{"id":78763,"date":"2026-02-28T07:51:38","date_gmt":"2026-02-27T23:51:38","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/78763.html"},"modified":"2026-02-28T07:51:38","modified_gmt":"2026-02-27T23:51:38","slug":"rag%e5%ba%94%e7%94%a8%e6%95%b0%e6%8d%ae%e5%ba%93%e9%80%89%e5%9e%8b%ef%bc%9aduckdb%e3%80%81milvus%e4%b8%8esurrealdb%e6%b7%b1%e5%ba%a6%e5%af%b9%e6%af%94%e5%88%86%e6%9e%90","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/78763.html","title":{"rendered":"RAG\u5e94\u7528\u6570\u636e\u5e93\u9009\u578b\uff1aDuckDB\u3001Milvus\u4e0eSurrealDB\u6df1\u5ea6\u5bf9\u6bd4\u5206\u6790"},"content":{"rendered":"

\u6458\u8981<\/h3>\n

\u68c0\u7d22\u589e\u5f3a\u751f\u6210（Retrieval-Augmented Generation，RAG）\u6280\u672f\u5df2\u6210\u4e3a\u5927\u8bed\u8a00\u6a21\u578b\u5e94\u7528\u843d\u5730\u7684\u6838\u5fc3\u67b6\u6784\u6a21\u5f0f\u3002\u5728RAG\u7cfb\u7edf\u4e2d，\u5411\u91cf\u6570\u636e\u5e93\u627f\u62c5\u7740\u8bed\u4e49\u68c0\u7d22\u7684\u5173\u952e\u804c\u8d23，\u5176\u9009\u578b\u76f4\u63a5\u5f71\u54cd\u7cfb\u7edf\u7684\u68c0\u7d22\u7cbe\u5ea6\u3001\u54cd\u5e94\u5ef6\u8fdf\u548c\u6269\u5c55\u80fd\u529b\u3002\u672c\u6587\u5c06\u4ece\u6280\u672f\u67b6\u6784\u3001\u6027\u80fd\u7279\u5f81\u3001\u9002\u7528\u573a\u666f\u7b49\u7ef4\u5ea6，\u5bf9DuckDB\u3001Milvus\u548cSurrealDB\u4e09\u6b3e\u6570\u636e\u5e93\u5728RAG\u5e94\u7528\u4e2d\u7684\u8868\u73b0\u8fdb\u884c\u7cfb\u7edf\u6027\u5bf9\u6bd4\u5206\u6790，\u4e3a\u6280\u672f\u9009\u578b\u63d0\u4f9b\u53c2\u8003\u4f9d\u636e\u3002<\/p>\n

1. RAG\u6280\u672f\u67b6\u6784\u4e0e\u6570\u636e\u5e93\u9700\u6c42<\/h3>\n

1.1 RAG\u5de5\u4f5c\u6d41\u7a0b<\/h4>\n

RAG\u7cfb\u7edf\u7684\u5178\u578b\u5de5\u4f5c\u6d41\u7a0b\u5305\u542b\u4ee5\u4e0b\u73af\u8282：<\/p>\n

  • \u6587\u6863\u5904\u7406：\u5bf9\u539f\u59cb\u6587\u6863\u8fdb\u884c\u5206\u5757\u3001\u6e05\u6d17\u548c\u9884\u5904\u7406<\/li>\n
  • \u5411\u91cf\u5316：\u901a\u8fc7Embedding\u6a21\u578b\u5c06\u6587\u672c\u8f6c\u6362\u4e3a\u9ad8\u7ef4\u5411\u91cf<\/li>\n
  • \u5411\u91cf\u5b58\u50a8：\u5c06\u5411\u91cf\u53ca\u5143\u6570\u636e\u6301\u4e45\u5316\u5b58\u50a8<\/li>\n
  • \u8bed\u4e49\u68c0\u7d22：\u6839\u636e\u67e5\u8be2\u5411\u91cf\u68c0\u7d22\u76f8\u4f3c\u6587\u6863<\/li>\n
  • \u4e0a\u4e0b\u6587\u589e\u5f3a：\u5c06\u68c0\u7d22\u7ed3\u679c\u4f5c\u4e3a\u4e0a\u4e0b\u6587\u8f93\u5165LLM\u751f\u6210\u56de\u7b54<\/li>\n

    1.2 \u6570\u636e\u5e93\u6838\u5fc3\u9700\u6c42<\/h4>\n

    RAG\u5e94\u7528\u5bf9\u6570\u636e\u5e93\u63d0\u51fa\u4e86\u591a\u7ef4\u5ea6\u7684\u6280\u672f\u8981\u6c42：<\/p>\n\n\u9700\u6c42\u7ef4\u5ea6\u5177\u4f53\u8981\u6c42<\/tr>\n\n\n\n\n\n\n
    \u5411\u91cf\u68c0\u7d22<\/td>\n\u652f\u6301\u9ad8\u7ef4\u5411\u91cf\u7684\u76f8\u4f3c\u5ea6\u8ba1\u7b97（\u4f59\u5f26\u3001\u6b27\u6c0f\u8ddd\u79bb\u7b49）<\/td>\n<\/tr>\n
    \u68c0\u7d22\u6027\u80fd<\/td>\n\u5927\u89c4\u6a21\u5411\u91cf\u96c6\u5408\u4e0b\u7684\u4f4e\u5ef6\u8fdf\u67e5\u8be2<\/td>\n<\/tr>\n
    \u6df7\u5408\u67e5\u8be2<\/td>\n\u5411\u91cf\u68c0\u7d22\u4e0e\u6807\u91cf\u8fc7\u6ee4\u7684\u7ec4\u5408\u80fd\u529b<\/td>\n<\/tr>\n
    \u6269\u5c55\u6027<\/td>\n\u6570\u636e\u91cf\u589e\u957f\u65f6\u7684\u6c34\u5e73\u6269\u5c55\u80fd\u529b<\/td>\n<\/tr>\n
    \u5143\u6570\u636e\u7ba1\u7406<\/td>\n\u6587\u6863\u5c5e\u6027\u3001\u6743\u9650\u7b49\u7ed3\u6784\u5316\u6570\u636e\u7684\u5b58\u50a8\u4e0e\u67e5\u8be2<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    2. \u6570\u636e\u5e93\u6280\u672f\u5b9a\u4f4d\u5bf9\u6bd4<\/h3>\n

    2.1 \u6838\u5fc3\u5b9a\u4f4d\u5dee\u5f02<\/h4>\n

    \u4e09\u6b3e\u6570\u636e\u5e93\u5728\u8bbe\u8ba1\u7406\u5ff5\u548c\u6280\u672f\u5b9a\u4f4d\u4e0a\u5b58\u5728\u672c\u8d28\u5dee\u5f02：<\/p>\n\n\u6570\u636e\u5e93\u7c7b\u578b\u5b9a\u4f4d\u8bbe\u8ba1\u76ee\u6807\u5f00\u6e90\u534f\u8bae<\/tr>\n\n\n\n\n
    DuckDB<\/td>\n\u5d4c\u5165\u5f0fOLAP\u6570\u636e\u5e93<\/td>\n\u9ad8\u6027\u80fd\u5206\u6790\u67e5\u8be2<\/td>\nMIT<\/td>\n<\/tr>\n
    Milvus<\/td>\n\u4e13\u7528\u5411\u91cf\u6570\u636e\u5e93<\/td>\n\u5927\u89c4\u6a21\u5411\u91cf\u76f8\u4f3c\u5ea6\u68c0\u7d22<\/td>\nApache 2.0<\/td>\n<\/tr>\n
    SurrealDB<\/td>\n\u591a\u6a21\u578b\u6570\u636e\u5e93<\/td>\n\u6587\u6863\u3001\u56fe\u3001\u5411\u91cf\u4e00\u4f53\u5316<\/td>\nBSL 1.1<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    2.2 \u67b6\u6784\u7279\u5f81\u5bf9\u6bd4<\/h4>\n

    DuckDB\u67b6\u6784\u7279\u5f81<\/h5>\n

    DuckDB\u91c7\u7528\u8fdb\u7a0b\u5185\u5d4c\u5165\u5f0f\u67b6\u6784，\u5176\u6838\u5fc3\u7279\u5f81\u5305\u62ec：<\/p>\n

      \n
    • \u5217\u5f0f\u5b58\u50a8\u5f15\u64ce：\u9488\u5bf9\u5206\u6790\u578b\u67e5\u8be2\u4f18\u5316\u7684\u5b58\u50a8\u683c\u5f0f<\/li>\n
    • \u5411\u91cf\u5316\u6267\u884c：\u6279\u91cf\u6570\u636e\u5904\u7406，\u5145\u5206\u5229\u7528CPU\u7f13\u5b58\u548cSIMD\u6307\u4ee4<\/li>\n
    • \u96f6\u5916\u90e8\u4f9d\u8d56：\u5355\u4e00\u5e93\u6587\u4ef6，\u65e0\u9700\u72ec\u7acb\u670d\u52a1\u8fdb\u7a0b<\/li>\n
    • SQL\u6807\u51c6\u517c\u5bb9：\u63d0\u4f9b\u4e30\u5bcc\u7684SQL\u65b9\u8a00\u652f\u6301<\/li>\n<\/ul>\n
      Milvus\u67b6\u6784\u7279\u5f81<\/h5>\n

      Milvus\u91c7\u7528\u5206\u5e03\u5f0f\u4e91\u539f\u751f\u67b6\u6784，\u6838\u5fc3\u7ec4\u4ef6\u5305\u62ec：<\/p>\n

        \n
      • \u5b58\u7b97\u5206\u79bb：\u8ba1\u7b97\u8282\u70b9\u4e0e\u5b58\u50a8\u8282\u70b9\u72ec\u7acb\u6269\u5c55<\/li>\n
      • \u591a\u7ea7\u7d22\u5f15：\u652f\u6301HNSW\u3001IVF_FLAT\u3001IVF_PQ\u3001ANNOY\u7b49\u591a\u79cdANN\u7d22\u5f15<\/li>\n
      • \u5206\u7247\u673a\u5236：\u6570\u636e\u81ea\u52a8\u5206\u7247，\u652f\u6301\u6c34\u5e73\u6269\u5c55<\/li>\n
      • \u4f9d\u8d56\u7ec4\u4ef6：etcd（\u5143\u6570\u636e）\u3001MinIO\/S3（\u5bf9\u8c61\u5b58\u50a8）\u3001Pulsar\/Kafka（\u6d88\u606f\u961f\u5217）<\/li>\n<\/ul>\n
        SurrealDB\u67b6\u6784\u7279\u5f81<\/h5>\n

        SurrealDB\u91c7\u7528\u591a\u6a21\u578b\u7edf\u4e00\u67b6\u6784：<\/p>\n

          \n
        • \u591a\u6a21\u578b\u652f\u6301：\u6587\u6863\u5b58\u50a8\u3001\u56fe\u6570\u636e\u5e93\u3001\u5411\u91cf\u68c0\u7d22\u4e09\u4f4d\u4e00\u4f53<\/li>\n
        • \u5b9e\u65f6\u80fd\u529b：\u5185\u7f6eWebSocket\u652f\u6301，\u6570\u636e\u53d8\u66f4\u5b9e\u65f6\u63a8\u9001<\/li>\n
        • \u5355\u4f53\u90e8\u7f72：\u5355\u4e00\u4e8c\u8fdb\u5236\u6587\u4ef6，\u652f\u6301\u5d4c\u5165\u5f0f\u548c\u670d\u52a1\u5668\u6a21\u5f0f<\/li>\n
        • SurrealQL：\u81ea\u7814\u67e5\u8be2\u8bed\u8a00，\u878d\u5408SQL\u4e0e\u56fe\u67e5\u8be2\u8bed\u6cd5<\/li>\n<\/ul>\n

          3. RAG\u573a\u666f\u80fd\u529b\u8be6\u7ec6\u5bf9\u6bd4<\/h3>\n

          3.1 \u5411\u91cf\u68c0\u7d22\u80fd\u529b<\/h4>\n

          DuckDB\u5411\u91cf\u68c0\u7d22<\/h5>\n

          DuckDB\u901a\u8fc7\u5185\u7f6e\u6570\u7ec4\u51fd\u6570\u63d0\u4f9b\u57fa\u7840\u5411\u91cf\u68c0\u7d22\u80fd\u529b：<\/p>\n

          — DuckDB\u5411\u91cf\u76f8\u4f3c\u5ea6\u67e5\u8be2<\/span>
          \nSELECT<\/span>
          \n doc_id,<\/span>
          \n content,<\/span>
          \n list_cosine_similarity(<\/span>embedding,<\/span> $query_embedding)<\/span> AS<\/span> similarity_score
          \nFROM<\/span> documents
          \nORDER<\/span> BY<\/span> similarity_score DESC<\/span>
          \nLIMIT<\/span> 10<\/span>;<\/span><\/p>\n

          \u6280\u672f\u7279\u70b9\u5206\u6790：<\/p>\n

            \n
          • \u91c7\u7528\u66b4\u529b\u641c\u7d22（Brute-Force）\u65b9\u5f0f，\u7cbe\u786e\u5ea6\u9ad8<\/li>\n
          • \u7f3a\u4e4f\u4e13\u4e1aANN\u7d22\u5f15\u652f\u6301，\u5927\u89c4\u6a21\u6570\u636e\u6027\u80fd\u53d7\u9650<\/li>\n
          • \u5411\u91cf\u7ef4\u5ea6\u548c\u6570\u636e\u89c4\u6a21\u53d7\u5185\u5b58\u9650\u5236<\/li>\n<\/ul>\n
            Milvus\u5411\u91cf\u68c0\u7d22<\/h5>\n

            Milvus\u63d0\u4f9b\u4e13\u4e1a\u7ea7\u5411\u91cf\u68c0\u7d22\u80fd\u529b：<\/p>\n

            from<\/span> pymilvus import<\/span> Collection,<\/span> connections<\/p>\n

            connections.<\/span>connect(<\/span>host=<\/span>"localhost"<\/span>,<\/span> port=<\/span>"19530"<\/span>)<\/span>
            \ncollection =<\/span> Collection(<\/span>"rag_documents"<\/span>)<\/span><\/p>\n

            search_params =<\/span> {<\/span>
            \n "metric_type"<\/span>:<\/span> "COSINE"<\/span>,<\/span>
            \n "params"<\/span>:<\/span> {<\/span>"nprobe"<\/span>:<\/span> 16<\/span>}<\/span>
            \n}<\/span><\/p>\n

            results =<\/span> collection.<\/span>search(<\/span>
            \n data=<\/span>[<\/span>query_embedding]<\/span>,<\/span>
            \n anns_field=<\/span>"embedding"<\/span>,<\/span>
            \n param=<\/span>search_params,<\/span>
            \n limit=<\/span>10<\/span>,<\/span>
            \n expr=<\/span>"category == 'technical' AND publish_date > '2024-01-01'"<\/span>,<\/span>
            \n output_fields=<\/span>[<\/span>"doc_id"<\/span>,<\/span> "content"<\/span>,<\/span> "metadata"<\/span>]<\/span>
            \n)<\/span><\/p>\n

            \u6280\u672f\u7279\u70b9\u5206\u6790：<\/p>\n

              \n
            • \u652f\u6301\u591a\u79cdANN\u7d22\u5f15\u7b97\u6cd5，\u53ef\u6839\u636e\u573a\u666f\u9009\u62e9\u7cbe\u5ea6\u4e0e\u6027\u80fd\u7684\u5e73\u8861\u70b9<\/li>\n
            • \u539f\u751f\u652f\u6301\u5411\u91cf\u68c0\u7d22\u4e0e\u6807\u91cf\u8fc7\u6ee4\u7684\u6df7\u5408\u67e5\u8be2<\/li>\n
            • \u4ebf\u7ea7\u5411\u91cf\u89c4\u6a21\u4e0b\u4ecd\u53ef\u4fdd\u6301\u6beb\u79d2\u7ea7\u54cd\u5e94<\/li>\n<\/ul>\n
              SurrealDB\u5411\u91cf\u68c0\u7d22<\/h5>\n

              SurrealDB\u901a\u8fc7\u5185\u7f6e\u5411\u91cf\u51fd\u6570\u63d0\u4f9b\u68c0\u7d22\u80fd\u529b：<\/p>\n

              — SurrealDB\u5411\u91cf\u68c0\u7d22<\/span>
              \nSELECT<\/span>
              \n id,<\/span>
              \n content,<\/span>
              \n vector::similarity::cosine(<\/span>embedding,<\/span> $query_vec)<\/span> AS<\/span> score
              \nFROM<\/span> document
              \nWHERE<\/span> score ><\/span> 0.7<\/span>
              \nORDER<\/span> BY<\/span> score DESC<\/span>
              \nLIMIT<\/span> 10<\/span>;<\/span><\/p>\n

              \u6280\u672f\u7279\u70b9\u5206\u6790：<\/p>\n

                \n
              • \u5411\u91cf\u68c0\u7d22\u4e0e\u6587\u6863\u3001\u56fe\u67e5\u8be2\u7edf\u4e00\u5728\u540c\u4e00\u67e5\u8be2\u8bed\u8a00\u4e2d<\/li>\n
              • \u652f\u6301\u57fa\u7840\u7684\u5411\u91cf\u76f8\u4f3c\u5ea6\u8ba1\u7b97<\/li>\n
              • \u7d22\u5f15\u80fd\u529b\u76f8\u5bf9\u6709\u9650，\u5927\u89c4\u6a21\u573a\u666f\u6027\u80fd\u5f85\u9a8c\u8bc1<\/li>\n<\/ul>\n

                3.2 \u6df7\u5408\u67e5\u8be2\u80fd\u529b<\/h4>\n

                RAG\u5e94\u7528\u901a\u5e38\u9700\u8981\u7ed3\u5408\u8bed\u4e49\u68c0\u7d22\u4e0e\u7ed3\u6784\u5316\u8fc7\u6ee4，\u4e09\u6b3e\u6570\u636e\u5e93\u7684\u6df7\u5408\u67e5\u8be2\u80fd\u529b\u5b58\u5728\u663e\u8457\u5dee\u5f02\u3002<\/p>\n

                DuckDB\u6df7\u5408\u67e5\u8be2<\/h5>\n

                — DuckDB\u6df7\u5408\u67e5\u8be2：\u5411\u91cf\u68c0\u7d22 + \u5173\u7cfb\u578b\u8fc7\u6ee4 + \u805a\u5408\u5206\u6790<\/span>
                \nWITH<\/span> semantic_matches AS<\/span> (<\/span>
                \n SELECT<\/span>
                \n doc_id,<\/span>
                \n content,<\/span>
                \n category,<\/span>
                \n list_cosine_similarity(<\/span>embedding,<\/span> $query_vec)<\/span> AS<\/span> score
                \n FROM<\/span> documents
                \n WHERE<\/span> publish_date >=<\/span> '2024-01-01'<\/span>
                \n AND<\/span> status<\/span> =<\/span> 'published'<\/span>
                \n)<\/span>
                \nSELECT<\/span>
                \n category,<\/span>
                \n COUNT<\/span>(<\/span>*<\/span>)<\/span> AS<\/span> match_count,<\/span>
                \n AVG<\/span>(<\/span>score)<\/span> AS<\/span> avg_relevance,<\/span>
                \n ARRAY_AGG(<\/span>content ORDER<\/span> BY<\/span> score DESC<\/span>)<\/span>[<\/span>1<\/span>:3<\/span>]<\/span> AS<\/span> top_contents
                \nFROM<\/span> semantic_matches
                \nWHERE<\/span> score ><\/span> 0.6<\/span>
                \nGROUP<\/span> BY<\/span> category
                \nORDER<\/span> BY<\/span> avg_relevance DESC<\/span>;<\/span><\/p>\n

                DuckDB\u5728\u590d\u6742SQL\u67e5\u8be2\u65b9\u9762\u5177\u6709\u663e\u8457\u4f18\u52bf，\u652f\u6301\u7a97\u53e3\u51fd\u6570\u3001CTE\u3001\u5b50\u67e5\u8be2\u7b49\u9ad8\u7ea7\u7279\u6027\u3002<\/p>\n

                Milvus\u6df7\u5408\u67e5\u8be2<\/h5>\n

                # Milvus\u6df7\u5408\u67e5\u8be2：\u5411\u91cf\u68c0\u7d22 + \u6807\u91cf\u8fc7\u6ee4<\/span>
                \nresults =<\/span> collection.<\/span>search(<\/span>
                \n data=<\/span>[<\/span>query_embedding]<\/span>,<\/span>
                \n anns_field=<\/span>"embedding"<\/span>,<\/span>
                \n param=<\/span>{<\/span>"metric_type"<\/span>:<\/span> "COSINE"<\/span>,<\/span> "params"<\/span>:<\/span> {<\/span>"nprobe"<\/span>:<\/span> 10<\/span>}<\/span>}<\/span>,<\/span>
                \n limit=<\/span>20<\/span>,<\/span>
                \n expr=<\/span>"""
                \n category in ['tech', 'science']
                \n AND word_count > 500
                \n AND publish_date > '2024-01-01'
                \n """<\/span>,<\/span>
                \n output_fields=<\/span>[<\/span>"doc_id"<\/span>,<\/span> "content"<\/span>,<\/span> "category"<\/span>,<\/span> "publish_date"<\/span>]<\/span>
                \n)<\/span><\/p>\n

                Milvus\u7684\u6807\u91cf\u8fc7\u6ee4\u80fd\u529b\u76f8\u5bf9\u57fa\u7840，\u4e0d\u652f\u6301\u590d\u6742\u7684JOIN\u548c\u805a\u5408\u64cd\u4f5c\u3002<\/p>\n

                SurrealDB\u6df7\u5408\u67e5\u8be2<\/h5>\n

                — SurrealDB\u6df7\u5408\u67e5\u8be2：\u5411\u91cf + \u56fe\u904d\u5386<\/span>
                \nSELECT<\/span>
                \n id,<\/span>
                \n content,<\/span>
                \n vector::similarity::cosine(<\/span>embedding,<\/span> $query_vec)<\/span> AS<\/span> score,<\/span>
                \n –<\/span>><\/span>cites–<\/span>><\/span>document.<\/span>title AS<\/span> cited_documents,<\/span>
                \n <<\/span>–<\/span>written_by<<\/span>–<\/span>author.<\/span>name AS<\/span> authors
                \nFROM<\/span> document
                \nWHERE<\/span> score ><\/span> 0.7<\/span>
                \n AND<\/span> –<\/span>><\/span>belongs_to–<\/span>><\/span>category.<\/span>name =<\/span> 'Machine Learning'<\/span>
                \nORDER<\/span> BY<\/span> score DESC<\/span>
                \nLIMIT<\/span> 10<\/span>;<\/span><\/p>\n

                SurrealDB\u7684\u72ec\u7279\u4f18\u52bf\u5728\u4e8e\u5c06\u5411\u91cf\u68c0\u7d22\u4e0e\u56fe\u904d\u5386\u67e5\u8be2\u65e0\u7f1d\u7ed3\u5408\u3002<\/p>\n

                3.3 \u6269\u5c55\u6027\u4e0e\u90e8\u7f72\u6a21\u5f0f<\/h4>\n\n\u7279\u6027DuckDBMilvusSurrealDB<\/tr>\n\n\n\n\n\n\n
                \u90e8\u7f72\u6a21\u5f0f<\/td>\n\u5d4c\u5165\u5f0f\/\u5355\u673a<\/td>\n\u5206\u5e03\u5f0f\u96c6\u7fa4<\/td>\n\u5355\u673a\/\u96c6\u7fa4<\/td>\n<\/tr>\n
                \u6c34\u5e73\u6269\u5c55<\/td>\n\u4e0d\u652f\u6301<\/td>\n\u539f\u751f\u652f\u6301<\/td>\n\u6709\u9650\u652f\u6301<\/td>\n<\/tr>\n
                \u9ad8\u53ef\u7528<\/td>\n\u4e0d\u652f\u6301<\/td>\n\u652f\u6301（\u591a\u526f\u672c）<\/td>\n\u652f\u6301（TiKV\u540e\u7aef）<\/td>\n<\/tr>\n
                \u90e8\u7f72\u590d\u6742\u5ea6<\/td>\n\u6781\u4f4e<\/td>\n\u8f83\u9ad8<\/td>\n\u4e2d\u7b49<\/td>\n<\/tr>\n
                \u8d44\u6e90\u5360\u7528<\/td>\n\u6781\u4f4e<\/td>\n\u8f83\u9ad8<\/td>\n\u4e2d\u7b49<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

                3.4 \u751f\u6001\u96c6\u6210\u80fd\u529b<\/h4>\n

                DuckDB\u751f\u6001\u96c6\u6210<\/h5>\n

                import<\/span> duckdb
                \nimport<\/span> pandas as<\/span> pd
                \nfrom<\/span> sentence_transformers import<\/span> SentenceTransformer<\/p>\n

                # \u4e0ePython\u6570\u636e\u79d1\u5b66\u751f\u6001\u65e0\u7f1d\u96c6\u6210<\/span>
                \nmodel =<\/span> SentenceTransformer(<\/span>'all-MiniLM-L6-v2'<\/span>)<\/span>
                \ndf =<\/span> pd.<\/span>read_parquet(<\/span>'documents.parquet'<\/span>)<\/span>
                \ndf[<\/span>'embedding'<\/span>]<\/span> =<\/span> df[<\/span>'content'<\/span>]<\/span>.<\/span>apply<\/span>(<\/span>lambda<\/span> x:<\/span> model.<\/span>encode(<\/span>x)<\/span>.<\/span>tolist(<\/span>)<\/span>)<\/span><\/p>\n

                # \u76f4\u63a5\u67e5\u8be2DataFrame<\/span>
                \nconn =<\/span> duckdb.<\/span>connect(<\/span>)<\/span>
                \nresults =<\/span> conn.<\/span>execute(<\/span>"""
                \n SELECT content, list_cosine_similarity(embedding, ?) AS score
                \n FROM df
                \n ORDER BY score DESC LIMIT 10
                \n"""<\/span>,<\/span> [<\/span>query_embedding]<\/span>)<\/span>.<\/span>fetchdf(<\/span>)<\/span><\/p>\n

                Milvus\u751f\u6001\u96c6\u6210<\/h5>\n

                from<\/span> langchain.<\/span>vectorstores import<\/span> Milvus
                \nfrom<\/span> langchain.<\/span>embeddings import<\/span> OpenAIEmbeddings<\/p>\n

                # LangChain\u539f\u751f\u96c6\u6210<\/span>
                \nvectorstore =<\/span> Milvus.<\/span>from_documents(<\/span>
                \n documents,<\/span>
                \n OpenAIEmbeddings(<\/span>)<\/span>,<\/span>
                \n connection_args=<\/span>{<\/span>"host"<\/span>:<\/span> "localhost"<\/span>,<\/span> "port"<\/span>:<\/span> "19530"<\/span>}<\/span>,<\/span>
                \n collection_name=<\/span>"langchain_docs"<\/span>
                \n)<\/span><\/p>\n

                # \u68c0\u7d22\u5668\u63a5\u53e3<\/span>
                \nretriever =<\/span> vectorstore.<\/span>as_retriever(<\/span>search_kwargs=<\/span>{<\/span>"k"<\/span>:<\/span> 5<\/span>}<\/span>)<\/span><\/p>\n

                SurrealDB\u751f\u6001\u96c6\u6210<\/h5>\n

                from<\/span> surrealdb import<\/span> Surreal<\/p>\n

                async<\/span> with<\/span> Surreal(<\/span>"ws:\/\/localhost:8000\/rpc"<\/span>)<\/span> as<\/span> db:<\/span>
                \n await<\/span> db.<\/span>signin(<\/span>{<\/span>"user"<\/span>:<\/span> "root"<\/span>,<\/span> "pass"<\/span>:<\/span> "root"<\/span>}<\/span>)<\/span>
                \n await<\/span> db.<\/span>use(<\/span>"namespace"<\/span>,<\/span> "database"<\/span>)<\/span><\/p>\n

                # \u63d2\u5165\u5e26\u5411\u91cf\u7684\u6587\u6863<\/span>
                \n await<\/span> db.<\/span>create(<\/span>"document"<\/span>,<\/span> {<\/span>
                \n "content"<\/span>:<\/span> "RAG\u6280\u672f\u6587\u6863"<\/span>,<\/span>
                \n "embedding"<\/span>:<\/span> embedding_vector,<\/span>
                \n "metadata"<\/span>:<\/span> {<\/span>"category"<\/span>:<\/span> "tech"<\/span>}<\/span>
                \n }<\/span>)<\/span><\/p>\n

                4. \u5178\u578b\u5e94\u7528\u573a\u666f\u63a8\u8350<\/h3>\n

                4.1 DuckDB\u9002\u7528\u573a\u666f<\/h4>\n

                \u63a8\u8350\u573a\u666f：<\/p>\n

                  \n
                • \u539f\u578b\u5f00\u53d1\u4e0e\u6982\u5ff5\u9a8c\u8bc1（POC）\u9636\u6bb5<\/li>\n
                • \u6587\u6863\u89c4\u6a21\u572810\u4e07\u6761\u4ee5\u5185\u7684\u5c0f\u578bRAG\u5e94\u7528<\/li>\n
                • \u9700\u8981\u590d\u6742SQL\u5206\u6790\u7684\u6df7\u5408\u67e5\u8be2\u573a\u666f<\/li>\n
                • \u8fb9\u7f18\u8bbe\u5907\u6216\u79bb\u7ebf\u73af\u5883\u90e8\u7f72<\/li>\n
                • \u6570\u636e\u79d1\u5b66\u5de5\u4f5c\u6d41\u4e2d\u7684\u4e34\u65f6\u5411\u91cf\u68c0\u7d22\u9700\u6c42<\/li>\n<\/ul>\n

                  \u5178\u578b\u7528\u4f8b：<\/p>\n

                  # \u672c\u5730\u77e5\u8bc6\u5e93\u95ee\u7b54\u539f\u578b<\/span>
                  \nimport<\/span> duckdb<\/p>\n

                  def<\/span> simple_rag_query<\/span>(<\/span>question:<\/span> str<\/span>,<\/span> conn:<\/span> duckdb.<\/span>DuckDBPyConnection)<\/span>:<\/span>
                  \n query_embedding =<\/span> embed_model.<\/span>encode(<\/span>question)<\/span><\/p>\n

                  results =<\/span> conn.<\/span>execute(<\/span>"""
                  \n SELECT content, metadata
                  \n FROM knowledge_base
                  \n WHERE list_cosine_similarity(embedding, ?) > 0.7
                  \n ORDER BY list_cosine_similarity(embedding, ?) DESC
                  \n LIMIT 5
                  \n """<\/span>,<\/span> [<\/span>query_embedding,<\/span> query_embedding]<\/span>)<\/span>.<\/span>fetchall(<\/span>)<\/span><\/p>\n

                  context =<\/span> "\\\\n"<\/span>.<\/span>join(<\/span>[<\/span>r[<\/span>0<\/span>]<\/span> for<\/span> r in<\/span> results]<\/span>)<\/span>
                  \n return<\/span> llm.<\/span>generate(<\/span>f"\u57fa\u4e8e\u4ee5\u4e0b\u5185\u5bb9\u56de\u7b54\u95ee\u9898：\\\\n<\/span>{<\/span>context}<\/span><\/span>\\\\n\\\\n\u95ee\u9898：<\/span>{<\/span>question}<\/span><\/span>"<\/span><\/span>)<\/span><\/p>\n

                  4.2 Milvus\u9002\u7528\u573a\u666f<\/h4>\n

                  \u63a8\u8350\u573a\u666f：<\/p>\n

                    \n
                  • \u751f\u4ea7\u7ea7\u5927\u89c4\u6a21RAG\u7cfb\u7edf（\u767e\u4e07\u81f3\u4ebf\u7ea7\u6587\u6863）<\/li>\n
                  • \u5bf9\u68c0\u7d22\u5ef6\u8fdf\u6709\u4e25\u683c\u8981\u6c42\u7684\u5b9e\u65f6\u5e94\u7528<\/li>\n
                  • \u9700\u8981\u6c34\u5e73\u6269\u5c55\u80fd\u529b\u7684\u4f01\u4e1a\u7ea7\u90e8\u7f72<\/li>\n
                  • \u591a\u79df\u6237SaaS\u5e73\u53f0\u7684\u5411\u91cf\u68c0\u7d22\u670d\u52a1<\/li>\n
                  • \u4e0eLangChain\u3001LlamaIndex\u7b49\u6846\u67b6\u6df1\u5ea6\u96c6\u6210\u7684\u9879\u76ee<\/li>\n<\/ul>\n

                    \u5178\u578b\u7528\u4f8b：<\/p>\n

                    # \u4f01\u4e1a\u7ea7RAG\u68c0\u7d22\u670d\u52a1<\/span>
                    \nfrom<\/span> pymilvus import<\/span> Collection,<\/span> utility<\/p>\n

                    class<\/span> ProductionRAGRetriever<\/span>:<\/span>
                    \n def<\/span> __init__<\/span>(<\/span>self,<\/span> collection_name:<\/span> str<\/span>)<\/span>:<\/span>
                    \n self.<\/span>collection =<\/span> Collection(<\/span>collection_name)<\/span>
                    \n self.<\/span>collection.<\/span>load(<\/span>)<\/span><\/p>\n

                    def<\/span> retrieve<\/span>(<\/span>self,<\/span> query_embedding,<\/span> filters:<\/span> dict<\/span>,<\/span> top_k:<\/span> int<\/span> =<\/span> 10<\/span>)<\/span>:<\/span>
                    \n expr =<\/span> self.<\/span>_build_filter_expr(<\/span>filters)<\/span><\/p>\n

                    results =<\/span> self.<\/span>collection.<\/span>search(<\/span>
                    \n data=<\/span>[<\/span>query_embedding]<\/span>,<\/span>
                    \n anns_field=<\/span>"embedding"<\/span>,<\/span>
                    \n param=<\/span>{<\/span>
                    \n "metric_type"<\/span>:<\/span> "COSINE"<\/span>,<\/span>
                    \n "params"<\/span>:<\/span> {<\/span>"nprobe"<\/span>:<\/span> 16<\/span>,<\/span> "ef"<\/span>:<\/span> 64<\/span>}<\/span>
                    \n }<\/span>,<\/span>
                    \n limit=<\/span>top_k,<\/span>
                    \n expr=<\/span>expr,<\/span>
                    \n output_fields=<\/span>[<\/span>"doc_id"<\/span>,<\/span> "content"<\/span>,<\/span> "metadata"<\/span>]<\/span>
                    \n )<\/span><\/p>\n

                    return<\/span> self.<\/span>_format_results(<\/span>results)<\/span><\/p>\n

                    def<\/span> _build_filter_expr<\/span>(<\/span>self,<\/span> filters:<\/span> dict<\/span>)<\/span> –<\/span>><\/span> str<\/span>:<\/span>
                    \n conditions =<\/span> [<\/span>]<\/span>
                    \n if<\/span> "tenant_id"<\/span> in<\/span> filters:<\/span>
                    \n conditions.<\/span>append(<\/span>f"tenant_id == '<\/span>{<\/span>filters[<\/span>'tenant_id'<\/span>]<\/span>}<\/span><\/span>'"<\/span><\/span>)<\/span>
                    \n if<\/span> "category"<\/span> in<\/span> filters:<\/span>
                    \n conditions.<\/span>append(<\/span>f"category in <\/span>{<\/span>filters[<\/span>'category'<\/span>]<\/span>}<\/span><\/span>"<\/span><\/span>)<\/span>
                    \n return<\/span> " AND "<\/span>.<\/span>join(<\/span>conditions)<\/span> if<\/span> conditions else<\/span> ""<\/span><\/p>\n

                    4.3 SurrealDB\u9002\u7528\u573a\u666f<\/h4>\n

                    \u63a8\u8350\u573a\u666f：<\/p>\n

                      \n
                    • \u77e5\u8bc6\u56fe\u8c31\u589e\u5f3a\u7684RAG\u5e94\u7528<\/li>\n
                    • \u9700\u8981\u56fe\u5173\u7cfb\u63a8\u7406\u7684\u590d\u6742\u68c0\u7d22\u573a\u666f<\/li>\n
                    • \u5168\u6808\u5e94\u7528，\u5e0c\u671b\u7b80\u5316\u6280\u672f\u6808<\/li>\n
                    • \u5b9e\u65f6\u534f\u4f5c\u5e94\u7528，\u9700\u8981\u6570\u636e\u53d8\u66f4\u8ba2\u9605<\/li>\n
                    • \u4e2d\u5c0f\u89c4\u6a21\u5e94\u7528，\u8ffd\u6c42\u5f00\u53d1\u6548\u7387<\/li>\n<\/ul>\n

                      \u5178\u578b\u7528\u4f8b：<\/p>\n

                      — \u77e5\u8bc6\u56fe\u8c31\u589e\u5f3aRAG\u67e5\u8be2<\/span>
                      \n— \u68c0\u7d22\u76f8\u5173\u6587\u6863\u5e76\u83b7\u53d6\u5173\u8054\u5b9e\u4f53\u4fe1\u606f<\/span>
                      \nSELECT<\/span>
                      \n doc.<\/span>id,<\/span>
                      \n doc.<\/span>content,<\/span>
                      \n vector::similarity::cosine(<\/span>doc.<\/span>embedding,<\/span> $query_vec)<\/span> AS<\/span> relevance,<\/span>
                      \n (<\/span>
                      \n SELECT<\/span> VALUE<\/span> entity.<\/span>name
                      \n FROM<\/span> doc–<\/span>><\/span>mentions–<\/span>><\/span>entity
                      \n WHERE<\/span> entity.<\/span>type<\/span> =<\/span> 'Concept'<\/span>
                      \n )<\/span> AS<\/span> related_concepts,<\/span>
                      \n (<\/span>
                      \n SELECT<\/span> VALUE<\/span> ref.<\/span>title
                      \n FROM<\/span> doc–<\/span>><\/span>references<\/span>–<\/span>><\/span>ref
                      \n )<\/span> AS<\/span> references<\/span>
                      \nFROM<\/span> document AS<\/span> doc
                      \nWHERE<\/span> relevance ><\/span> 0.6<\/span>
                      \nORDER<\/span> BY<\/span> relevance DESC<\/span>
                      \nLIMIT<\/span> 10<\/span>;<\/span><\/p>\n

                      5. \u6df7\u5408\u67b6\u6784\u65b9\u6848<\/h3>\n

                      \u5728\u5b9e\u9645\u751f\u4ea7\u73af\u5883\u4e2d，\u5355\u4e00\u6570\u636e\u5e93\u5f80\u5f80\u96be\u4ee5\u6ee1\u8db3RAG\u7cfb\u7edf\u7684\u5168\u90e8\u9700\u6c42\u3002\u6df7\u5408\u67b6\u6784\u65b9\u6848\u53ef\u5145\u5206\u53d1\u6325\u5404\u6570\u636e\u5e93\u7684\u4f18\u52bf\u3002<\/p>\n

                      5.1 DuckDB + Milvus\u6df7\u5408\u67b6\u6784<\/h4>\n

                      \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510
                      \n\u2502 RAG Application \u2502
                      \n\u251c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524
                      \n\u2502 \u2502
                      \n\u2502 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510 \u2502
                      \n\u2502 \u2502 DuckDB \u2502 \u2502 Milvus \u2502 \u2502 LLM \u2502 \u2502
                      \n\u2502 \u2502 \u2502 \u2502 \u2502 \u2502 \u2502 \u2502
                      \n\u2502 \u2502 \u2022 \u6570\u636e\u9884\u5904\u7406 \u2502\u2500\u2500\u2500\u25b6\u2502 \u2022 \u5411\u91cf\u5b58\u50a8 \u2502\u2500\u2500\u2500\u25b6\u2502 \u2022 \u751f\u6210\u56de\u7b54 \u2502 \u2502
                      \n\u2502 \u2502 \u2022 \u5143\u6570\u636e\u7ba1\u7406 \u2502 \u2502 \u2022 \u8bed\u4e49\u68c0\u7d22 \u2502 \u2502 \u2502 \u2502
                      \n\u2502 \u2502 \u2022 \u5206\u6790\u62a5\u8868 \u2502\u25c0\u2500\u2500\u2500\u2502 \u2022 ANN\u7d22\u5f15 \u2502 \u2502 \u2502 \u2502
                      \n\u2502 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518 \u2502
                      \n\u2502 \u2502
                      \n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518<\/p>\n

                      \u804c\u8d23\u5206\u5de5：<\/p>\n

                        \n
                      • DuckDB：\u6587\u6863\u9884\u5904\u7406\u3001\u7279\u5f81\u5de5\u7a0b\u3001\u5143\u6570\u636e\u7ba1\u7406\u3001\u5206\u6790\u62a5\u8868<\/li>\n
                      • Milvus：\u5411\u91cf\u5b58\u50a8\u3001\u8bed\u4e49\u68c0\u7d22\u3001\u76f8\u4f3c\u5ea6\u8ba1\u7b97<\/li>\n<\/ul>\n

                        5.2 \u5b9e\u73b0\u793a\u4f8b<\/h4>\n

                        import<\/span> duckdb
                        \nfrom<\/span> pymilvus import<\/span> Collection<\/p>\n

                        class<\/span> HybridRAGSystem<\/span>:<\/span>
                        \n def<\/span> __init__<\/span>(<\/span>self)<\/span>:<\/span>
                        \n self.<\/span>duckdb_conn =<\/span> duckdb.<\/span>connect(<\/span>'metadata.duckdb'<\/span>)<\/span>
                        \n self.<\/span>milvus_collection =<\/span> Collection(<\/span>"documents"<\/span>)<\/span><\/p>\n

                        def<\/span> ingest_documents<\/span>(<\/span>self,<\/span> documents:<\/span> list<\/span>)<\/span>:<\/span>
                        \n # DuckDB\u5904\u7406\u5143\u6570\u636e<\/span>
                        \n metadata_df =<\/span> self.<\/span>_extract_metadata(<\/span>documents)<\/span>
                        \n self.<\/span>duckdb_conn.<\/span>execute(<\/span>
                        \n "INSERT INTO document_metadata SELECT * FROM metadata_df"<\/span>
                        \n )<\/span><\/p>\n

                        # Milvus\u5b58\u50a8\u5411\u91cf<\/span>
                        \n vectors =<\/span> self.<\/span>_generate_embeddings(<\/span>documents)<\/span>
                        \n self.<\/span>milvus_collection.<\/span>insert(<\/span>[<\/span>
                        \n [<\/span>doc[<\/span>'id'<\/span>]<\/span> for<\/span> doc in<\/span> documents]<\/span>,<\/span>
                        \n vectors,<\/span>
                        \n [<\/span>doc[<\/span>'content'<\/span>]<\/span> for<\/span> doc in<\/span> documents]<\/span>
                        \n ]<\/span>)<\/span><\/p>\n

                        def<\/span> hybrid_search<\/span>(<\/span>self,<\/span> query:<\/span> str<\/span>,<\/span> filters:<\/span> dict<\/span>)<\/span>:<\/span>
                        \n # Step 1: DuckDB\u9884\u8fc7\u6ee4\u83b7\u53d6\u5019\u9009\u6587\u6863ID<\/span>
                        \n candidate_ids =<\/span> self.<\/span>duckdb_conn.<\/span>execute(<\/span>"""
                        \n SELECT doc_id FROM document_metadata
                        \n WHERE category = ? AND publish_date > ?
                        \n """<\/span>,<\/span> [<\/span>filters[<\/span>'category'<\/span>]<\/span>,<\/span> filters[<\/span>'date'<\/span>]<\/span>]<\/span>)<\/span>.<\/span>fetchall(<\/span>)<\/span><\/p>\n

                        # Step 2: Milvus\u5728\u5019\u9009\u96c6\u4e2d\u8fdb\u884c\u5411\u91cf\u68c0\u7d22<\/span>
                        \n query_embedding =<\/span> self.<\/span>_embed(<\/span>query)<\/span>
                        \n id_filter =<\/span> f"doc_id in <\/span>{<\/span>[<\/span>id<\/span>[<\/span>0<\/span>]<\/span> for<\/span> id<\/span> in<\/span> candidate_ids]<\/span>}<\/span><\/span>"<\/span><\/span><\/p>\n

                        results =<\/span> self.<\/span>milvus_collection.<\/span>search(<\/span>
                        \n data=<\/span>[<\/span>query_embedding]<\/span>,<\/span>
                        \n anns_field=<\/span>"embedding"<\/span>,<\/span>
                        \n param=<\/span>{<\/span>"metric_type"<\/span>:<\/span> "COSINE"<\/span>}<\/span>,<\/span>
                        \n limit=<\/span>10<\/span>,<\/span>
                        \n expr=<\/span>id_filter
                        \n )<\/span><\/p>\n

                        # Step 3: DuckDB\u8865\u5145\u5b8c\u6574\u5143\u6570\u636e<\/span>
                        \n doc_ids =<\/span> [<\/span>hit.<\/span>id<\/span> for<\/span> hit in<\/span> results[<\/span>0<\/span>]<\/span>]<\/span>
                        \n full_results =<\/span> self.<\/span>duckdb_conn.<\/span>execute(<\/span>"""
                        \n SELECT * FROM document_metadata
                        \n WHERE doc_id IN (SELECT UNNEST(?))
                        \n """<\/span>,<\/span> [<\/span>doc_ids]<\/span>)<\/span>.<\/span>fetchdf(<\/span>)<\/span><\/p>\n

                        return<\/span> full_results<\/p>\n

                        6. \u9009\u578b\u51b3\u7b56\u6846\u67b6<\/h3>\n

                        6.1 \u6838\u5fc3\u80fd\u529b\u5bf9\u6bd4<\/h4>\n\n\u80fd\u529b\u7ef4\u5ea6DuckDBMilvusSurrealDB<\/tr>\n\n\n\n\n\n\n\n
                        \u5411\u91cf\u68c0\u7d22<\/td>\n\u66b4\u529b\u641c\u7d22，\u65e0ANN\u7d22\u5f15<\/td>\n\u591a\u79cdANN\u7d22\u5f15，\u4ebf\u7ea7\u652f\u6301<\/td>\n\u57fa\u7840\u652f\u6301<\/td>\n<\/tr>\n
                        \u90e8\u7f72\u590d\u6742\u5ea6<\/td>\n\u96f6\u4f9d\u8d56，\u5d4c\u5165\u5f0f<\/td>\n\u9700etcd\/MinIO\u7b49\u7ec4\u4ef6<\/td>\n\u5355\u4e8c\u8fdb\u5236，\u8f83\u7b80\u5355<\/td>\n<\/tr>\n
                        \u6c34\u5e73\u6269\u5c55<\/td>\n\u4e0d\u652f\u6301<\/td>\n\u539f\u751f\u5206\u5e03\u5f0f<\/td>\n\u6709\u9650\u652f\u6301<\/td>\n<\/tr>\n
                        SQL\u80fd\u529b<\/td>\n\u5b8c\u6574SQL\u652f\u6301<\/td>\n\u4ec5\u6807\u91cf\u8fc7\u6ee4<\/td>\nSurrealQL，\u8f83\u5b8c\u6574<\/td>\n<\/tr>\n
                        \u56fe\u67e5\u8be2<\/td>\n\u4e0d\u652f\u6301<\/td>\n\u4e0d\u652f\u6301<\/td>\n\u539f\u751f\u652f\u6301<\/td>\n<\/tr>\n
                        \u751f\u6001\u6210\u719f\u5ea6<\/td>\n\u6210\u719f，\u793e\u533a\u6d3b\u8dc3<\/td>\n\u6210\u719f，LangChain\u7b49\u6df1\u5ea6\u96c6\u6210<\/td>\n\u8f83\u65b0，\u751f\u6001\u5efa\u8bbe\u4e2d<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

                        6.2 \u5feb\u901f\u51b3\u7b56\u6307\u5357<\/h4>\n

                        \u5f00\u59cb
                        \n \u2502
                        \n \u25bc
                        \n\u6587\u6863\u89c4\u6a21\u662f\u5426\u8d85\u8fc7100\u4e07？
                        \n \u2502
                        \n \u251c\u2500 \u662f \u2500\u2500\u25b6 \u9009\u62e9 Milvus
                        \n \u2502
                        \n \u25bc \u5426
                        \n\u662f\u5426\u9700\u8981\u56fe\u5173\u7cfb\u67e5\u8be2？
                        \n \u2502
                        \n \u251c\u2500 \u662f \u2500\u2500\u25b6 \u9009\u62e9 SurrealDB
                        \n \u2502
                        \n \u25bc \u5426
                        \n\u662f\u5426\u4e3a\u539f\u578b\/POC\u9636\u6bb5？
                        \n \u2502
                        \n \u251c\u2500 \u662f \u2500\u2500\u25b6 \u9009\u62e9 DuckDB
                        \n \u2502
                        \n \u25bc \u5426
                        \n\u662f\u5426\u9700\u8981\u590d\u6742SQL\u5206\u6790？
                        \n \u2502
                        \n \u251c\u2500 \u662f \u2500\u2500\u25b6 \u9009\u62e9 DuckDB + Milvus\u6df7\u5408\u67b6\u6784
                        \n \u2502
                        \n \u25bc \u5426
                        \n\u9009\u62e9 Milvus（\u751f\u4ea7\u73af\u5883\u9996\u9009）<\/p>\n

                        7. \u603b\u7ed3\u4e0e\u5efa\u8bae<\/h3>\n

                        7.1 \u6838\u5fc3\u7ed3\u8bba<\/h4>\n

                        \u4e09\u6b3e\u6570\u636e\u5e93\u5728RAG\u5e94\u7528\u573a\u666f\u4e2d\u5404\u5177\u7279\u8272：<\/p>\n

                          \n
                        • DuckDB：\u4ee5\u6781\u4f4e\u7684\u90e8\u7f72\u6210\u672c\u548c\u5f3a\u5927\u7684SQL\u80fd\u529b\u89c1\u957f，\u9002\u5408\u539f\u578b\u5f00\u53d1\u548c\u5c0f\u89c4\u6a21\u5e94\u7528，\u4f46\u5411\u91cf\u68c0\u7d22\u80fd\u529b\u6709\u9650<\/li>\n
                        • Milvus：\u4e13\u4e1a\u5411\u91cf\u6570\u636e\u5e93\u7684\u4ee3\u8868，\u5728\u5927\u89c4\u6a21\u5411\u91cf\u68c0\u7d22\u573a\u666f\u4e0b\u6027\u80fd\u5353\u8d8a，\u662f\u751f\u4ea7\u7ea7RAG\u7cfb\u7edf\u7684\u9996\u9009<\/li>\n
                        • SurrealDB：\u591a\u6a21\u578b\u7edf\u4e00\u7684\u521b\u65b0\u8bbe\u8ba1，\u5728\u77e5\u8bc6\u56fe\u8c31\u589e\u5f3aRAG\u573a\u666f\u4e2d\u5177\u6709\u72ec\u7279\u4f18\u52bf<\/li>\n<\/ul>\n

                          7.2 \u5b9e\u8df5\u5efa\u8bae<\/h4>\n
                        • \u539f\u578b\u9636\u6bb5：\u4f18\u5148\u9009\u62e9DuckDB，\u5feb\u901f\u9a8c\u8bc1\u4e1a\u52a1\u903b\u8f91<\/li>\n
                        • \u751f\u4ea7\u90e8\u7f72：\u6839\u636e\u89c4\u6a21\u9009\u62e9Milvus\u6216\u6df7\u5408\u67b6\u6784<\/li>\n
                        • \u77e5\u8bc6\u56fe\u8c31\u573a\u666f：\u8003\u8651SurrealDB\u6216Neo4j + \u5411\u91cf\u63d2\u4ef6\u65b9\u6848<\/li>\n
                        • \u6301\u7eed\u6f14\u8fdb：\u5efa\u7acb\u62bd\u8c61\u5c42，\u4fdd\u6301\u6570\u636e\u5e93\u5207\u6362\u7684\u7075\u6d3b\u6027<\/li>\n

                          7.3 \u672a\u6765\u5c55\u671b<\/h4>\n

                          \u968f\u7740RAG\u6280\u672f\u7684\u6301\u7eed\u6f14\u8fdb，\u6570\u636e\u5e93\u9886\u57df\u4e5f\u5728\u5feb\u901f\u53d1\u5c55：<\/p>\n

                            \n
                          • DuckDB\u6b63\u5728\u589e\u5f3a\u5411\u91cf\u68c0\u7d22\u80fd\u529b，\u672a\u6765\u7248\u672c\u53ef\u80fd\u5f15\u5165ANN\u7d22\u5f15\u652f\u6301<\/li>\n
                          • Milvus\u6301\u7eed\u4f18\u5316\u6df7\u5408\u67e5\u8be2\u80fd\u529b，\u5411\u5168\u529f\u80fd\u6570\u636e\u5e93\u65b9\u5411\u6f14\u8fdb<\/li>\n
                          • SurrealDB\u4f5c\u4e3a\u65b0\u5174\u9879\u76ee，\u591a\u6a21\u578b\u7edf\u4e00\u7684\u7406\u5ff5\u5177\u6709\u957f\u671f\u4ef7\u503c<\/li>\n<\/ul>\n

                            \u6280\u672f\u9009\u578b\u5e94\u7ed3\u5408\u56e2\u961f\u80fd\u529b\u3001\u4e1a\u52a1\u9700\u6c42\u548c\u957f\u671f\u89c4\u5212\u7efc\u5408\u8003\u91cf，\u907f\u514d\u8fc7\u5ea6\u8bbe\u8ba1，\u4e5f\u8981\u4e3a\u672a\u6765\u6269\u5c55\u9884\u7559\u7a7a\u95f4\u3002<\/p>\n

                            \n

                            \u53c2\u8003\u8d44\u6e90：<\/p>\n

                              \n
                            • DuckDB\u5b98\u65b9\u6587\u6863：https:\/\/duckdb.org\/docs\/<\/li>\n
                            • Milvus\u5b98\u65b9\u6587\u6863：https:\/\/milvus.io\/docs\/<\/li>\n
                            • SurrealDB\u5b98\u65b9\u6587\u6863：https:\/\/surrealdb.com\/docs\/<\/li>\n
                            • LangChain\u5411\u91cf\u5b58\u50a8\u96c6\u6210：https:\/\/python.langchain.com\/docs\/integrations\/vectorstores\/<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"

                              \u6458\u8981
                              \n\u68c0\u7d22\u589e\u5f3a\u751f\u6210（Retrieval-Augmented Generation，RAG）\u6280\u672f\u5df2\u6210\u4e3a\u5927\u8bed\u8a00\u6a21\u578b\u5e94\u7528\u843d\u5730\u7684\u6838\u5fc3\u67b6\u6784\u6a21\u5f0f\u3002\u5728RAG\u7cfb\u7edf\u4e2d，\u5411\u91cf\u6570\u636e\u5e93\u627f\u62c5\u7740\u8bed\u4e49\u68c0\u7d22\u7684\u5173\u952e\u804c\u8d23，\u5176\u9009\u578b\u76f4\u63a5\u5f71\u54cd\u7cfb\u7edf\u7684\u68c0\u7d22\u7cbe\u5ea6\u3001\u54cd\u5e94\u5ef6\u8fdf\u548c\u6269\u5c55\u80fd\u529b\u3002\u672c\u6587\u5c06\u4ece\u6280\u672f\u67b6\u6784\u3001\u6027\u80fd\u7279\u5f81\u3001\u9002\u7528\u573a\u666f\u7b49\u7ef4\u5ea6，\u5bf9DuckDB\u3001Milvus\u548cSurrealDB\u4e09\u6b3e\u6570\u636e\u5e93\u5728RAG\u5e94\u7528\u4e2d\u7684\u8868\u73b0\u8fdb\u884c\u7cfb\u7edf\u6027\u5bf9\u6bd4\u5206\u6790，\u4e3a\u6280<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[4119,50,100],"topic":[],"class_list":["post-78763","post","type-post","status-publish","format-standard","hentry","category-server","tag-rag","tag-50","tag-100"],"yoast_head":"\nRAG\u5e94\u7528\u6570\u636e\u5e93\u9009\u578b\uff1aDuckDB\u3001Milvus\u4e0eSurrealDB\u6df1\u5ea6\u5bf9\u6bd4\u5206\u6790 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/78763.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"RAG\u5e94\u7528\u6570\u636e\u5e93\u9009\u578b\uff1aDuckDB\u3001Milvus\u4e0eSurrealDB\u6df1\u5ea6\u5bf9\u6bd4\u5206\u6790 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u6458\u8981 \u68c0\u7d22\u589e\u5f3a\u751f\u6210（Retrieval-Augmented Generation，RAG）\u6280\u672f\u5df2\u6210\u4e3a\u5927\u8bed\u8a00\u6a21\u578b\u5e94\u7528\u843d\u5730\u7684\u6838\u5fc3\u67b6\u6784\u6a21\u5f0f\u3002\u5728RAG\u7cfb\u7edf\u4e2d，\u5411\u91cf\u6570\u636e\u5e93\u627f\u62c5\u7740\u8bed\u4e49\u68c0\u7d22\u7684\u5173\u952e\u804c\u8d23，\u5176\u9009\u578b\u76f4\u63a5\u5f71\u54cd\u7cfb\u7edf\u7684\u68c0\u7d22\u7cbe\u5ea6\u3001\u54cd\u5e94\u5ef6\u8fdf\u548c\u6269\u5c55\u80fd\u529b\u3002\u672c\u6587\u5c06\u4ece\u6280\u672f\u67b6\u6784\u3001\u6027\u80fd\u7279\u5f81\u3001\u9002\u7528\u573a\u666f\u7b49\u7ef4\u5ea6，\u5bf9DuckDB\u3001Milvus\u548cSurrealDB\u4e09\u6b3e\u6570\u636e\u5e93\u5728RAG\u5e94\u7528\u4e2d\u7684\u8868\u73b0\u8fdb\u884c\u7cfb\u7edf\u6027\u5bf9\u6bd4\u5206\u6790，\u4e3a\u6280\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/78763.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-02-27T23:51:38+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/78763.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/78763.html\",\"name\":\"RAG\u5e94\u7528\u6570\u636e\u5e93\u9009\u578b\uff1aDuckDB\u3001Milvus\u4e0eSurrealDB\u6df1\u5ea6\u5bf9\u6bd4\u5206\u6790 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-02-27T23:51:38+00:00\",\"dateModified\":\"2026-02-27T23:51:38+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/78763.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/78763.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/78763.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"RAG\u5e94\u7528\u6570\u636e\u5e93\u9009\u578b\uff1aDuckDB\u3001Milvus\u4e0eSurrealDB\u6df1\u5ea6\u5bf9\u6bd4\u5206\u6790\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"RAG\u5e94\u7528\u6570\u636e\u5e93\u9009\u578b\uff1aDuckDB\u3001Milvus\u4e0eSurrealDB\u6df1\u5ea6\u5bf9\u6bd4\u5206\u6790 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/78763.html","og_locale":"zh_CN","og_type":"article","og_title":"RAG\u5e94\u7528\u6570\u636e\u5e93\u9009\u578b\uff1aDuckDB\u3001Milvus\u4e0eSurrealDB\u6df1\u5ea6\u5bf9\u6bd4\u5206\u6790 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u6458\u8981 \u68c0\u7d22\u589e\u5f3a\u751f\u6210（Retrieval-Augmented Generation，RAG）\u6280\u672f\u5df2\u6210\u4e3a\u5927\u8bed\u8a00\u6a21\u578b\u5e94\u7528\u843d\u5730\u7684\u6838\u5fc3\u67b6\u6784\u6a21\u5f0f\u3002\u5728RAG\u7cfb\u7edf\u4e2d，\u5411\u91cf\u6570\u636e\u5e93\u627f\u62c5\u7740\u8bed\u4e49\u68c0\u7d22\u7684\u5173\u952e\u804c\u8d23，\u5176\u9009\u578b\u76f4\u63a5\u5f71\u54cd\u7cfb\u7edf\u7684\u68c0\u7d22\u7cbe\u5ea6\u3001\u54cd\u5e94\u5ef6\u8fdf\u548c\u6269\u5c55\u80fd\u529b\u3002\u672c\u6587\u5c06\u4ece\u6280\u672f\u67b6\u6784\u3001\u6027\u80fd\u7279\u5f81\u3001\u9002\u7528\u573a\u666f\u7b49\u7ef4\u5ea6，\u5bf9DuckDB\u3001Milvus\u548cSurrealDB\u4e09\u6b3e\u6570\u636e\u5e93\u5728RAG\u5e94\u7528\u4e2d\u7684\u8868\u73b0\u8fdb\u884c\u7cfb\u7edf\u6027\u5bf9\u6bd4\u5206\u6790，\u4e3a\u6280","og_url":"https:\/\/www.wsisp.com\/helps\/78763.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-02-27T23:51:38+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"6 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/78763.html","url":"https:\/\/www.wsisp.com\/helps\/78763.html","name":"RAG\u5e94\u7528\u6570\u636e\u5e93\u9009\u578b\uff1aDuckDB\u3001Milvus\u4e0eSurrealDB\u6df1\u5ea6\u5bf9\u6bd4\u5206\u6790 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-02-27T23:51:38+00:00","dateModified":"2026-02-27T23:51:38+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/78763.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/78763.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/78763.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"RAG\u5e94\u7528\u6570\u636e\u5e93\u9009\u578b\uff1aDuckDB\u3001Milvus\u4e0eSurrealDB\u6df1\u5ea6\u5bf9\u6bd4\u5206\u6790"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/78763","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=78763"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/78763\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=78763"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=78763"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=78763"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=78763"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}