Update main.ts

main.ts

@@ -1,10 +1,10 @@
 import { serve } from "https://deno.land/std@0.208.0/http/server.ts";
-import { decode } from "https://deno.land/std@0.208.0/encoding/base64.ts";
+import wav from 'wav'
 
 // --- 常量定义 ---
-const MAX_DOCUMENT_SIZE_MB = 20;
+const MAX_DOCUMENT_SIZE_MB = 20; // 设置最大文档大小限制（单位：MB）
 const MAX_DOCUMENT_SIZE_BYTES = MAX_DOCUMENT_SIZE_MB * 1024 * 1024;
-const MODELS_CACHE_DURATION = 60000;
+const MODELS_CACHE_DURATION = 60000; // 1分钟模型缓存
 
 // --- 接口定义 ---
 interface OpenAIMessage {
@@ -12,8 +12,8 @@ interface OpenAIMessage {
   content: string | Array<{
     type: string;
     text?: string;
-    image_url?: { url:string };
-    document?: { url: string; type: string };
+    image_url?: { url: string };
+    document?: { url: string; type: string }; // 支持多种文档类型
   }>;
 }
 
@@ -25,14 +25,16 @@ interface OpenAIRequest {
   stream?: boolean;
 }
 
+// [新增] OpenAI TTS 请求接口定义
 interface OpenAITTSRequest {
-  model: 'tts-1' | 'tts-1-hd';
-  input: string;
-  voice: string;
-  response_format?: 'mp3' | 'opus' | 'aac' | 'flac';
-  speed?: number;
+    model: string; // e.g., 'tts-1', 'tts-1-hd'
+    input: string; // The text to synthesize
+    voice: 'Puck' | 'Charon' | 'Kore' | 'Fenrir' | 'Leda' | 'Aoede';
+    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
 }
 
+
 class GoogleAIService {
   public apiKeys: string[];
   public currentKeyIndex = 0;
@@ -49,6 +51,7 @@ class GoogleAIService {
       this.apiKeys.push(key);
       i++;
     }
+
     if (this.apiKeys.length === 0) {
       throw new Error("No Google AI API keys found in environment variables (e.g., GOOGLE_AI_KEY_1, GOOGLE_AI_KEY)");
     }
@@ -60,35 +63,112 @@ class GoogleAIService {
     return key;
   }
 
-  async synthesizeSpeech(input: string, voiceName: string): Promise<Uint8Array> {
+  async function saveWaveFile(
+    filename,
+    pcmData,
+    channels = 1,
+    rate = 24000,
+    sampleWidth = 2,
+    ) {
+    return new Promise((resolve, reject) => {
+        const writer = new wav.FileWriter(filename, {
+                channels,
+                sampleRate: rate,
+                bitDepth: sampleWidth * 8,
+        });
+
+        writer.on('finish', resolve);
+        writer.on('error', reject);
+
+        writer.write(pcmData);
+        writer.end();
+    });
+    }
+  
+  // --- [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 } = {
+        'Puck': 'Puck',      // A good default, versatile voice'Puck' | 'Charon' | 'Kore' | 'Fenrir' | 'Leda' | 'Aoede'
+        'Charon': 'Charon',        // Another male voice option
+        'Kore': 'Kore',     // Female, narrative style
+        'Fenrir': 'Fenrir',       // Deep, male voice
+        'Leda': 'Leda',    // Energetic female voice
+        'Aoede': 'Aoede',    // Gentle female voice
+        // Fallback to a default if the voice is not in the map
+        'default': 'Puck'
+    };
+    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();
-    console.log(`Synthesizing speech with voice: ${voiceName}`);
+    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 = {
-      "input": { "text": input },
-      "voice": { "name": voiceName },
-      "audioConfig": { "audioEncoding": "MP3" }
+        "contents": [{
+            "parts":[{
+                "text": input
+            }]
+        }],
+        "generationConfig": {
+            "responseModalities": ["AUDIO"],
+            "speechConfig": {
+                "voiceConfig": {
+                    "prebuiltVoiceConfig": {
+                        "voiceName": googleVoice
+                    }
+                }
+            }
+        },
+        "model": ttsModel,
     };
+
     const response = await fetch(
-      `https://texttospeech.googleapis.com/v1beta/text:synthesize?key=${apiKey}`,
-      {
-        method: "POST",
-        headers: { "Content-Type": "application/json" },
-        body: JSON.stringify(requestBody),
-      }
+        `https://generativelanguage.googleapis.com/v1beta/models/${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);
-      throw new Error(`Google TTS API request failed: ${response.status}: ${errorMessage}`);
+        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();
-    if (!data.audioContent) {
-      throw new Error("TTS synthesis failed, no audio content in response.");
-    }
-    return decode(data.audioContent);
+    
+    // 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. The response might be blocked or empty.");
+    }   
+    const audioBuffer = Buffer.from(audioContentBase64, 'base64');
+    const fileName = 'out.wav';
+    return await saveWaveFile(fileName, audioBuffer);
   }
   
-  // 省略其他 GoogleAIService 方法，它们与之前相同...
+  // --- 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) {
@@ -127,7 +207,9 @@ class GoogleAIService {
     return [
       { 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"] }
+      { 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" }
     ];
   }
 
@@ -148,19 +230,37 @@ class GoogleAIService {
 
   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 provided as a standard base64 data URL (e.g., 'data:application/pdf;base64,...').");
+
+    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.");
+    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 exceeds the ${MAX_DOCUMENT_SIZE_MB}MB limit.`);
+    if (approxSizeInBytes > MAX_DOCUMENT_SIZE_BYTES) {
+        throw new Error(`Document size (${(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') {
       try {
         const textContent = atob(base64Data);
         return { mimeType, data: base64Data, text: textContent, docType };
-      } catch (error) { throw new Error(`Invalid base64 encoding for ${docType} document.`); }
+      } catch (error) {
+        console.error(`Failed to decode base64 content for ${docType}:`, error);
+        throw new Error(`Invalid base64 encoding for ${docType} document.`);
+      }
     }
+    
     const finalMimeType = docType === 'pdf' ? 'application/pdf' : mimeType;
     return { mimeType: finalMimeType, data: base64Data, docType };
   }
@@ -178,13 +278,265 @@ class GoogleAIService {
   }
 
   async generateContentWithDocument(messages: OpenAIMessage[], modelName: string): Promise<string> {
-    // ... code from previous answer ...
-    return "Not implemented for brevity";
+    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 }] };
+        }
+
+        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 (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 { role: msg.role === "assistant" ? "model" : "user", parts: messageParts.filter(p => p.text || p.inlineData) };
+      });
+    } catch (error) {
+      throw error;
+    }
+
+    const requestBody = {
+      contents,
+      generationConfig: { temperature: 0.7, maxOutputTokens: 8192 }
+    };
+
+    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 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}: ${errorMessage}`);
+    }
+
+    const data = await response.json();
+    const promptFeedback = data.promptFeedback;
+    if (promptFeedback && promptFeedback.blockReason) {
+      const reason = promptFeedback.blockReason;
+      const safetyRatings = promptFeedback.safetyRatings?.map((r: any) => `${r.category}: ${r.probability}`).join(', ') || 'N/A';
+      throw new Error(`Request blocked by Google API. Reason: ${reason}. Safety Ratings: [${safetyRatings}]`);
+    }
+
+    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> {
-    // ... code from previous answer ...
-     return "Not implemented for brevity";
+    const hasDocument = messages.some(msg => Array.isArray(msg.content) && msg.content.some(part => part.type === "document"));
+    if (hasDocument) {
+      return await this.generateContentWithDocument(messages, modelName);
+    }
+
+    const apiKey = this.getNextApiKey();
+    const fullModelName = modelName.startsWith('models/') ? modelName : `models/${modelName}`;
+
+    const contents = messages.map(msg => {
+      if (typeof msg.content === "string") {
+        return { role: msg.role === "assistant" ? "model" : "user", parts: [{ text: msg.content }] };
+      } else {
+        const messageParts = msg.content.map(part => {
+          if (part.type === "text") {
+            return { text: part.text };
+          } else if (part.type === "image_url" && part.image_url) {
+            const imageData = part.image_url.url;
+            if (imageData.startsWith("data:image/")) {
+              const { mimeType, data } = this.extractImageData(imageData);
+              return { inlineData: { mimeType, data } };
+            } else {
+              return { fileData: { mimeType: "image/jpeg", fileUri: imageData } };
+            }
+          }
+          return { text: "" };
+        });
+        return { role: msg.role === "assistant" ? "model" : "user", parts: messageParts };
+      }
+    });
+
+    const requestBody: any = {
+      contents,
+      generationConfig: { temperature: 0.7, maxOutputTokens: 4096 }
+    };
+    if (enableSearch) {
+      requestBody.tools = [{ googleSearchRetrieval: {} }];
+    }
+
+    const response = await fetch(
+      `https://generativelanguage.googleapis.com/v1beta/${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} - ${errorText}`);
+    }
+    const data = await response.json();
+    if (!data.candidates || data.candidates.length === 0) {
+      throw new Error("No response generated from Google AI");
+    }
+    const candidate = data.candidates[0];
+    if (candidate.finishReason === "SAFETY") {
+      throw new Error("Response blocked due to safety filters");
+    }
+    return candidate.content?.parts[0]?.text || "No response generated";
+  }
+
+  async generateOrEditImageWithGemini(prompt: string, modelName: string = "gemini-2.0-flash-preview-image-generation", inputImage?: { mimeType: string; data: string }): Promise<{ text?: string; imageBase64?: string; imageUrl?: string }> {
+    const apiKey = this.getNextApiKey();
+    const fullModelName = modelName.startsWith('models/') ? modelName : `models/${modelName}`;
+    const requestParts: any[] = [{ text: prompt }];
+
+    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) {
+      const errorText = await response.text();
+      throw new Error(`Image ${inputImage ? 'editing' : 'generation'} failed: ${response.status} - ${errorText}`);
+    }
+    const data = await response.json();
+    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 ${inputImage ? 'editing' : 'generation'} blocked due to safety filters`);
+    }
+
+    const responseParts = candidate.content?.parts || [];
+    let textResponse = "";
+    let imageBase64 = "";
+
+    for (const part of responseParts) {
+      if (part.text) textResponse += part.text;
+      if (part.inlineData?.data) imageBase64 = part.inlineData.data;
+      if (part.inline_data?.data) imageBase64 = part.inline_data.data;
+    }
+
+    const result: { text?: string; imageBase64?: string; imageUrl?: string } = {};
+    if (textResponse) result.text = textResponse;
+    if (imageBase64) {
+      result.imageBase64 = imageBase64;
+      result.imageUrl = `data:image/png;base64,${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.`;
   }
 }
 
@@ -211,59 +563,184 @@ 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> {
     try {
-      // 增加一个 Content-Type 检查，这是良好的实践
-      if (request.headers.get("Content-Type") !== "application/json") {
-          return new Response(JSON.stringify({ error: { message: "Content-Type must be application/json" } }), { status: 415, headers: { "Content-Type": "application/json" } });
-      }
-
-      const body: OpenAITTSRequest = await request.json();
-
-      if (!body.input || !body.voice) {
-        throw new Error("Missing required parameters: 'input' and 'voice' are required.");
-      }
-
-      const audioData = await this.googleAI.synthesizeSpeech(body.input, body.voice);
-
-      return new Response(audioData, {
-        status: 200,
-        headers: { "Content-Type": "audio/mpeg" },
-      });
+        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 audioFile = 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/wav",
+                "Access-Control-Allow-Origin": "*",
+            }
+        });
     } catch (error) {
-      console.error("Error in /v1/audio/speech:", error.message);
-      const status = error.message.includes("required parameter") ? 400 : 500;
-      return new Response(JSON.stringify({ error: { message: error.message, type: "api_error" } }), { status, headers: { "Content-Type": "application/json" } });
+        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" } }
+        );
     }
   }
-
+  
   private async handleChatCompletions(request: Request): Promise<Response> {
-    // 省略此处的实现细节，与之前版本相同
     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}`);
+
+      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"));
+      
+      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 (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 (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);
+      }
+
+      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: "gemini-pro",
-          choices: [{ index: 0, message: { role: "assistant", content: "Chat completions logic is correct." }, finish_reason: "stop" }],
+          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) {
-        return new Response(JSON.stringify({ error: { message: error.message } }), { status: 500 });
+      }
+    } catch (error) {
+      console.error("Error in chat completions:", error.message);
+      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" } }
+      );
     }
   }
   
+  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({
+        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`));
+        }
+    });
+  }
+  
   private async handleModels(): Promise<Response> {
-    // 省略此处的实现细节，与之前版本相同
     try {
-      const models = { object: "list", data: [
-          { id: "tts-1", object: "model", owned_by: "google" },
-          { id: "tts-1-hd", object: "model", owned_by: "google" },
-          { id: "gemini-1.5-pro", object: "model", owned_by: "google" }
-      ]};
+      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) {
-        return new Response(JSON.stringify({ error: { message: "Failed to fetch models." } }), { status: 500 });
+      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: "2.5.0",
+          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"
+      };
+      return new Response(JSON.stringify(status), { headers: { "Content-Type": "application/json" } });
+  }
 
   async handleRequest(request: Request): Promise<Response> {
     const corsHeaders = {
@@ -279,38 +756,53 @@ class OpenAICompatibleServer {
     const url = new URL(request.url);
     let response: Response;
 
-    if (!this.authenticate(request)) {
+    // 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 if (url.pathname === "/v1/audio/speech" && request.method === "POST") {
-      response = await this.handleAudioSpeech(request);
     } else {
       response = new Response("Not Found", { status: 404 });
     }
 
-    // --- [ 这是关键的修正 ] ---
-    // 直接修改返回的 Response 对象的 headers，而不是创建一个新的 Response。
+    // Add CORS headers to all responses
+    const finalHeaders = new Headers(response.headers);
     for (const [key, value] of Object.entries(corsHeaders)) {
-      response.headers.set(key, value);
+      finalHeaders.set(key, value);
     }
 
-    return response; // 返回被修改过的原始 response 对象
+    return new Response(response.body, { status: response.status, headers: finalHeaders });
   }
 }
 
 // --- 服务器启动 ---
 const server = new OpenAICompatibleServer();
 
-console.log("🚀 OpenAI Compatible Server starting on port 7860...");
+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.`);
+
+// 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.`);
+});
+
 console.log("\n🔗 Endpoints:");
 console.log("   POST /v1/chat/completions");
-console.log("   POST /v1/audio/speech");
+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 }