apiVersion: v1 kind: KnowledgeEngine metadata: name: langrag label: en_US: LangRAG zh_Hans: LangRAG description: en_US: Simple Knowledge Engine with document chunking, vector, full-text, and hybrid retrieval zh_Hans: 简单 RAG 引擎，支持文档分块及向量、全文、混合检索 spec: creation_schema: - name: embedding_model_uuid label: en_US: Embedding Model zh_Hans: 嵌入模型 description: en_US: Select embedding model for text vectorization zh_Hans: 选择用于文本向量化的嵌入模型 type: embedding-model-selector required: true default: '' - name: index_type label: en_US: Index Type zh_Hans: 索引类型 description: en_US: 'Chunk: flat chunking; Parent-Child: two-level chunking for richer context; QA: LLM-generated question-answer pairs for better query matching' zh_Hans: 'Chunk：平坦分块；Parent-Child：双层分块，检索返回更丰富上下文；QA：LLM 生成问答对，提升查询匹配质量' type: select required: false default: chunk options: - name: chunk label: en_US: Chunk zh_Hans: 分块 - name: parent_child label: en_US: Parent-Child zh_Hans: 父子索引 - name: qa label: en_US: QA Index zh_Hans: 问答索引 - name: chunk_size label: en_US: Chunk Size zh_Hans: 分块大小 type: integer required: false default: 512 - name: overlap label: en_US: Chunk Overlap zh_Hans: 分块重叠 type: integer required: false default: 50 - name: parent_chunk_size label: en_US: Parent Chunk Size zh_Hans: 父块大小 description: en_US: Size of parent chunks (only used with Parent-Child index type) zh_Hans: 父块大小（仅在 Parent-Child 索引类型下生效） type: integer required: false default: 2048 show_if: field: index_type operator: eq value: parent_child - name: child_chunk_size label: en_US: Child Chunk Size zh_Hans: 子块大小 description: en_US: Size of child chunks (only used with Parent-Child index type) zh_Hans: 子块大小（仅在 Parent-Child 索引类型下生效） type: integer required: false default: 256 show_if: field: index_type operator: eq value: parent_child - name: qa_llm_model_uuid label: en_US: QA Generation LLM zh_Hans: QA 生成用 LLM description: en_US: LLM model for generating Q&A pairs (required when index type is QA) zh_Hans: 用于生成问答对的 LLM 模型（索引类型为 QA 时需要） type: llm-model-selector required: false default: '' show_if: field: index_type operator: eq value: qa - name: questions_per_chunk label: en_US: Questions Per Chunk zh_Hans: 每块生成问题数 description: en_US: Number of Q&A pairs to generate per chunk (only used with QA index type) zh_Hans: 每个文本块生成的问答对数量（仅在 QA 索引类型下生效） type: integer required: false default: 1 show_if: field: index_type operator: eq value: qa retrieval_schema: - name: top_k label: en_US: Top K zh_Hans: 召回数量 type: integer required: false default: 5 - name: search_type label: en_US: Search Type zh_Hans: 检索类型 description: en_US: 'Vector: semantic similarity; Full-text: keyword matching; Hybrid: combined' zh_Hans: 'Vector：语义相似度；Full-text：关键词匹配；Hybrid：混合检索' type: select required: false default: vector options: - name: vector label: en_US: Vector zh_Hans: 向量检索 - name: full_text label: en_US: Full-text zh_Hans: 全文检索 - name: hybrid label: en_US: Hybrid zh_Hans: 混合检索 - name: vector_weight label: en_US: Vector Search Weight zh_Hans: 向量搜索权重 description: en_US: 'Balance between vector and keyword search in hybrid mode (0.0 = keyword only, 1.0 = vector only)' zh_Hans: '混合检索中向量搜索的权重（0.0 = 仅关键词，1.0 = 仅向量）' type: float required: false default: 0.7 show_if: field: search_type operator: eq value: hybrid - name: query_rewrite label: en_US: Query Rewriting zh_Hans: 查询重写 description: en_US: 'Off: direct search; HyDE: hypothetical document; Multi-Query: query expansion; Step-Back: abstract generalization' zh_Hans: 'Off：直接检索；HyDE：假设文档嵌入；Multi-Query：多查询扩展；Step-Back：回退抽象提问' type: select required: false default: "off" options: - name: "off" label: { en_US: "Off", zh_Hans: "关闭" } - name: hyde label: { en_US: "HyDE", zh_Hans: "HyDE（假设文档）" } - name: multi_query label: { en_US: "Multi-Query", zh_Hans: "多查询扩展" } - name: step_back label: { en_US: "Step-Back", zh_Hans: "回退提问" } - name: rewrite_llm_model_uuid label: en_US: Query Rewrite LLM zh_Hans: 查询重写 LLM description: en_US: LLM model for query rewriting (required when query rewriting is enabled) zh_Hans: 用于查询重写的 LLM 模型（启用查询重写时需要） type: llm-model-selector required: false default: '' show_if: field: query_rewrite operator: neq value: "off" - name: rerank label: en_US: Reranking zh_Hans: 重排序 description: en_US: 'Off: no reranking; Rerank Model: use a LangBot rerank model; LLM: use a language model to rerank candidates by relevance' zh_Hans: 'Off：不重排序；Rerank Model：使用 LangBot 重排序模型；LLM：使用大语言模型按相关性重排候选结果' type: select required: false default: "off" options: - name: "off" label: { en_US: "Off", zh_Hans: "关闭" } - name: "rerank_model" label: { en_US: "Rerank Model", zh_Hans: "重排序模型" } - name: "llm" label: { en_US: "LLM", zh_Hans: "LLM 重排序" } - name: rerank_model_uuid label: en_US: Rerank Model zh_Hans: 重排序模型 description: en_US: LangBot rerank model for reranking retrieved candidates zh_Hans: 用于对检索候选结果重排序的 LangBot 重排序模型 type: rerank-model-selector required: false default: '' show_if: field: rerank operator: eq value: "rerank_model" - name: rerank_llm_model_uuid label: en_US: Rerank LLM zh_Hans: 重排序 LLM description: en_US: LLM model for reranking retrieved candidates zh_Hans: 用于对检索候选结果重排序的 LLM 模型 type: llm-model-selector required: false default: '' show_if: field: rerank operator: eq value: "llm" - name: context_window label: en_US: Context Window zh_Hans: 上下文窗口 description: en_US: 'Number of adjacent chunks to include as context for each hit (0 = disabled). Requires vector store ID-based lookup support.' zh_Hans: '为每个命中 chunk 附加的相邻 chunk 数量（0 = 禁用）。需要向量库支持按 ID 查询。' type: integer required: false default: 0 execution: python: path: langrag.py attr: LangRAG