.

chatdesk-ui/app/api/retrieval/retrieve/route.ts

@@ -1,4 +1,5 @@
 import { generateLocalEmbedding } from "@/lib/generate-local-embedding"
+import { generateBgeM3Embedding } from "@/lib/generate-bge-m3-embedding"  // 新增
 import { checkApiKey, getServerProfile } from "@/lib/server/server-chat-helpers"
 import { Database } from "@/supabase/types"
 import { createClient } from "@supabase/supabase-js"
@@ -86,6 +87,18 @@ export async function POST(request: Request) {
       chunks = localFileItems
     } else if (embeddingsProvider === "bge-m3"){
       // 示例：调用你自己的 BGE-M3 API 或本地函数
+      // 新增：使用 BGE-M3 嵌入
+      const bgeEmbedding = await generateBgeM3Embedding(userInput)
+
+      // 调用对应的 RPC，需要在数据库侧提前定义 match_file_items_bge_m3
+      const { data: bgeFileItems, error: bgeError } =
+        await supabaseAdmin.rpc("match_file_items_bge_m3", {
+          query_embedding: bgeEmbedding as any,
+          match_count: sourceCount,
+          file_ids: uniqueFileIds
+        })
+      if (bgeError) throw bgeError
+      chunks = bgeFileItems
     }
 
     const mostSimilarChunks = chunks?.sort(

chatdesk-ui/lib/generate-bgem3-embedding.ts

@@ -1,19 +1,36 @@
 export async function generateBgeM3Embedding(text: string): Promise<number[] | null> {
   try {
-    const response = await fetch("http://localhost:8000/embedding", {
+    // 动态获取当前协议和主机（不含端口），然后指定后端端口 8001
+    const { protocol, host } = window.location;
+    const hostname = host.split(":")[0];
+    const apiUrl = `${protocol}//${hostname}:8001/v1/embeddings`;
+
+    const response = await fetch(apiUrl, {
       method: "POST",
       headers: { "Content-Type": "application/json" },
-      body: JSON.stringify({ input: text })
+      body: JSON.stringify({
+        // OpenAI 兼容请求字段
+        input: text,
+        model: "text-embedding-bge-m3"
       })
+    });
 
     if (!response.ok) {
-      throw new Error(`Failed to fetch BGE-M3 embedding: ${response.status}`)
+      throw new Error(`Failed to fetch BGE-M3 embedding: ${response.status}`);
     }
 
-    const result = await response.json()
-    return result.embedding as number[]
+    // 返回结构为 { object, data: [{ embedding, … }], model, usage }
+    const result = await response.json();
+
+    // 取 data[0].embedding
+    if (Array.isArray(result.data) && result.data.length > 0) {
+      return result.data[0].embedding as number[];
+    } else {
+      console.error("Unexpected embedding response format:", result);
+      return null;
+    }
   } catch (err) {
-    console.error("Error in generateBgeM3Embedding:", err)
-    return null
+    console.error("Error in generateBgeM3Embedding:", err);
+    return null;
   }
 }