diff --git a/src/app/api/chat/route.ts b/src/app/api/chat/route.ts
index cbb338b..0913b25 100644
--- a/src/app/api/chat/route.ts
+++ b/src/app/api/chat/route.ts
@@ -72,6 +72,11 @@ function isCodexModel(modelId: string): boolean {
   return modelId.startsWith('gpt-') && modelId.includes('codex');
 }
 
+// 规范化 URL（移除末尾斜杠，避免拼接时出现双斜杠）
+function normalizeBaseUrl(url: string): string {
+  return url.replace(/\/+$/, '');
+}
+
 // 默认系统提示词 - 用于生成更详细、更有结构的回复
 const DEFAULT_SYSTEM_PROMPT = `你是一个专业、友好的 AI 助手。请遵循以下规则来回复用户：
 
@@ -282,6 +287,7 @@ export async function POST(request: Request) {
       async start(controller) {
         try {
           const cchUrl = settings.cchUrl || process.env.CCH_DEFAULT_URL || 'https://claude.leocoder.cn/';
+          const apiFormat = (settings.apiFormat as 'claude' | 'openai') || 'claude';
 
           // 获取系统提示词（叠加模式）
           // 1. 始终使用 DEFAULT_SYSTEM_PROMPT 作为基础
@@ -312,8 +318,41 @@ export async function POST(request: Request) {
           let totalInputTokens = 0;
           let totalOutputTokens = 0;
 
-          if (isCodex) {
-            // ==================== Codex 模型处理（OpenAI 格式） ====================
+          // 【重要】处理器选择优先级说明：
+          // 1. 首先检查 apiFormat === 'openai'：如果用户选择了 "OpenAI 兼容" 格式，
+          //    则所有模型（包括 Codex 模型）都统一使用 /v1/chat/completions 端点
+          //    这是因为第三方中转站通常只支持 OpenAI 兼容的 /v1/chat/completions 端点
+          // 2. 然后检查 isCodex：如果是 Claude 原生格式 + Codex 模型，才使用 /v1/responses 端点
+          // 3. 最后是普通的 Claude 原生格式，使用 /v1/messages 端点
+
+          if (apiFormat === 'openai') {
+            // ==================== OpenAI 兼容格式处理 ====================
+            // 当用户选择 "OpenAI 兼容" 时，无论什么模型都走这个分支
+            // 第三方中转站统一使用 /v1/chat/completions 端点
+            console.log('[API/chat] 使用 OpenAI 兼容格式，模型:', useModel, '(isCodex:', isCodex, ')');
+            const result = await handleOpenAICompatibleChat({
+              cchUrl,
+              apiKey: decryptedApiKey,
+              model: useModel,
+              systemPrompt,
+              temperature,
+              historyMessages,
+              message,
+              tools: tools || (conversation.tools as string[]) || [],
+              controller,
+              encoder,
+              images,
+            });
+
+            fullContent = result.fullContent;
+            thinkingContent = result.thinkingContent;
+            totalInputTokens = result.inputTokens;
+            totalOutputTokens = result.outputTokens;
+          } else if (isCodex) {
+            // ==================== Codex 模型处理（使用 Codex Response API） ====================
+            // 仅当使用 Claude 原生格式 + Codex 模型时，才使用 /v1/responses 端点
+            // 这是 CCH 项目特有的 Codex Response API
+            console.log('[API/chat] 使用 Codex Response API (Claude 原生格式 + Codex 模型)');
             const result = await handleCodexChat({
               cchUrl,
               apiKey: decryptedApiKey,
@@ -332,7 +371,8 @@ export async function POST(request: Request) {
             totalInputTokens = result.inputTokens;
             totalOutputTokens = result.outputTokens;
           } else {
-            // ==================== Claude 模型处理（原有逻辑） ====================
+            // ==================== Claude 原生格式处理 ====================
+            console.log('[API/chat] 使用 Claude 原生格式 (/v1/messages)');
             const result = await handleClaudeChat({
               cchUrl,
               apiKey: decryptedApiKey,
@@ -476,13 +516,15 @@ async function handleCodexChat(params: CodexChatParams): Promise<{
     images,
   } = params;
 
-  // 构建 Codex Response API 格式的输入
+  // 构建 Codex Response API 格式的输入（过滤空内容的消息）
   const inputItems: CodexInputItem[] = [
-    ...historyMessages.map((msg) => ({
-      type: 'message' as const,
-      role: msg.role as 'user' | 'assistant',
-      content: msg.content,
-    })),
+    ...historyMessages
+      .filter((msg) => msg.content && msg.content.trim() !== '')
+      .map((msg) => ({
+        type: 'message' as const,
+        role: msg.role as 'user' | 'assistant',
+        content: msg.content,
+      })),
   ];
 
   // 添加当前用户消息（支持多模态内容）
@@ -556,7 +598,7 @@ async function handleCodexChat(params: CodexChatParams): Promise<{
     }
 
     // 使用 Codex Response API 端点
-    const response = await fetch(`${cchUrl}/v1/responses`, {
+    const response = await fetch(`${normalizeBaseUrl(cchUrl)}/v1/responses`, {
       method: 'POST',
       headers: {
         'Content-Type': 'application/json',
@@ -796,11 +838,13 @@ async function handleClaudeChat(params: ClaudeChatParams): Promise<{
     images,
   } = params;
 
-  // 构建消息历史
-  const messageHistory: APIMessage[] = historyMessages.map((msg) => ({
-    role: msg.role as 'user' | 'assistant',
-    content: msg.content,
-  }));
+  // 构建消息历史（过滤空内容的消息）
+  const messageHistory: APIMessage[] = historyMessages
+    .filter((msg) => msg.content && msg.content.trim() !== '')
+    .map((msg) => ({
+      role: msg.role as 'user' | 'assistant',
+      content: msg.content,
+    }));
 
   // 添加当前用户消息（支持多模态内容）
   if (images && images.length > 0) {
@@ -900,7 +944,7 @@ async function handleClaudeChat(params: ClaudeChatParams): Promise<{
       requestBody.temperature = temperature;
     }
 
-    const response = await fetch(`${cchUrl}/v1/messages`, {
+    const response = await fetch(`${normalizeBaseUrl(cchUrl)}/v1/messages`, {
       method: 'POST',
       headers: {
         'Content-Type': 'application/json',
@@ -1126,6 +1170,462 @@ async function handleClaudeChat(params: ClaudeChatParams): Promise<{
   };
 }
 
+// ==================== OpenAI 兼容格式处理函数 ====================
+interface OpenAICompatibleChatParams {
+  cchUrl: string;
+  apiKey: string;
+  model: string;
+  systemPrompt: string;
+  temperature: number;
+  historyMessages: { role: string; content: string }[];
+  message: string;
+  tools: string[];
+  controller: ReadableStreamDefaultController;
+  encoder: TextEncoder;
+  images?: {
+    type: 'image';
+    media_type: string;
+    data: string;
+  }[];
+}
+
+// OpenAI 消息格式
+interface OpenAIMessageContent {
+  type: 'text' | 'image_url';
+  text?: string;
+  image_url?: {
+    url: string;
+  };
+}
+
+interface OpenAICompatibleMessage {
+  role: 'system' | 'user' | 'assistant' | 'tool';
+  content: string | OpenAIMessageContent[] | null;
+  tool_calls?: {
+    id: string;
+    type: 'function';
+    function: {
+      name: string;
+      arguments: string;
+    };
+  }[];
+  tool_call_id?: string;
+}
+
+async function handleOpenAICompatibleChat(params: OpenAICompatibleChatParams): Promise<{
+  fullContent: string;
+  thinkingContent: string;
+  inputTokens: number;
+  outputTokens: number;
+}> {
+  const {
+    cchUrl,
+    apiKey,
+    model,
+    systemPrompt,
+    temperature,
+    historyMessages,
+    message,
+    tools,
+    controller,
+    encoder,
+    images,
+  } = params;
+
+  // 构建 OpenAI 格式的消息历史（过滤空内容的消息）
+  const openaiMessages: OpenAICompatibleMessage[] = [
+    { role: 'system', content: systemPrompt },
+    ...historyMessages
+      .filter((msg) => msg.content && msg.content.trim() !== '')
+      .map((msg) => ({
+        role: msg.role as 'user' | 'assistant',
+        content: msg.content,
+      })),
+  ];
+
+  // 添加当前用户消息（支持多模态）
+  if (images && images.length > 0) {
+    console.log('[handleOpenAICompatibleChat] Building multimodal message with', images.length, 'images');
+    const multimodalContent: OpenAIMessageContent[] = [];
+
+    // 添加图片
+    for (const img of images) {
+      console.log('[handleOpenAICompatibleChat] Adding image:', {
+        type: img.type,
+        media_type: img.media_type,
+        dataLength: img.data?.length || 0,
+      });
+      multimodalContent.push({
+        type: 'image_url',
+        image_url: {
+          url: `data:${img.media_type};base64,${img.data}`,
+        },
+      });
+    }
+
+    // 添加文本
+    if (message) {
+      multimodalContent.push({
+        type: 'text',
+        text: message,
+      });
+    }
+
+    openaiMessages.push({
+      role: 'user',
+      content: multimodalContent,
+    });
+  } else {
+    console.log('[handleOpenAICompatibleChat] No images, using simple text message');
+    openaiMessages.push({
+      role: 'user',
+      content: message,
+    });
+  }
+
+  // 构建 OpenAI 格式的工具定义
+  const openaiTools = buildOpenAIToolDefinitions(tools);
+
+  let fullContent = '';
+  let thinkingContent = ''; // 用于收集 <think> 标签中的思考内容
+  let totalInputTokens = 0;
+  let totalOutputTokens = 0;
+  let loopCount = 0;
+  const maxLoops = 10;
+
+  // 用于处理 <think> 标签的状态变量（跨 chunk 处理）
+  let isInThinkingMode = false;
+  let pendingBuffer = ''; // 用于处理标签可能跨 chunk 的情况
+
+  while (loopCount < maxLoops) {
+    loopCount++;
+
+    // 构建请求体
+    const requestBody: Record<string, unknown> = {
+      model,
+      messages: openaiMessages,
+      stream: true,
+      temperature,
+    };
+
+    if (openaiTools.length > 0) {
+      requestBody.tools = openaiTools;
+    }
+
+    console.log('[handleOpenAICompatibleChat] Sending to OpenAI-compatible API:', {
+      model,
+      messagesCount: openaiMessages.length,
+      url: `${normalizeBaseUrl(cchUrl)}/v1/chat/completions`,
+    });
+
+    const response = await fetch(`${normalizeBaseUrl(cchUrl)}/v1/chat/completions`, {
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json',
+        'Authorization': `Bearer ${apiKey}`,
+      },
+      body: JSON.stringify(requestBody),
+    });
+
+    if (!response.ok) {
+      const errorText = await response.text();
+      throw new Error(`OpenAI-compatible API error: ${response.status} - ${errorText}`);
+    }
+
+    const reader = response.body?.getReader();
+    if (!reader) {
+      throw new Error('No response body');
+    }
+
+    // 收集当前轮次的内容
+    let currentTextContent = '';
+    const toolCalls: { id: string; name: string; arguments: string }[] = [];
+    const currentToolCallsMap: Map<number, { id: string; name: string; arguments: string }> = new Map();
+    let stopReason: string | null = null;
+
+    const decoder = new TextDecoder();
+    let buffer = '';
+
+    // 处理流式响应
+    while (true) {
+      const { done, value } = await reader.read();
+      if (done) break;
+
+      buffer += decoder.decode(value, { stream: true });
+      const lines = buffer.split('\n');
+      buffer = lines.pop() || '';
+
+      for (const line of lines) {
+        if (line.startsWith('data: ')) {
+          const data = line.slice(6).trim();
+          if (data === '[DONE]') continue;
+          if (!data) continue;
+
+          try {
+            const event = JSON.parse(data);
+
+            // 处理文本内容（包含 <think> 标签过滤）
+            if (event.choices?.[0]?.delta?.content) {
+              const rawContent = event.choices[0].delta.content;
+
+              // 将新内容追加到待处理缓冲区
+              pendingBuffer += rawContent;
+
+              // 处理缓冲区中的内容
+              while (pendingBuffer.length > 0) {
+                if (isInThinkingMode) {
+                  // 在 thinking 模式中，查找 </think> 结束标签
+                  const endTagIndex = pendingBuffer.indexOf('</think>');
+
+                  if (endTagIndex !== -1) {
+                    // 找到结束标签，提取 thinking 内容
+                    const thinkPart = pendingBuffer.slice(0, endTagIndex);
+                    if (thinkPart) {
+                      thinkingContent += thinkPart;
+                      // 发送 thinking 事件（让前端可以展示折叠的思考内容）
+                      controller.enqueue(encoder.encode(`data: ${JSON.stringify({
+                        type: 'thinking',
+                        content: thinkPart,
+                      })}\n\n`));
+                    }
+                    // 移除已处理的内容和结束标签
+                    pendingBuffer = pendingBuffer.slice(endTagIndex + 8); // 8 = '</think>'.length
+                    isInThinkingMode = false;
+                  } else {
+                    // 没找到结束标签，检查是否可能是部分标签
+                    // 保留最后 7 个字符以防 </think> 被截断（7 = '</think'.length - 1）
+                    if (pendingBuffer.length > 7) {
+                      const safePart = pendingBuffer.slice(0, -7);
+                      const keepPart = pendingBuffer.slice(-7);
+
+                      if (safePart) {
+                        thinkingContent += safePart;
+                        controller.enqueue(encoder.encode(`data: ${JSON.stringify({
+                          type: 'thinking',
+                          content: safePart,
+                        })}\n\n`));
+                      }
+                      pendingBuffer = keepPart;
+                    }
+                    // 等待更多数据
+                    break;
+                  }
+                } else {
+                  // 不在 thinking 模式，查找 <think> 开始标签
+                  const startTagIndex = pendingBuffer.indexOf('<think>');
+
+                  if (startTagIndex !== -1) {
+                    // 找到开始标签，先输出标签前的普通文本
+                    const textPart = pendingBuffer.slice(0, startTagIndex);
+                    if (textPart) {
+                      currentTextContent += textPart;
+                      fullContent += textPart;
+                      controller.enqueue(encoder.encode(`data: ${JSON.stringify({
+                        type: 'text',
+                        content: textPart,
+                      })}\n\n`));
+                    }
+                    // 移除已处理的内容和开始标签
+                    pendingBuffer = pendingBuffer.slice(startTagIndex + 7); // 7 = '<think>'.length
+                    isInThinkingMode = true;
+                  } else {
+                    // 没找到开始标签，检查是否可能是部分标签
+                    // 保留最后 6 个字符以防 <think> 被截断（6 = '<think'.length - 1）
+                    const potentialTagStart = pendingBuffer.lastIndexOf('<');
+
+                    if (potentialTagStart !== -1 && potentialTagStart > pendingBuffer.length - 7) {
+                      // 可能是部分的 <think> 标签，保留这部分
+                      const safePart = pendingBuffer.slice(0, potentialTagStart);
+                      const keepPart = pendingBuffer.slice(potentialTagStart);
+
+                      if (safePart) {
+                        currentTextContent += safePart;
+                        fullContent += safePart;
+                        controller.enqueue(encoder.encode(`data: ${JSON.stringify({
+                          type: 'text',
+                          content: safePart,
+                        })}\n\n`));
+                      }
+                      pendingBuffer = keepPart;
+                      // 等待更多数据
+                      break;
+                    } else {
+                      // 安全输出所有内容
+                      currentTextContent += pendingBuffer;
+                      fullContent += pendingBuffer;
+                      controller.enqueue(encoder.encode(`data: ${JSON.stringify({
+                        type: 'text',
+                        content: pendingBuffer,
+                      })}\n\n`));
+                      pendingBuffer = '';
+                    }
+                  }
+                }
+              }
+            }
+
+            // 处理工具调用
+            if (event.choices?.[0]?.delta?.tool_calls) {
+              for (const toolCall of event.choices[0].delta.tool_calls) {
+                const index = toolCall.index ?? 0;
+
+                if (!currentToolCallsMap.has(index)) {
+                  // 新的工具调用开始
+                  currentToolCallsMap.set(index, {
+                    id: toolCall.id || '',
+                    name: toolCall.function?.name || '',
+                    arguments: '',
+                  });
+
+                  if (toolCall.id) {
+                    controller.enqueue(encoder.encode(`data: ${JSON.stringify({
+                      type: 'tool_use_start',
+                      id: toolCall.id,
+                      name: toolCall.function?.name || '',
+                    })}\n\n`));
+                  }
+                }
+
+                const current = currentToolCallsMap.get(index)!;
+
+                // 更新工具调用信息
+                if (toolCall.id) current.id = toolCall.id;
+                if (toolCall.function?.name) current.name = toolCall.function.name;
+                if (toolCall.function?.arguments) {
+                  current.arguments += toolCall.function.arguments;
+                }
+              }
+            }
+
+            // 处理结束原因
+            if (event.choices?.[0]?.finish_reason) {
+              stopReason = event.choices[0].finish_reason;
+            }
+
+            // 处理 usage 信息
+            if (event.usage) {
+              totalInputTokens = event.usage.prompt_tokens || 0;
+              totalOutputTokens = event.usage.completion_tokens || 0;
+            }
+          } catch (e) {
+            console.error('[handleOpenAICompatibleChat] Parse error:', e, 'Line:', line);
+          }
+        }
+      }
+    }
+
+    // 流结束后，处理可能残留在 pendingBuffer 中的内容
+    if (pendingBuffer.length > 0) {
+      if (isInThinkingMode) {
+        // 如果还在 thinking 模式，说明 </think> 标签没有正常闭合
+        // 将剩余内容作为 thinking 内容处理
+        thinkingContent += pendingBuffer;
+        controller.enqueue(encoder.encode(`data: ${JSON.stringify({
+          type: 'thinking',
+          content: pendingBuffer,
+        })}\n\n`));
+      } else {
+        // 普通文本模式，输出剩余内容
+        currentTextContent += pendingBuffer;
+        fullContent += pendingBuffer;
+        controller.enqueue(encoder.encode(`data: ${JSON.stringify({
+          type: 'text',
+          content: pendingBuffer,
+        })}\n\n`));
+      }
+      pendingBuffer = '';
+    }
+
+    // 收集所有工具调用
+    for (const [, tc] of currentToolCallsMap) {
+      if (tc.id && tc.name) {
+        toolCalls.push(tc);
+        controller.enqueue(encoder.encode(`data: ${JSON.stringify({
+          type: 'tool_use_complete',
+          id: tc.id,
+          name: tc.name,
+          input: JSON.parse(tc.arguments || '{}'),
+        })}\n\n`));
+      }
+    }
+
+    // 检查是否需要执行工具
+    if ((stopReason === 'tool_calls' || toolCalls.length > 0) && toolCalls.length > 0) {
+      // 将助手消息添加到历史
+      const assistantMessage: OpenAICompatibleMessage = {
+        role: 'assistant',
+        content: currentTextContent || null,
+        tool_calls: toolCalls.map((tc) => ({
+          id: tc.id,
+          type: 'function' as const,
+          function: {
+            name: tc.name,
+            arguments: tc.arguments,
+          },
+        })),
+      };
+      openaiMessages.push(assistantMessage);
+
+      // 执行所有工具并收集结果
+      for (const tc of toolCalls) {
+        controller.enqueue(encoder.encode(`data: ${JSON.stringify({
+          type: 'tool_execution_start',
+          id: tc.id,
+          name: tc.name,
+        })}\n\n`));
+
+        // 解析工具参数
+        let toolInput: Record<string, unknown> = {};
+        try {
+          toolInput = JSON.parse(tc.arguments || '{}');
+        } catch {
+          console.error('[handleOpenAICompatibleChat] Failed to parse tool arguments');
+        }
+
+        // 执行工具
+        const result = await executeTool(tc.name, toolInput);
+
+        controller.enqueue(encoder.encode(`data: ${JSON.stringify({
+          type: 'tool_execution_result',
+          id: tc.id,
+          name: tc.name,
+          success: result.success,
+          result: result.displayResult,
+          images: result.images,
+        })}\n\n`));
+
+        // 将工具结果显示给用户
+        const toolDisplayText = `\n\n${result.displayResult}\n\n`;
+        fullContent += toolDisplayText;
+        controller.enqueue(encoder.encode(`data: ${JSON.stringify({
+          type: 'text',
+          content: toolDisplayText,
+        })}\n\n`));
+
+        // 将工具结果添加到消息历史（OpenAI 格式）
+        openaiMessages.push({
+          role: 'tool',
+          content: result.fullResult,
+          tool_call_id: tc.id,
+        });
+      }
+
+      // 继续循环，让 AI 基于工具结果继续回复
+      continue;
+    }
+
+    // 如果没有工具调用，结束循环
+    break;
+  }
+
+  return {
+    fullContent,
+    thinkingContent,
+    inputTokens: totalInputTokens,
+    outputTokens: totalOutputTokens,
+  };
+}
+
 // 构建 Claude 工具定义
 function buildClaudeToolDefinitions(toolIds: string[]) {
   const toolMap: Record<string, object> = {