feat: multimodal RAG with PDF image extraction and display

Extract images from PDFs using pymupdf, generate descriptions via
Qwen3-VL-8B, store in ChromaDB alongside text chunks, and render
images in chat answers. Includes image proxy rewrite, force re-process
endpoint, and VLM API timeout.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

web/next.config.js

@@ -32,6 +32,10 @@ const nextConfig = {
         source: '/generated_images/:path*',
         destination: `${backendUrl}/generated_images/:path*`,
       },
+      {
+        source: '/images/:path*',
+        destination: `${backendUrl}/images/:path*`,
+      },
     ];
   },
   // 禁用静态生成，避免 SSR 时使用浏览器 API 的错误

web/src/components/chat/message-item.tsx

@@ -252,6 +252,22 @@ export default function MessageItem({ message, selectedModel }: MessageItemProps
                     td: ({ children }) => (
                       <td className="border border-border px-2 py-1.5">{children}</td>
                     ),
+                    img: ({ src, alt }: any) => (
+                      <a href={src} target="_blank" rel="noopener noreferrer" className="block my-3 group">
+                        <img
+                          src={src}
+                          alt={alt || "图片来源"}
+                          loading="lazy"
+                          className="max-w-full max-h-80 rounded-lg border border-border/50 cursor-pointer
+                            hover:border-primary/40 transition-colors object-contain bg-muted/20"
+                        />
+                        {alt && (
+                          <span className="block text-[11px] text-muted-foreground/70 mt-1 text-center">
+                            {alt}
+                          </span>
+                        )}
+                      </a>
+                    ),
                   }}
                 >
                   {preprocessCitations(message.content, String(message.id))}

web/src/components/chat/source-references.tsx

@@ -1,7 +1,7 @@
 "use client";
 
 import { useState, useMemo } from "react";
-import { Database, Globe, ChevronDown, ChevronRight, Star } from "lucide-react";
+import { Database, Globe, ChevronDown, ChevronRight, Star, Image } from "lucide-react";
 import { Button } from "@/components/ui/button";
 import { cn } from "@/lib/utils";
 
@@ -13,7 +13,8 @@ interface SourceReference {
   score?: number;
   preview: string;
   url?: string;
-  source_type?: "web" | "rag";
+  source_type?: "web" | "rag" | "image";
+  image_url?: string;
 }
 
 interface SourceReferencesProps {
@@ -188,7 +189,8 @@ export default function SourceReferences({
   const showExpandButton = sources.length > maxSources;
   const visibleSources = expanded ? sources : displaySources;
 
-  const ragSources = visibleSources.filter((s) => s.source_type !== "web");
+  const ragSources = visibleSources.filter((s) => s.source_type !== "web" && s.source_type !== "image");
+  const imageSources = visibleSources.filter((s) => s.source_type === "image");
   const webSources = visibleSources.filter((s) => s.source_type === "web");
 
   const sourceSection = (
@@ -218,6 +220,31 @@ export default function SourceReferences({
         </div>
       )}
 
+      {imageSources.length > 0 && (
+        <div className="space-y-1">
+          <div className="flex items-center gap-1.5 text-[11px] font-medium text-muted-foreground px-0.5">
+            <Image className="h-3 w-3 text-purple-500" />
+            <span>
+              图片来源 ({imageSources.length}
+              {expanded && showExpandButton
+                ? `/${sources.filter((s) => s.source_type === "image").length}`
+                : ""}
+              )
+            </span>
+          </div>
+          <div className="divide-y divide-border/30">
+            {imageSources.map((source, i) => (
+              <SourceRow
+                key={i}
+                source={source}
+                answerContent={answerContent}
+                messageId={messageId}
+              />
+            ))}
+          </div>
+        </div>
+      )}
+
       {webSources.length > 0 && (
         <div className="space-y-1">
           <div className="flex items-center gap-1.5 text-[11px] font-medium text-muted-foreground px-0.5">

web/src/types/index.ts

@@ -101,7 +101,8 @@ export interface SourceInfo {
   score?: number;
   preview: string;
   url?: string;
-  source_type?: "web" | "rag";
+  source_type?: "web" | "rag" | "image";
+  image_url?: string;
 }
 
 export interface ChatResponse {