LangBot/plugins/KnowledgeEngine/LangRAG/components/knowledge_engine/langrag.yaml

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