-api/docs/text-to-speech#supported_voices
+   */
   private getGoogleVoice(openAIVoice: string): string {
     const voiceMap: { [key: string]: string } = {
-        'alloy': 'Kore', 'echo': 'Sal', 'fable': 'Polly', 'onyx': 'Onyx',
-        'nova': 'Sparkle', 'shimmer': 'Luna', 'default': 'Kore'
+        'alloy': 'Kore',      // A good default, versatile voice
+        'echo': 'Sal',        // Another male voice option
+        'fable': 'Polly',     // Female, narrative style
+        'onyx': 'Onyx',       // Deep, male voice
+        'nova': 'Sparkle',    // Energetic female voice
+        'shimmer': 'Luna',    // Gentle female voice
+        // Fallback to a default if the voice is not in the map
+        'default': 'Kore'
     };
     return voiceMap[openAIVoice] || voiceMap['default'];
   }
 
+  /**
+   * [新增] 调用 Google Gemini TTS API 生成语音。
+   * @param input - 要转换为语音的文本。
+   * @param model - 请求的模型（在Google端，我们硬编码为TTS模型）。
+   * @param voice - OpenAI 格式的语音名称。
+   * @returns 返回包含音频数据的 ArrayBuffer。
+   */
   async generateSpeech(input: string, model: string, voice: string): Promise<ArrayBuffer> {
     const apiKey = this.getNextApiKey();
     const googleVoice = this.getGoogleVoice(voice);
+    // 根据 curl 命令，模型是固定的 TTS 模型
     const ttsModel = "gemini-2.5-flash-preview-tts"; 
 
     console.log(`Generating speech with model: ${ttsModel}, voice: ${googleVoice} (mapped from OpenAI's '${voice}')`);
+
     const requestBody = {
-        "contents": [{"parts":[{"text": input}]}],
-        "generationConfig": {"responseModalities": ["AUDIO"],"speechConfig": {"voiceConfig": {"prebuiltVoiceConfig": {"voiceName": googleVoice}}}},
+        "contents": [{
+            "parts":[{
+                "text": input
+            }]
+        }],
+        "generationConfig": {
+            "responseModalities": ["AUDIO"],
+            "speechConfig": {
+                "voiceConfig": {
+                    "prebuiltVoiceConfig": {
+                        "voiceName": googleVoice
+                    }
+                }
+            }
+        },
         "model": ttsModel,
     };
-    const response = await fetch(`https://generativelanguage.googleapis.com/v1beta/models/${ttsModel}:generateContent?key=${apiKey}`,
-        { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify(requestBody) });
+
+    const response = await fetch(
+        `https://generativelanguage.googleapis.com/v1beta/models/${ttsModel}:generateContent?key=${apiKey}`,
+        {
+            method: "POST",
+            headers: { "Content-Type": "application/json" },
+            body: JSON.stringify(requestBody),
+        }
+    );
+
     if (!response.ok) {
         const errorBody = await response.json().catch(() => response.text());
         const errorMessage = errorBody?.error?.message || JSON.stringify(errorBody);
+        console.error(`Google TTS API Error: ${response.status} - ${errorMessage}`);
         throw new Error(`Google TTS API request failed with status ${response.status}: ${errorMessage}`);
     }
+
     const data = await response.json();
+    
+    // 提取 base64 编码的音频数据
     const audioContentBase64 = data.candidates?.[0]?.content?.parts?.[0]?.inlineData?.data;
-    if (!audioContentBase64) throw new Error("No audio data returned from Google API.");
+    if (!audioContentBase64) {
+        throw new Error("No audio data returned from Google API. The response might be blocked or empty.");
+    }
+    
+    // 将 base64 字符串解码为二进制数据 (ArrayBuffer)
     const binaryString = atob(audioContentBase64);
-    const bytes = new Uint8Array(binaryString.length);
-    for (let i = 0; i < binaryString.length; i++) {
+    const len = binaryString.length;
+    const bytes = new Uint8Array(len);
+    for (let i = 0; i < len; i++) {
         bytes[i] = binaryString.charCodeAt(i);
     }
     return bytes.buffer;
   }
   
+  // --- 现有代码保持不变 ---
+  
   async fetchOfficialModels(): Promise<any[]> {
     const now = Date.now();
-    if (this.cachedModels.length > 0 && (now - this.modelsLastFetch) < MODELS_CACHE_DURATION) return this.cachedModels;
+    if (this.cachedModels.length > 0 && (now - this.modelsLastFetch) < MODELS_CACHE_DURATION) {
+      return this.cachedModels;
+    }
+
     const apiKey = this.getNextApiKey();
     try {
-      const response = await fetch(`https://generativelanguage.googleapis.com/v1beta/models?key=${apiKey}`);
+      const response = await fetch(
+        `https://generativelanguage.googleapis.com/v1beta/models?key=${apiKey}`,
+        { method: "GET", headers: { "Content-Type": "application/json" } }
+      );
+
       if (!response.ok) {
         console.warn(`Failed to fetch models from Google AI: ${response.status}. Using fallback models.`);
         return this.getFallbackModels();
       }
+
       const data = await response.json();
       if (data.models && Array.isArray(data.models)) {
-        this.cachedModels = data.models.filter((model: any) => model.supportedGenerationMethods?.includes('generateContent'));
+        this.cachedModels = data.models.filter((model: any) =>
+          model.supportedGenerationMethods?.includes('generateContent')
+        );
         this.modelsLastFetch = now;
+        console.log(`Fetched ${this.cachedModels.length} models from Google AI`);
         return this.cachedModels;
       }
       return this.getFallbackModels();
@@ -124,188 +188,338 @@ class GoogleAIService {
 
   private getFallbackModels(): any[] {
     return [
-      { name: "models/gemini-1.5-pro", displayName: "Gemini 1.5 Pro", description: "Mid-size multimodal model that supports up to 1 million tokens.", supportedGenerationMethods: ["generateContent"], maxTokens: 1000000 },
-      { name: "models/gemini-1.5-flash", displayName: "Gemini 1.5 Flash", description: "Fast and versatile multimodal model.", supportedGenerationMethods: ["generateContent"], maxTokens: 1000000 },
-      { name: "models/gemini-2.0-flash-preview-image-generation", displayName: "Gemini 2.0 Flash Image Generation", description: "Advanced model for generating images.", supportedGenerationMethods: ["generateContent"], maxTokens: 100000 },
-      { name: "models/gemini-2.5-flash-preview-tts", displayName: "Gemini 2.5 Flash TTS", description: "Text-to-speech model.", id: "gemini-2.5-flash-preview-tts" }
+      { name: "models/gemini-1.5-pro", displayName: "Gemini 1.5 Pro", description: "Mid-size multimodal model that supports up to 1 million tokens, images, and documents (PDF, TXT, MD)", supportedGenerationMethods: ["generateContent"], maxTokens: 1000000, supportsDocuments: true },
+      { name: "models/gemini-1.5-flash", displayName: "Gemini 1.5 Flash", description: "Fast and versatile multimodal model for diverse tasks, supports images and documents (PDF, TXT, MD)", supportedGenerationMethods: ["generateContent"], maxTokens: 1000000, supportsDocuments: true },
+      { name: "models/gemini-2.0-flash-preview-image-generation", displayName: "Gemini 2.0 Flash Image Generation", description: "Advanced model for generating and editing high-quality images with text and image outputs", supportedGenerationMethods: ["generateContent"], maxTokens: 100000, capabilities: ["text", "image_generation", "image_editing"] },
+      // [新增] 在模型列表中添加TTS模型，使其在 /v1/models 接口可见
+      { name: "models/gemini-2.5-flash-preview-tts", displayName: "Gemini 2.5 Flash TTS", description: "Text-to-speech model for generating high-quality audio.", supportedGenerationMethods: ["generateContent"], id: "gemini-2.5-flash-preview-tts" }
     ];
   }
 
   public isVisionModel = (modelName: string): boolean => modelName.toLowerCase().includes('vision') || modelName.toLowerCase().includes('pro');
-  public isImageGenerationModel = (modelName: string): boolean => modelName.includes('image-generation');
-  public isImageEditingModel = (modelName: string): boolean => modelName.includes('image-generation');
-  public isDocumentModel = (modelName: string): boolean => modelName.toLowerCase().includes('gemini-1.5');
+  public isImageGenerationModel = (modelName: string): boolean => modelName.includes('image-generation') || modelName === 'gemini-2.0-flash-preview-image-generation';
+  public isImageEditingModel = (modelName: string): boolean => modelName.includes('image-generation') || modelName === 'gemini-2.0-flash-preview-image-generation';
+  public isDocumentModel = (modelName: string): boolean => modelName.toLowerCase().includes('gemini-1.5') || modelName.toLowerCase().includes('pro') || modelName.toLowerCase().includes('flash');
 
   private getDocumentType(url: string): string {
     const lowerUrl = url.toLowerCase();
     if (lowerUrl.startsWith('data:application/pdf') || lowerUrl.includes('.pdf')) return 'pdf';
     if (lowerUrl.startsWith('data:text/plain') || lowerUrl.includes('.txt')) return 'txt';
     if (lowerUrl.startsWith('data:text/markdown') || lowerUrl.includes('.md')) return 'md';
+    if (lowerUrl.startsWith('data:application/msword') || lowerUrl.includes('.doc')) return 'doc';
+    if (lowerUrl.startsWith('data:application/vnd.openxmlformats-officedocument.wordprocessingml.document') || lowerUrl.includes('.docx')) return 'docx';
     return 'unknown';
   }
 
   private extractDocumentData(documentUrl: string): { mimeType: string; data: string; text?: string; docType: string } {
     const docType = this.getDocumentType(documentUrl);
-    if (!documentUrl.startsWith("data:")) throw new Error("Document must be a base64 data URL.");
-    const [mimeInfo, base64Data] = documentUrl.split(",");
-    if (base64Data.length * 0.75 > MAX_DOCUMENT_SIZE_BYTES) throw new Error(`Document size exceeds ${MAX_DOCUMENT_SIZE_MB}MB.`);
+
+    if (!documentUrl.startsWith("data:")) {
+      if (documentUrl.startsWith("http")) {
+        throw new Error("Document URL downloads are not supported. Please provide base64 encoded data URLs.");
+      }
+      throw new Error("Document must be provided as a standard base64 data URL (e.g., 'data:application/pdf;base64,...').");
+    }
+
+    const parts = documentUrl.split(",");
+    if (parts.length !== 2) {
+        throw new Error("Invalid data URL format for document. Expected 'data:[mime];base64,[data]'.");
+    }
+    const [mimeInfo, base64Data] = parts;
+
+    const approxSizeInBytes = base64Data.length * 0.75;
+    if (approxSizeInBytes > MAX_DOCUMENT_SIZE_BYTES) {
+        throw new Error(`Document size (${(approxSizeInBytes / 1024 / 1024).toFixed(2)}MB) exceeds the ${MAX_DOCUMENT_SIZE_MB}MB limit.`);
+    }
+
     const mimeType = mimeInfo.split(":")[1]?.split(";")[0] || 'application/octet-stream';
+
     if (docType === 'txt' || docType === 'md') {
-      return { mimeType, data: base64Data, text: atob(base64Data), docType };
+      try {
+        const textContent = atob(base64Data);
+        return { mimeType, data: base64Data, text: textContent, docType };
+      } catch (error) {
+        console.error(`Failed to decode base64 content for ${docType}:`, error);
+        throw new Error(`Invalid base64 encoding for ${docType} document.`);
+      }
     }
-    return { mimeType: docType === 'pdf' ? 'application/pdf' : mimeType, data: base64Data, docType };
+    
+    const finalMimeType = docType === 'pdf' ? 'application/pdf' : mimeType;
+    return { mimeType: finalMimeType, data: base64Data, docType };
   }
   
   private extractImageData(imageUrl: string): { mimeType: string; data: string } {
     if (imageUrl.startsWith("data:image/")) {
       const [mimeInfo, base64Data] = imageUrl.split(",");
-      return { mimeType: mimeInfo.split(":")[1].split(";")[0], data: base64Data };
+      const mimeType = mimeInfo.split(":")[1].split(";")[0];
+      return { mimeType, data: base64Data };
     } else if (imageUrl.startsWith("http")) {
-      throw new Error("URL images are not supported. Please use base64 data URLs.");
+      throw new Error("URL images are not supported yet. Please provide base64 encoded images.");
+    } else {
+      return { mimeType: "image/jpeg", data: imageUrl };
     }
-    return { mimeType: "image/jpeg", data: imageUrl };
   }
 
-  private buildGoogleContent(messages: OpenAIMessage[]) {
-    return messages.map(msg => {
-      const role = msg.role === "assistant" ? "model" : "user";
-      if (typeof msg.content === "string") {
-        return { role, parts: [{ text: msg.content }] };
-      }
-      const parts = msg.content.map(part => {
-        if (part.type === "text") return { text: part.text };
-        if (part.type === "image_url" && part.image_url) {
-          const { mimeType, data } = this.extractImageData(part.image_url.url);
-          return { inlineData: { mimeType, data } };
+  async generateContentWithDocument(messages: OpenAIMessage[], modelName: string): Promise<string> {
+    const apiKey = this.getNextApiKey();
+    const fullModelName = modelName.startsWith('models/') ? modelName : `models/${modelName}`;
+    const documentModel = this.isDocumentModel(fullModelName) ? fullModelName : 'models/gemini-1.5-pro-latest';
+
+    console.log(`Processing document with model: ${documentModel}`);
+
+    let contents;
+    try {
+      contents = messages.map(msg => {
+        if (typeof msg.content === "string") {
+          return { role: msg.role === "assistant" ? "model" : "user", parts: [{ text: msg.content }] };
         }
-        if (part.type === "document" && part.document) {
-          const docData = this.extractDocumentData(part.document.url);
-          if (docData.docType === 'txt' || docData.docType === 'md') {
-            return { text: `${docData.docType === 'md' ? 'Markdown' : 'Text'} document content:\n${docData.text}` };
+
+        const messageParts = msg.content.map(part => {
+          if (part.type === "text") return { text: part.text };
+
+          if (part.type === "image_url" && part.image_url) {
+            const { mimeType, data } = this.extractImageData(part.image_url.url);
+            return { inlineData: { mimeType, data } };
           }
-          if (docData.docType === 'pdf') {
-            return { inlineData: { mimeType: docData.mimeType, data: docData.data } };
+
+          if (part.type === "document" && part.document) {
+            const docData = this.extractDocumentData(part.document.url);
+            console.log(`Processing document: ${docData.docType}, mime: ${docData.mimeType}, size: ${(docData.data.length * 0.75 / 1024).toFixed(2)} KB`);
+
+            if (docData.docType === 'txt' || docData.docType === 'md') {
+              const prefix = docData.docType === 'md' ? 'Markdown document content:\n' : 'Text document content:\n';
+              return { text: `${prefix}${docData.text}` };
+            }
+            if (docData.docType === 'pdf') {
+              return { inlineData: { mimeType: docData.mimeType, data: docData.data } };
+            }
+            return { text: `[Document type '${docData.docType}' is not supported for direct processing. Please convert to PDF, TXT, or MD.]` };
           }
-        }
-        return { text: "" };
+          return { text: "" };
+        });
+        return { role: msg.role === "assistant" ? "model" : "user", parts: messageParts.filter(p => p.text || p.inlineData) };
       });
-      return { role, parts: parts.filter(p => p && (p.text || p.inlineData)) };
-    });
-  }
+    } catch (error) {
+      throw error;
+    }
 
-  async generateContent(messages: OpenAIMessage[], modelName: string): Promise<string> {
-    const apiKey = this.getNextApiKey();
-    const fullModelName = modelName.startsWith('models/') ? modelName : `models/${modelName}`;
-    const contents = this.buildGoogleContent(messages);
-    const requestBody = { contents, generationConfig: { temperature: 0.7, maxOutputTokens: 8192 } };
+    const requestBody = {
+      contents,
+      generationConfig: { temperature: 0.7, maxOutputTokens: 8192 }
+    };
 
-    const response = await fetch(`https://generativelanguage.googleapis.com/v1beta/${fullModelName}:generateContent?key=${apiKey}`,
-        { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify(requestBody) });
+    const response = await fetch(
+      `https://generativelanguage.googleapis.com/v1beta/${documentModel}:generateContent?key=${apiKey}`,
+      {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify(requestBody),
+      }
+    );
 
     if (!response.ok) {
-        const errorText = await response.text();
-        throw new Error(`Google AI API error: ${response.status} - ${errorText}`);
+      const errorBody = await response.json().catch(() => response.text());
+      const errorMessage = errorBody?.error?.message || JSON.stringify(errorBody);
+      console.error(`Google API Error: ${response.status} - ${errorMessage}`);
+      throw new Error(`Google API request failed with status ${response.status}: ${errorMessage}`);
     }
+
     const data = await response.json();
-    if (data.promptFeedback?.blockReason) {
-        throw new Error(`Request blocked by Google. Reason: ${data.promptFeedback.blockReason}`);
+    const promptFeedback = data.promptFeedback;
+    if (promptFeedback && promptFeedback.blockReason) {
+      const reason = promptFeedback.blockReason;
+      const safetyRatings = promptFeedback.safetyRatings?.map((r: any) => `${r.category}: ${r.probability}`).join(', ') || 'N/A';
+      throw new Error(`Request blocked by Google API. Reason: ${reason}. Safety Ratings: [${safetyRatings}]`);
     }
-    const candidate = data.candidates?.[0];
-    if (!candidate) throw new Error("No response generated from Google AI.");
-    if (candidate.finishReason === "SAFETY") throw new Error("Response blocked for safety reasons.");
-    return candidate.content?.parts?.[0]?.text || "No text response generated.";
+
+    if (!data.candidates || data.candidates.length === 0) {
+      throw new Error("No response generated for document content. The content might be empty or unreadable.");
+    }
+
+    const candidate = data.candidates[0];
+    if (candidate.finishReason === "SAFETY") {
+        throw new Error("Response blocked due to safety filters. Check content for sensitive topics.");
+    }
+    if (candidate.finishReason === "RECITATION") {
+        throw new Error("Response blocked due to recitation policy. The model's output was too similar to a copyrighted source.");
+    }
+
+    return candidate.content?.parts[0]?.text || "Document processed, but no text response was generated.";
   }
+  
+  async generateContent(messages: OpenAIMessage[], modelName: string, enableSearch: boolean = false): Promise<string> {
+    const hasDocument = messages.some(msg => Array.isArray(msg.content) && msg.content.some(part => part.type === "document"));
+    if (hasDocument) {
+      return await this.generateContentWithDocument(messages, modelName);
+    }
 
-  /**
-   * [新增] 真正的流式内容生成函数
-   * 使用 Google 的 streamGenerateContent 端点进行流式响���处理
-   */
-  async * streamGenerateContent(messages: OpenAIMessage[], modelName: string): AsyncGenerator<string> {
     const apiKey = this.getNextApiKey();
     const fullModelName = modelName.startsWith('models/') ? modelName : `models/${modelName}`;
-    const contents = this.buildGoogleContent(messages);
-    const requestBody = { contents, generationConfig: { temperature: 0.7, maxOutputTokens: 8192 } };
 
-    // [关键] 使用 :streamGenerateContent 端点
-    const response = await fetch(`https://generativelanguage.googleapis.com/v1beta/${fullModelName}:streamGenerateContent?key=${apiKey}`,
-        { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify(requestBody) });
+    const contents = messages.map(msg => {
+      if (typeof msg.content === "string") {
+        return { role: msg.role === "assistant" ? "model" : "user", parts: [{ text: msg.content }] };
+      } else {
+        const messageParts = msg.content.map(part => {
+          if (part.type === "text") {
+            return { text: part.text };
+          } else if (part.type === "image_url" && part.image_url) {
+            const imageData = part.image_url.url;
+            if (imageData.startsWith("data:image/")) {
+              const { mimeType, data } = this.extractImageData(imageData);
+              return { inlineData: { mimeType, data } };
+            } else {
+              return { fileData: { mimeType: "image/jpeg", fileUri: imageData } };
+            }
+          }
+          return { text: "" };
+        });
+        return { role: msg.role === "assistant" ? "model" : "user", parts: messageParts };
+      }
+    });
 
-    if (!response.ok || !response.body) {
-      const errorText = await response.text();
-      throw new Error(`Google AI streaming API error: ${response.status} - ${errorText}`);
+    const requestBody: any = {
+      contents,
+      generationConfig: { temperature: 0.7, maxOutputTokens: 4096 }
+    };
+    if (enableSearch) {
+      requestBody.tools = [{ googleSearchRetrieval: {} }];
     }
-    
-    // [关键] 读取并解析流
-    const reader = response.body.getReader();
-    const decoder = new TextDecoder();
-    let buffer = "";
 
-    while (true) {
-      const { done, value } = await reader.read();
-      if (done) break;
-      
-      buffer += decoder.decode(value, { stream: true });
-      
-      // Google 的流式响应可能在一个数据包里包含多个以 "data: " 开头的 JSON 对象
-      const lines = buffer.split('\n');
-      buffer = lines.pop() || ''; // Keep the last, possibly incomplete, line in buffer
-
-      for (const line of lines) {
-          if (line.startsWith('data: ')) {
-              try {
-                  const jsonStr = line.substring(6);
-                  const chunk = JSON.parse(jsonStr);
-                  
-                  if (chunk.error) throw new Error(`Google API stream error: ${chunk.error.message}`);
-
-                  const text = chunk.candidates?.[0]?.content?.parts?.[0]?.text;
-                  if (text) {
-                      yield text; // 产生一个文本块
-                  }
-              } catch (e) {
-                  console.warn("Could not parse stream chunk:", line, e.message);
-              }
-          }
-      }
+    const response = await fetch(
+      `https://generativelanguage.googleapis.com/v1beta/${fullModelName}:generateContent?key=${apiKey}`,
+      { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify(requestBody) }
+    );
+
+    if (!response.ok) {
+      const errorText = await response.text();
+      throw new Error(`Google AI API error: ${response.status} - ${errorText}`);
     }
+    const data = await response.json();
+    if (!data.candidates || data.candidates.length === 0) {
+      throw new Error("No response generated from Google AI");
+    }
+    const candidate = data.candidates[0];
+    if (candidate.finishReason === "SAFETY") {
+      throw new Error("Response blocked due to safety filters");
+    }
+    return candidate.content?.parts[0]?.text || "No response generated";
   }
 
-  async generateOrEditImage(prompt: string, modelName: string, inputImages?: any[]): Promise<string> {
+  async generateOrEditImageWithGemini(prompt: string, modelName: string = "gemini-2.0-flash-preview-image-generation", inputImage?: { mimeType: string; data: string }): Promise<{ text?: string; imageBase64?: string; imageUrl?: string }> {
     const apiKey = this.getNextApiKey();
     const fullModelName = modelName.startsWith('models/') ? modelName : `models/${modelName}`;
     const requestParts: any[] = [{ text: prompt }];
-    let inputImage;
-    if (inputImages && inputImages.length > 0) {
-        inputImage = this.extractImageData(inputImages[0].url);
-        requestParts.push({ inline_data: { mime_type: inputImage.mimeType, data: inputImage.data } });
+
+    if (inputImage) {
+      requestParts.push({ inline_data: { mime_type: inputImage.mimeType, data: inputImage.data } });
+      console.log(`Editing image with model: ${fullModelName}`);
+    } else {
+      console.log(`Generating image with model: ${fullModelName}`);
     }
 
     const requestBody = {
       contents: [{ parts: requestParts }],
       generationConfig: { responseModalities: ["TEXT", "IMAGE"], temperature: 0.7 }
     };
-    const response = await fetch(`https://generativelanguage.googleapis.com/v1beta/${fullModelName}:generateContent?key=${apiKey}`,
-        { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify(requestBody) });
 
-    if (!response.ok) throw new Error(`Image processing failed: ${response.status} - ${await response.text()}`);
+    const response = await fetch(
+      `https://generativelanguage.googleapis.com/v1beta/${fullModelName}:generateContent?key=${apiKey}`,
+      { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify(requestBody) }
+    );
+
+    if (!response.ok) {
+      const errorText = await response.text();
+      throw new Error(`Image ${inputImage ? 'editing' : 'generation'} failed: ${response.status} - ${errorText}`);
+    }
     const data = await response.json();
-    const candidate = data.candidates?.[0];
-    if (!candidate) throw new Error("No image response from Google.");
-    if (candidate.finishReason === "SAFETY") throw new Error("Image request blocked for safety reasons.");
+    if (!data.candidates || data.candidates.length === 0) {
+      throw new Error(`No ${inputImage ? 'edited' : 'generated'} image returned`);
+    }
 
+    const candidate = data.candidates[0];
+    if (candidate.finishReason === "SAFETY") {
+      throw new Error(`Image ${inputImage ? 'editing' : 'generation'} blocked due to safety filters`);
+    }
+
+    const responseParts = candidate.content?.parts || [];
     let textResponse = "";
     let imageBase64 = "";
-    (candidate.content?.parts || []).forEach((part: any) => {
-        if (part.text) textResponse += part.text;
-        if (part.inlineData?.data || part.inline_data?.data) {
-            imageBase64 = part.inlineData?.data || part.inline_data.data;
-        }
-    });
 
-    let result = "";
-    if (textResponse) result += textResponse + "\\\\n\\\\n";
-    if (imageBase64) result += `${inputImage ? 'Edited' : 'Generated'} image:\\\\n"data:image/png;base64,${imageBase64}"`;
-    return result || "Image processing complete.";
+    for (const part of responseParts) {
+      if (part.text) textResponse += part.text;
+      if (part.inlineData?.data) imageBase64 = part.inlineData.data;
+      if (part.inline_data?.data) imageBase64 = part.inline_data.data;
+    }
+
+    const result: { text?: string; imageBase64?: string; imageUrl?: string } = {};
+    if (textResponse) result.text = textResponse;
+    if (imageBase64) {
+      result.imageBase64 = imageBase64;
+      result.imageUrl = `data:image/png;base64,${imageBase64}`;
+    }
+    return result;
+  }
+    
+  async generateContentWithGrounding(messages: OpenAIMessage[], modelName: string): Promise<string> {
+    const apiKey = this.getNextApiKey();
+    const fullModelName = modelName.startsWith('models/') ? modelName : `models/${modelName}`;
+    const contents = messages.map(msg => ({ role: msg.role === 'assistant' ? 'model' : 'user', parts: [{ text: typeof msg.content === 'string' ? msg.content : '' }] }));
+
+    const requestBody = {
+      contents,
+      tools: [{ googleSearch: {} }],
+      generationConfig: { temperature: 0.7, maxOutputTokens: 4096 }
+    };
+
+    const response = await fetch(
+      `https://generativelanguage.googleapis.com/v1beta/${fullModelName}:generateContent?key=${apiKey}`,
+      { method: "POST", headers: { "Content-Type": "application/json" }, body: JSON.stringify(requestBody) }
+    );
+
+    if (!response.ok) {
+      console.warn(`Google Search API failed: ${response.status}, trying alternative.`);
+      return await this.generateContentWithSearchPrompt(messages, modelName);
+    }
+
+    const data = await response.json();
+    if (!data.candidates || data.candidates.length === 0) {
+      return await this.generateContentWithSearchPrompt(messages, modelName);
+    }
+
+    const candidate = data.candidates[0];
+    if (candidate.finishReason === "SAFETY") {
+      throw new Error("Response blocked due to safety filters");
+    }
+    return candidate.content?.parts[0]?.text || "No response generated";
+  }
+
+  async generateContentWithSearchPrompt(messages: OpenAIMessage[], modelName: string): Promise<string> {
+    const enhancedMessages = [...messages];
+    const lastMessage = enhancedMessages[enhancedMessages.length - 1];
+    if (typeof lastMessage.content === "string") {
+      lastMessage.content = `Please provide the most current and accurate information available about: ${lastMessage.content}.`;
+    }
+    return await this.generateContent(enhancedMessages, modelName, false);
+  }
+
+  async generateOrEditImage(prompt: string, modelName: string, inputImages?: any[]): Promise<string> {
+    if (this.isImageGenerationModel(modelName)) {
+      try {
+        let inputImage: { mimeType: string; data: string } | undefined;
+        if (inputImages && inputImages.length > 0) {
+          inputImage = this.extractImageData(inputImages[0].url);
+        }
+        const result = await this.generateOrEditImageWithGemini(prompt, modelName, inputImage);
+        let response = "";
+        if (result.text) response += result.text + "\\\\n\\\\n";
+        if (result.imageUrl) response += `${inputImage ? 'Edited' : 'Generated'} image:\\\\n${result.imageUrl}`;
+        return response || `Image processing complete.`;
+      } catch (error) {
+        return `Image processing failed: ${error.message}`;
+      }
+    }
+    return `Model ${modelName} does not support image generation. Use a model like gemini-2.0-flash-preview-image-generation.`;
   }
 }
 
@@ -320,7 +534,8 @@ class OpenAICompatibleServer {
 
   private authenticate(request: Request): boolean {
     if (!this.authKey) return true;
-    return request.headers.get("Authorization")?.replace("Bearer ", "") === this.authKey;
+    const authHeader = request.headers.get("Authorization");
+    return authHeader ? authHeader.replace("Bearer ", "") === this.authKey : false;
   }
 
   private isDocumentContent(url?: string): boolean {
@@ -331,114 +546,178 @@ class OpenAICompatibleServer {
            lowerUrl.includes('.md') || lowerUrl.startsWith('data:text/markdown');
   }
 
+  /**
+   * [新增] 处理 OpenAI 兼容的 TTS 请求
+   */
   private async handleAudioSpeech(request: Request): Promise<Response> {
-    const body: OpenAITTSRequest = await request.json();
-    if (!body.input || !body.voice || !body.model) {
-        return new Response(JSON.stringify({ error: "Missing required fields: input, voice, model." }), { status: 400 });
+    try {
+        const body: OpenAITTSRequest = await request.json();
+        
+        if (!body.input || !body.voice || !body.model) {
+            return new Response(JSON.stringify({ error: { message: "Missing required fields: input, voice, and model.", type: "invalid_request_error" } }), { status: 400 });
+        }
+
+        const audioBuffer = await this.googleAI.generateSpeech(body.input, body.model, body.voice);
+
+        // Google Gemini TTS API 生成的是 MP3 格式的音频。
+        return new Response(audioBuffer, {
+            headers: {
+                "Content-Type": "audio/mp3",
+                "Access-Control-Allow-Origin": "*",
+            }
+        });
+    } catch (error) {
+        console.error("Error in audio speech generation:", error.message);
+        return new Response(
+            JSON.stringify({
+                error: {
+                    message: error.message,
+                    type: "api_error",
+                    code: null
+                }
+            }),
+            { status: 500, headers: { "Content-Type": "application/json" } }
+        );
     }
-    const audioBuffer = await this.googleAI.generateSpeech(body.input, body.model, body.voice);
-    return new Response(audioBuffer, { headers: { "Content-Type": "audio/mp3" } });
   }
   
   private async handleChatCompletions(request: Request): Promise<Response> {
     try {
       const body: OpenAIRequest = await request.json();
       const requestedModel = body.model || "gemini-1.5-pro";
+      const stream = body.stream || false;
+      console.log(`Request for model: ${requestedModel}, stream: ${stream}`);
 
-      // [修改] 流式请求处理
-      if (body.stream) {
-        const googleStream = this.googleAI.streamGenerateContent(body.messages, requestedModel);
-        const openAIStream = this.streamGoogleResponseAsOpenAI(googleStream, requestedModel);
-        return new Response(openAIStream, {
-          headers: { "Content-Type": "text/event-stream", "Cache-Control": "no-cache", "Connection": "keep-alive" }
-        });
-      }
-
-      // [保持不变] 非流式请求处理
       const lastMessage = body.messages[body.messages.length - 1];
-      const content = typeof lastMessage.content === "string" ? lastMessage.content : (Array.isArray(lastMessage.content) ? lastMessage.content.map(p => p.text || "").join(" ") : "");
-      const hasDocument = body.messages.some(msg => Array.isArray(msg.content) && msg.content.some(part => part.type === "document" || this.isDocumentContent(part.document?.url)));
-      const hasImages = body.messages.some(msg => Array.isArray(msg.content) && msg.content.some(part => part.type === "image_url"));
+      const content = typeof lastMessage.content === "string"
+        ? lastMessage.content
+        : (Array.isArray(lastMessage.content) ? lastMessage.content.map(p => p.text || "").join(" ") : "");
+
+      const hasDocument = body.messages.some(msg =>
+        Array.isArray(msg.content) &&
+        msg.content.some(part => part.type === "document" || this.isDocumentContent(part.document?.url))
+      );
 
+      const hasImages = body.messages.some(msg => Array.isArray(msg.content) && msg.content.some(part => part.type === "image_url"));
+      
+      let inputImages: any[] = [];
+      if (hasImages) {
+        body.messages.forEach(msg => {
+          if (Array.isArray(msg.content)) {
+            msg.content.forEach(part => {
+              if (part.type === "image_url" && part.image_url) inputImages.push({ url: part.image_url.url });
+            });
+          }
+        });
+      }
+      
       let responseText: string;
-      if (this.googleAI.isImageGenerationModel(requestedModel)) {
-        const inputImages = hasImages ? body.messages.flatMap(msg => Array.isArray(msg.content) ? msg.content.filter(p => p.type === 'image_url' && p.image_url).map(p => ({ url: p.image_url!.url })) : []) : undefined;
+
+      if (hasDocument) {
+        responseText = await this.googleAI.generateContentWithDocument(body.messages, requestedModel);
+      } else if (this.googleAI.isImageEditingModel(requestedModel) && hasImages) {
         responseText = await this.googleAI.generateOrEditImage(content, requestedModel, inputImages);
-      } else if(hasDocument || hasImages) { // Vision/Document models
-        responseText = await this.googleAI.generateContent(body.messages, requestedModel);
-      } else { // Standard text
-        responseText = await this.googleAI.generateContent(body.messages, requestedModel);
+      } else if (this.googleAI.isImageGenerationModel(requestedModel)) {
+        responseText = await this.googleAI.generateOrEditImage(content, requestedModel);
+      } else if (content.toLowerCase().startsWith("/search:")) {
+        const query = content.substring(8).trim();
+        const searchMessages = [{ ...lastMessage, content: query }];
+        responseText = await this.googleAI.generateContentWithGrounding(searchMessages, requestedModel);
+      } else {
+        responseText = await this.googleAI.generateContent(body.messages, requestedModel, false);
       }
 
-      const responsePayload = {
-        id: `chatcmpl-${Date.now()}`, object: "chat.completion", created: Math.floor(Date.now() / 1000), model: requestedModel,
-        choices: [{ index: 0, message: { role: "assistant", content: responseText }, finish_reason: "stop" }],
-        usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 }
-      };
-      return new Response(JSON.stringify(responsePayload), { headers: { "Content-Type": "application/json" } });
-
+      if (stream) {
+        const streamResponse = await this.streamStringAsOpenAIResponse(responseText, requestedModel);
+        return new Response(streamResponse, {
+          headers: { "Content-Type": "text/event-stream", "Cache-Control": "no-cache", "Connection": "keep-alive", "Access-Control-Allow-Origin": "*" }
+        });
+      } else {
+        const responsePayload = {
+          id: `chatcmpl-${Date.now()}`, object: "chat.completion", created: Math.floor(Date.now() / 1000), model: requestedModel,
+          choices: [{ index: 0, message: { role: "assistant", content: responseText }, finish_reason: "stop" }],
+          usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 }
+        };
+        return new Response(JSON.stringify(responsePayload), { headers: { "Content-Type": "application/json" } });
+      }
     } catch (error) {
       console.error("Error in chat completions:", error.message);
-      const status = error.message.includes("exceeds") || error.message.includes("Invalid") ? 400 : 500;
-      return new Response(JSON.stringify({ error: { message: error.message, type: "api_error" } }),
-        { status, headers: { "Content-Type": "application/json" } });
+      const status = error.message.includes("exceeds the limit") || error.message.includes("Invalid") ? 400 : 500;
+      return new Response(
+        JSON.stringify({
+          error: {
+            message: error.message,
+            type: status === 400 ? "invalid_request_error" : "api_error",
+            code: null
+          }
+        }),
+        { status, headers: { "Content-Type": "application/json" } }
+      );
     }
   }
   
-  /**
-   * [新增] 真正的流式响应转换函数
-   * 将 Google 的数据流转换为 OpenAI 兼容的 Server-Sent Events 流
-   */
-  private streamGoogleResponseAsOpenAI(googleStream: AsyncGenerator<string>, modelName: string): ReadableStream<Uint8Array> {
+  private async streamStringAsOpenAIResponse(content: string, modelName: string): Promise<ReadableStream<Uint8Array>> {
     const encoder = new TextEncoder();
     const streamId = `chatcmpl-${Date.now()}`;
     const creationTime = Math.floor(Date.now() / 1000);
+    let contentQueue = content.split('');
 
     return new ReadableStream({
-      async start(controller) {
-        // 发送第一个包含角色的数据块
-        const initialChunk = { id: streamId, object: 'chat.completion.chunk', created: creationTime, model: modelName, choices: [{ index: 0, delta: { role: 'assistant', content: '' }, finish_reason: null }] };
-        controller.enqueue(encoder.encode(`data: ${JSON.stringify(initialChunk)}\n\n`));
-        
-        // 迭代从 Google API 收到的文本块并转发
-        for await (const textChunk of googleStream) {
-            if (textChunk) {
-                const chunk = {
-                    id: streamId, object: 'chat.completion.chunk', created: creationTime, model: modelName,
-                    choices: [{ index: 0, delta: { content: textChunk }, finish_reason: null }]
-                };
-                controller.enqueue(encoder.encode(`data: ${JSON.stringify(chunk)}\n\n`));
+        start(controller) {
+            const initialChunk = { id: streamId, object: 'chat.completion.chunk', created: creationTime, model: modelName, choices: [{ index: 0, delta: { role: 'assistant', content: '' }, finish_reason: null }] };
+            controller.enqueue(encoder.encode(`data: ${JSON.stringify(initialChunk)}\n\n`));
+        },
+        pull(controller) {
+            if (contentQueue.length === 0) {
+                const finalChunk = { id: streamId, object: 'chat.completion.chunk', created: creationTime, model: modelName, choices: [{ index: 0, delta: {}, finish_reason: 'stop' }] };
+                controller.enqueue(encoder.encode(`data: ${JSON.stringify(finalChunk)}\n\n`));
+                controller.enqueue(encoder.encode('data: [DONE]\n\n'));
+                controller.close();
+                return;
             }
+            const char = contentQueue.shift();
+            const chunk = { id: streamId, object: 'chat.completion.chunk', created: creationTime, model: modelName, choices: [{ index: 0, delta: { content: char }, finish_reason: null }] };
+            controller.enqueue(encoder.encode(`data: ${JSON.stringify(chunk)}\n\n`));
         }
-        
-        // 发送结束信号
-        const finalChunk = { id: streamId, object: 'chat.completion.chunk', created: creationTime, model: modelName, choices: [{ index: 0, delta: {}, finish_reason: 'stop' }] };
-        controller.enqueue(encoder.encode(`data: ${JSON.stringify(finalChunk)}\n\n`));
-        controller.enqueue(encoder.encode('data: [DONE]\n\n'));
-        controller.close();
-      }
     });
   }
   
   private async handleModels(): Promise<Response> {
-    const googleModels = await this.googleAI.fetchOfficialModels();
-    const fallbackModels = this.googleAI['getFallbackModels']();
-    
-    const uniqueModelMap = new Map();
-    [...googleModels, ...fallbackModels].forEach(model => {
-        const modelId = model.id || model.name.replace('models/', '');
-        if (!uniqueModelMap.has(modelId)) {
-            uniqueModelMap.set(modelId, {
-                id: modelId, object: "model", created: Math.floor(Date.now() / 1000), owned_by: "google"
-            });
-        }
-    });
-    return new Response(JSON.stringify({ object: "list", data: Array.from(uniqueModelMap.values()) }), { headers: { "Content-Type": "application/json" } });
+    try {
+      const googleModels = await this.googleAI.fetchOfficialModels();
+      const fallbackModels = this.googleAI['getFallbackModels'](); // Access private method for a complete list
+      
+      const allModels = [...googleModels, ...fallbackModels];
+      const uniqueModelMap = new Map();
+      allModels.forEach(model => {
+          const modelId = model.id || model.name.replace('models/', '');
+          if (!uniqueModelMap.has(modelId)) {
+              uniqueModelMap.set(modelId, {
+                  id: modelId, 
+                  object: "model", 
+                  created: Math.floor(Date.now() / 1000), 
+                  owned_by: "google",
+                  description: model.description || model.displayName, 
+                  maxTokens: model.inputTokenLimit || model.maxTokens
+              });
+          }
+      });
+
+      const models = {
+        object: "list",
+        data: Array.from(uniqueModelMap.values()),
+      };
+      
+      return new Response(JSON.stringify(models), { headers: { "Content-Type": "application/json" } });
+    } catch (error) {
+      console.error("Error fetching models:", error);
+      return new Response(JSON.stringify({ error: { message: "Failed to fetch models." } }), { status: 500 });
+    }
   }
   
   private async handleStatus(): Promise<Response> {
       const status = {
-          status: "healthy", timestamp: new Date().toISOString(), version: "2.5.1",
+          status: "healthy", timestamp: new Date().toISOString(), version: "2.5.0",
           api_keys_loaded: this.googleAI.apiKeys.length,
           models_in_cache: this.googleAI.cachedModels.length,
           models_last_fetched: this.googleAI.modelsLastFetch > 0 ? new Date(this.googleAI.modelsLastFetch).toISOString() : "never"
@@ -460,52 +739,53 @@ class OpenAICompatibleServer {
     const url = new URL(request.url);
     let response: Response;
 
-    try {
-        if (url.pathname === "/health" || url.pathname === "/status") {
-          response = await this.handleStatus();
-        } else if (!this.authenticate(request)) {
-          response = new Response(JSON.stringify({ error: { message: "Unauthorized" } }), { status: 401 });
-        } else if (url.pathname === "/v1/audio/speech" && request.method === "POST") {
-          response = await this.handleAudioSpeech(request);
-        } else if (url.pathname === "/v1/chat/completions" && request.method === "POST") {
-          response = await this.handleChatCompletions(request);
-        } else if (url.pathname === "/v1/models" && request.method === "GET") {
-          response = await this.handleModels();
-        } else {
-          response = new Response("Not Found", { status: 404 });
-        }
-    } catch (error) {
-        console.error("Unhandled error:", error);
-        response = new Response(JSON.stringify({ error: { message: error.message || "An internal server error occurred." } }), { status: 500 });
+    // Handle routes
+    if (url.pathname === "/health" || url.pathname === "/status") {
+      response = await this.handleStatus();
+    } else if (!this.authenticate(request)) {
+      response = new Response(JSON.stringify({ error: { message: "Unauthorized" } }), { status: 401 });
+    // [修改] 添加 TTS 路由
+    } else if (url.pathname === "/v1/audio/speech" && request.method === "POST") {
+      response = await this.handleAudioSpeech(request);
+    } else if (url.pathname === "/v1/chat/completions" && request.method === "POST") {
+      response = await this.handleChatCompletions(request);
+    } else if (url.pathname === "/v1/models" && request.method === "GET") {
+      response = await this.handleModels();
+    } else {
+      response = new Response("Not Found", { status: 404 });
     }
 
-    // 为所有响应添加CORS头
+    // Add CORS headers to all responses
     const finalHeaders = new Headers(response.headers);
     for (const [key, value] of Object.entries(corsHeaders)) {
       finalHeaders.set(key, value);
     }
-    return new Response(response.body, { status: response.status, statusText: response.statusText, headers: finalHeaders });
+
+    return new Response(response.body, { status: response.status, headers: finalHeaders });
   }
 }
 
 // --- 服务器启动 ---
 const server = new OpenAICompatibleServer();
 
-console.log("🚀 OpenAI Compatible Server with Google AI starting...");
-console.log(`✅ Loaded ${server['googleAI'].apiKeys.length} API key(s).`);
+console.log("🚀 OpenAI Compatible Server with Google AI starting on port 8000...");
+console.log(`✅ Loaded ${server.googleAI.apiKeys.length} API key(s).`);
+console.log(`📄 Max document size set to ${MAX_DOCUMENT_SIZE_MB}MB.`);
 
-server['googleAI'].fetchOfficialModels().catch(error => {
-  console.warn(`⚠️ Could not pre-fetch models: ${error.message}. Will use fallbacks.`);
+// Pre-fetch models at startup
+server.googleAI.fetchOfficialModels().then(models => {
+  console.log(`✅ Successfully fetched ${models.length} models from Google AI.`);
+}).catch(error => {
+  console.warn(`⚠️ Could not pre-fetch models: ${error.message}. Will use fallbacks or fetch on first request.`);
 });
 
-const port = 7860;
-console.log(`Server listening on http://localhost:${port}`);
 console.log("\n🔗 Endpoints:");
-console.log(`   POST /v1/chat/completions`);
-console.log(`   POST /v1/audio/speech`);
-console.log(`   GET  /v1/models`);
-console.log(`   GET  /status`);
+console.log("   POST /v1/chat/completions");
+console.log("   POST /v1/audio/speech      <-- [NEW] OpenAI TTS compatible endpoint"); // [修改] 更新启动日志
+console.log("   GET  /v1/models");
+console.log("   GET  /status");
 
+// [修改] 端口从 7860 改为 8000，与日志一致。您可以根据需要改回 7860。
 await serve(
   (request: Request) => server.handleRequest(request),
   { port: 7860 }