feat(API): 聊天接口支持文档解析和多模态输入

- 支持PDF文档使用Claude原生document类型
- 集成Office文档解析(Word/Excel)
- 支持图片+文档+文本混合输入
- 改进错误处理: 文档解析失败时通知用户
- 解析后的文档内容添加到AI上下文

src/app/api/chat/route.ts

@@ -6,6 +6,24 @@ import { nanoid } from 'nanoid';
 import { executeTool } from '@/services/tools';
 import { getCurrentUser } from '@/lib/auth';
 import { decryptApiKey } from '@/lib/crypto';
+import { parseDocument, type ParsedDocument } from '@/lib/document-parser';
+
+// PDF 文档数据（从前端传来，直接传给 Claude API）
+interface PdfDocumentData {
+  name: string;
+  size: number;
+  data: string; // Base64 编码
+  media_type: 'application/pdf';
+}
+
+// Office 文档数据（从前端传来，需要后端解析）
+interface OfficeDocumentData {
+  name: string;
+  size: number;
+  data: string; // Base64 编码
+  type: 'word' | 'excel';
+  mimeType: string;
+}
 
 interface ChatRequest {
   conversationId: string;
@@ -29,6 +47,10 @@ interface ChatRequest {
     type: string;
     content: string;
   }[];
+  // PDF 文档（直接传给 Claude API 使用原生 document 类型）
+  pdfDocuments?: PdfDocumentData[];
+  // Office 文档（Word/Excel，需要后端解析）
+  officeDocuments?: OfficeDocumentData[];
 }
 
 // 消息内容块类型（Claude）
@@ -240,9 +262,9 @@ const DEFAULT_SYSTEM_PROMPT = `你是一个专业、友好的 AI 助手。请遵
 export async function POST(request: Request) {
   try {
     const body: ChatRequest = await request.json();
-    const { conversationId, message, displayMessage, model, tools, enableThinking, images, uploadedImages, uploadedDocuments } = body;
+    const { conversationId, message, displayMessage, model, tools, enableThinking, images, uploadedImages, uploadedDocuments, pdfDocuments, officeDocuments } = body;
 
-    // 调试日志：确认接收到的图片数据
+    // 调试日志：确认接收到的图片和文档数据
     console.log('[API/chat] Received request with:', {
       conversationId,
       messageLength: message?.length,
@@ -253,11 +275,24 @@ export async function POST(request: Request) {
       imagesCount: images?.length || 0,
       uploadedImagesCount: uploadedImages?.length || 0,
       uploadedDocumentsCount: uploadedDocuments?.length || 0,
+      pdfDocumentsCount: pdfDocuments?.length || 0,
+      officeDocumentsCount: officeDocuments?.length || 0,
       images: images ? images.map(img => ({
         type: img.type,
         media_type: img.media_type,
         dataLength: img.data?.length || 0,
       })) : undefined,
+      pdfDocuments: pdfDocuments ? pdfDocuments.map(doc => ({
+        name: doc.name,
+        size: doc.size,
+        dataLength: doc.data?.length || 0,
+      })) : undefined,
+      officeDocuments: officeDocuments ? officeDocuments.map(doc => ({
+        name: doc.name,
+        size: doc.size,
+        type: doc.type,
+        dataLength: doc.data?.length || 0,
+      })) : undefined,
     });
 
     // 获取当前登录用户
@@ -360,12 +395,43 @@ export async function POST(request: Request) {
           // 获取温度参数
           const temperature = parseFloat(conversation.temperature || settings.temperature || '0.7');
 
+          // 解析 Office 文档（Word/Excel）
+          const parsedOfficeDocuments: ParsedDocument[] = [];
+          const documentParseErrors: { name: string; error: string }[] = []; // 记录解析失败的文档
+          if (officeDocuments && officeDocuments.length > 0) {
+            console.log('[API/chat] 📄 开始解析 Office 文档，数量:', officeDocuments.length);
+            for (const doc of officeDocuments) {
+              try {
+                const parsed = await parseDocument(doc.data, doc.name, doc.mimeType);
+                if (parsed) {
+                  parsedOfficeDocuments.push(parsed);
+                  console.log('[API/chat] ✅ 成功解析文档:', doc.name, '内容长度:', parsed.content.length);
+                }
+              } catch (parseError) {
+                const errorMsg = parseError instanceof Error ? parseError.message : '未知错误';
+                console.error('[API/chat] ❌ 解析文档失败:', doc.name, parseError);
+                documentParseErrors.push({ name: doc.name, error: errorMsg });
+              }
+            }
+          }
+
           let fullContent = '';
           let thinkingContent = '';
           let totalInputTokens = 0;
           let totalOutputTokens = 0;
           let usedTools: string[] = []; // 收集使用过的工具名称
 
+          // 如果有文档解析失败，将警告添加到内容开头
+          if (documentParseErrors.length > 0) {
+            const errorMessages = documentParseErrors.map(e => `- ${e.name}: ${e.error}`).join('\n');
+            const warningText = `⚠️ 以下文档解析失败，无法分析其内容：\n${errorMessages}\n\n`;
+            fullContent = warningText;
+            safeWriter.write({
+              type: 'text',
+              content: warningText,
+            });
+          }
+
           // 【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
@@ -391,6 +457,8 @@ export async function POST(request: Request) {
               encoder,
               images,
               metasoApiKey: decryptedMetasoApiKey,
+              pdfDocuments,
+              parsedOfficeDocuments,
             });
 
             fullContent = result.fullContent;
@@ -416,6 +484,8 @@ export async function POST(request: Request) {
               encoder,
               images, // 传递用户上传的图片
               metasoApiKey: decryptedMetasoApiKey,
+              pdfDocuments,
+              parsedOfficeDocuments,
             });
 
             fullContent = result.fullContent;
@@ -439,6 +509,8 @@ export async function POST(request: Request) {
               encoder,
               images, // 传递用户上传的图片
               metasoApiKey: decryptedMetasoApiKey,
+              pdfDocuments,
+              parsedOfficeDocuments,
             });
 
             fullContent = result.fullContent;
@@ -534,6 +606,10 @@ interface CodexChatParams {
   }[];
   // 秘塔 API Key
   metasoApiKey?: string;
+  // PDF 文档（直接传给 API）
+  pdfDocuments?: PdfDocumentData[];
+  // 解析后的 Office 文档
+  parsedOfficeDocuments?: ParsedDocument[];
 }
 
 // Codex Response API 的输入项类型
@@ -574,6 +650,8 @@ async function handleCodexChat(params: CodexChatParams): Promise<{
     encoder,
     images,
     metasoApiKey,
+    pdfDocuments,
+    parsedOfficeDocuments,
   } = params;
 
   // 创建安全的 stream 写入器
@@ -591,13 +669,37 @@ async function handleCodexChat(params: CodexChatParams): Promise<{
   ];
 
   // 添加当前用户消息（支持多模态内容）
-  if (images && images.length > 0) {
+  const hasImages = images && images.length > 0;
-    console.log('[handleCodexChat] Building multimodal message with', images.length, 'images');
+  const hasPdfDocuments = pdfDocuments && pdfDocuments.length > 0;
+  const hasOfficeDocuments = parsedOfficeDocuments && parsedOfficeDocuments.length > 0;
+
+  // 构建消息文本（包含用户消息、PDF 提示和解析后的 Office 文档内容）
+  let textContent = message;
+
+  // Codex/OpenAI 格式不支持原生 PDF，添加提示信息
+  if (hasPdfDocuments) {
+    const pdfInfo = pdfDocuments!.map(doc => `[PDF 文档: ${doc.name}]`).join('\n');
+    textContent = `${textContent}\n\n注意：用户上传了以下 PDF 文件，但当前 API 格式不支持原生 PDF 解析。建议切换到 Claude 原生格式以获得完整的 PDF 支持。\n${pdfInfo}`;
+    console.log('[handleCodexChat] PDF documents detected but not supported in Codex format');
+  }
+
+  // 添加解析后的 Office 文档内容
+  if (hasOfficeDocuments) {
+    const officeContents = parsedOfficeDocuments!.map(doc => {
+      const typeLabel = doc.type === 'word' ? 'Word' : doc.type === 'excel' ? 'Excel' : '文档';
+      return `\n\n--- ${typeLabel} 文件：${doc.name} ---\n${doc.content}\n--- 文件结束 ---`;
+    }).join('\n');
+    textContent = textContent + officeContents;
+    console.log('[handleCodexChat] Added Office documents content, total length:', textContent.length);
+  }
+
+  if (hasImages) {
+    console.log('[handleCodexChat] Building multimodal message with', images!.length, 'images');
     // 如果有图片，构建多模态消息
     const multimodalContent: CodexMultimodalContent[] = [];
 
     // 先添加图片
-    for (const img of images) {
+    for (const img of images!) {
       console.log('[handleCodexChat] Adding image:', {
         type: img.type,
         media_type: img.media_type,
@@ -612,10 +714,10 @@ async function handleCodexChat(params: CodexChatParams): Promise<{
     }
 
     // 再添加文本
-    if (message) {
+    if (textContent) {
       multimodalContent.push({
         type: 'input_text',
-        text: message,
+        text: textContent,
       });
     }
 
@@ -626,12 +728,12 @@ async function handleCodexChat(params: CodexChatParams): Promise<{
       content: multimodalContent,
     });
   } else {
-    console.log('[handleCodexChat] No images, using simple text message');
+    console.log('[handleCodexChat] No images, using text message');
-    // 没有图片，使用简单文本消息
+    // 没有图片，使用文本消息
     inputItems.push({
       type: 'message' as const,
       role: 'user' as const,
-      content: message,
+      content: textContent,
     });
   }
 
@@ -912,6 +1014,10 @@ interface ClaudeChatParams {
   }[];
   // 秘塔 API Key
   metasoApiKey?: string;
+  // PDF 文档（直接传给 Claude API 使用原生 document 类型）
+  pdfDocuments?: PdfDocumentData[];
+  // 解析后的 Office 文档
+  parsedOfficeDocuments?: ParsedDocument[];
 }
 
 async function handleClaudeChat(params: ClaudeChatParams): Promise<{
@@ -935,6 +1041,8 @@ async function handleClaudeChat(params: ClaudeChatParams): Promise<{
     encoder,
     images,
     metasoApiKey,
+    pdfDocuments,
+    parsedOfficeDocuments,
   } = params;
 
   // 创建安全的 stream 写入器
@@ -948,35 +1056,72 @@ async function handleClaudeChat(params: ClaudeChatParams): Promise<{
       content: msg.content,
     }));
 
-  // 添加当前用户消息（支持多模态内容）
+  // 添加当前用户消息（支持多模态内容：图片、PDF、Office文档）
-  if (images && images.length > 0) {
+  const hasImages = images && images.length > 0;
-    console.log('[handleClaudeChat] Building multimodal message with', images.length, 'images');
+  const hasPdfDocuments = pdfDocuments && pdfDocuments.length > 0;
-    // 如果有图片，构建多模态消息
+  const hasOfficeDocuments = parsedOfficeDocuments && parsedOfficeDocuments.length > 0;
+
+  if (hasImages || hasPdfDocuments || hasOfficeDocuments) {
+    console.log('[handleClaudeChat] Building multimodal message with:', {
+      images: images?.length || 0,
+      pdfDocuments: pdfDocuments?.length || 0,
+      officeDocuments: parsedOfficeDocuments?.length || 0,
+    });
+    // 构建多模态消息
     const multimodalContent: ContentBlock[] = [];
 
-    // 先添加图片
+    // 1. 添加 PDF 文档（使用 Claude 原生 document 类型）
-    for (const img of images) {
+    if (hasPdfDocuments) {
-      console.log('[handleClaudeChat] Adding image:', {
+      for (const pdf of pdfDocuments!) {
-        type: img.type,
+        console.log('[handleClaudeChat] Adding PDF document:', pdf.name, 'size:', pdf.size);
-        media_type: img.media_type,
+        multimodalContent.push({
-        dataLength: img.data?.length || 0,
+          type: 'document' as unknown as 'text',
-      });
+          // @ts-expect-error - Claude API 支持 document 类型但 TypeScript 类型定义不完整
-      multimodalContent.push({
+          source: {
-        type: 'image' as unknown as 'text',
+            type: 'base64',
-        // @ts-expect-error - Claude API 支持 image 类型但 TypeScript 类型定义不完整
+            media_type: 'application/pdf',
-        source: {
+            data: pdf.data,
-          type: 'base64',
+          },
-          media_type: img.media_type,
+        });
-          data: img.data,
+      }
-        },
-      });
     }
 
-    // 再添加文本
+    // 2. 添加图片
-    if (message) {
+    if (hasImages) {
+      for (const img of images!) {
+        console.log('[handleClaudeChat] Adding image:', {
+          type: img.type,
+          media_type: img.media_type,
+          dataLength: img.data?.length || 0,
+        });
+        multimodalContent.push({
+          type: 'image' as unknown as 'text',
+          // @ts-expect-error - Claude API 支持 image 类型但 TypeScript 类型定义不完整
+          source: {
+            type: 'base64',
+            media_type: img.media_type,
+            data: img.data,
+          },
+        });
+      }
+    }
+
+    // 3. 构建消息文本（包含用户消息和解析后的 Office 文档内容）
+    let textContent = message;
+    if (hasOfficeDocuments) {
+      const officeContents = parsedOfficeDocuments!.map(doc => {
+        const typeLabel = doc.type === 'word' ? 'Word' : doc.type === 'excel' ? 'Excel' : '文档';
+        return `\n\n--- ${typeLabel} 文件：${doc.name} ---\n${doc.content}\n--- 文件结束 ---`;
+      }).join('\n');
+      textContent = message + officeContents;
+      console.log('[handleClaudeChat] Added Office documents content, total length:', textContent.length);
+    }
+
+    // 4. 添加文本内容
+    if (textContent) {
       multimodalContent.push({
         type: 'text',
-        text: message,
+        text: textContent,
       });
     }
 
@@ -986,8 +1131,8 @@ async function handleClaudeChat(params: ClaudeChatParams): Promise<{
       content: multimodalContent,
     });
   } else {
-    console.log('[handleClaudeChat] No images, using simple text message');
+    console.log('[handleClaudeChat] No multimodal content, using simple text message');
-    // 没有图片，使用简单文本消息
+    // 没有多模态内容，使用简单文本消息
     messageHistory.push({
       role: 'user',
       content: message,
@@ -1323,6 +1468,10 @@ interface OpenAICompatibleChatParams {
   }[];
   // 秘塔 API Key
   metasoApiKey?: string;
+  // PDF 文档
+  pdfDocuments?: PdfDocumentData[];
+  // 解析后的 Office 文档
+  parsedOfficeDocuments?: ParsedDocument[];
 }
 
 // OpenAI 消息格式
@@ -1368,6 +1517,8 @@ async function handleOpenAICompatibleChat(params: OpenAICompatibleChatParams): P
     encoder,
     images,
     metasoApiKey,
+    pdfDocuments,
+    parsedOfficeDocuments,
   } = params;
 
   // 创建安全的 stream 写入器
@@ -1385,12 +1536,36 @@ async function handleOpenAICompatibleChat(params: OpenAICompatibleChatParams): P
   ];
 
   // 添加当前用户消息（支持多模态）
-  if (images && images.length > 0) {
+  const hasImages = images && images.length > 0;
-    console.log('[handleOpenAICompatibleChat] Building multimodal message with', images.length, 'images');
+  const hasPdfDocuments = pdfDocuments && pdfDocuments.length > 0;
+  const hasOfficeDocuments = parsedOfficeDocuments && parsedOfficeDocuments.length > 0;
+
+  // 构建消息文本（包含用户消息、PDF 提示和解析后的 Office 文档内容）
+  let textContent = message;
+
+  // OpenAI 格式不支持原生 PDF，添加提示信息
+  if (hasPdfDocuments) {
+    const pdfInfo = pdfDocuments!.map(doc => `[PDF 文档: ${doc.name}]`).join('\n');
+    textContent = `${textContent}\n\n注意：用户上传了以下 PDF 文件，但当前 API 格式不支持原生 PDF 解析。建议切换到 Claude 原生格式以获得完整的 PDF 支持。\n${pdfInfo}`;
+    console.log('[handleOpenAICompatibleChat] PDF documents detected but not supported in OpenAI format');
+  }
+
+  // 添加解析后的 Office 文档内容
+  if (hasOfficeDocuments) {
+    const officeContents = parsedOfficeDocuments!.map(doc => {
+      const typeLabel = doc.type === 'word' ? 'Word' : doc.type === 'excel' ? 'Excel' : '文档';
+      return `\n\n--- ${typeLabel} 文件：${doc.name} ---\n${doc.content}\n--- 文件结束 ---`;
+    }).join('\n');
+    textContent = textContent + officeContents;
+    console.log('[handleOpenAICompatibleChat] Added Office documents content, total length:', textContent.length);
+  }
+
+  if (hasImages) {
+    console.log('[handleOpenAICompatibleChat] Building multimodal message with', images!.length, 'images');
     const multimodalContent: OpenAIMessageContent[] = [];
 
     // 添加图片
-    for (const img of images) {
+    for (const img of images!) {
       console.log('[handleOpenAICompatibleChat] Adding image:', {
         type: img.type,
         media_type: img.media_type,
@@ -1405,10 +1580,10 @@ async function handleOpenAICompatibleChat(params: OpenAICompatibleChatParams): P
     }
 
     // 添加文本
-    if (message) {
+    if (textContent) {
       multimodalContent.push({
         type: 'text',
-        text: message,
+        text: textContent,
       });
     }
 
@@ -1417,10 +1592,10 @@ async function handleOpenAICompatibleChat(params: OpenAICompatibleChatParams): P
       content: multimodalContent,
     });
   } else {
-    console.log('[handleOpenAICompatibleChat] No images, using simple text message');
+    console.log('[handleOpenAICompatibleChat] No images, using text message');
     openaiMessages.push({
       role: 'user',
-      content: message,
+      content: textContent,
     });
   }