after_sales_chatflow.yaml

version: 0.6.0
kind: app
app:
  name: 售后咨询分流
  mode: advanced-chat
  icon: 🤖
  icon_type: emoji
  icon_background: '#FFEAD5'
  description: 模拟电商售后问答与分流的 chatflow
  use_icon_as_answer_icon: false
workflow:
  graph:
    nodes:
    - id: start_node
      type: custom
      data:
        type: start
        title: 用户售后问题
        desc: ''
        selected: false
        variables:
        - variable: query
          label: 请输入售后问题
          type: paragraph
          required: true
          options: []
          value_selector: []
      position:
        x: 30.0
        y: 245.0
      positionAbsolute:
        x: 30.0
        y: 245.0
      width: 244
      height: 90
      selected: false
      sourcePosition: right
      targetPosition: left
    - id: llm_node
      type: custom
      data:
        type: llm
        title: 复杂售后转人工
        desc: ''
        selected: false
        model:
          provider: openai
          name: gpt-4o-mini
          mode: chat
          completion_params:
            temperature: 0.2
        prompt_template:
        - role: system
          text: 你是电商平台高级售后协调专员。用于处理发票、保修、价格保护、投诉升级和无法明确归类的复杂售后问题。请用中文回复：先概括用户问题；再说明需要补充的关键信息；然后给出转人工或升级处理建议，并说明会由对应专员继续跟进；不要虚构已完成的处理动作。回复控制在
            4 到 6 行。
        - role: user
          text: '{{#start_node.query#}}'
        vision:
          enabled: false
          configs:
            variable_selector: []
        memory:
          enabled: true
          window:
            enabled: true
            size: 10
        context:
          enabled: false
          variable_selector: []
        structured_output:
          enabled: false
        retry_config:
          enabled: false
          max_retries: 1
          retry_interval: 1000
          exponential_backoff:
            enabled: false
            multiplier: 2
            max_interval: 10000
      position:
        x: 718.0
        y: 425.0
      positionAbsolute:
        x: 718.0
        y: 425.0
      width: 244
      height: 90
      selected: false
      sourcePosition: right
      targetPosition: left
    - id: answer_node
      type: custom
      data:
        type: answer
        title: 复杂售后答复
        desc: ''
        selected: false
        variables: []
        answer: '{{#llm_node.text#}}'
      position:
        x: 1062.0
        y: 425.0
      positionAbsolute:
        x: 1062.0
        y: 425.0
      width: 244
      height: 90
      selected: false
      sourcePosition: right
      targetPosition: left
    - id: aftersales_classifier
      type: custom
      data:
        type: question-classifier
        title: 售后问题分类
        desc: ''
        selected: false
        model:
          provider: openai
          name: gpt-4o-mini
          mode: chat
          completion_params:
            temperature: 0.1
        vision:
          enabled: false
          configs:
            variable_selector: []
        query_variable_selector:
        - start_node
        - query
        classes:
        - id: refund_return
          name: 退款退货
        - id: logistics
          name: 物流催件
        - id: quality_issue
          name: 质量问题
        - id: manual_service
          name: 人工处理
        instruction: 你是售后分流分类器。只能从 4 个类别中选择 1 个。退款退货：退款、退货、拒收、仅退款、七天无理由。物流催件：未发货、发货慢、物流停滞、签收异常、催派送。质量问题：商品破损、少件、错发、功能故障、与描述不符。人工处理：发票、保修、价格保护、补偿争议、投诉升级、无法明确归类的问题。优先按用户核心诉求分类。
      position:
        x: 374.0
        y: 245.0
      positionAbsolute:
        x: 374.0
        y: 245.0
      width: 244
      height: 90
      selected: false
      sourcePosition: right
      targetPosition: left
    - id: refund_llm
      type: custom
      data:
        type: llm
        title: 退款退货专员
        desc: ''
        selected: false
        model:
          provider: openai
          name: gpt-4o-mini
          mode: chat
          completion_params:
            temperature: 0.3
        prompt_template:
        - role: system
          text: 你是电商平台售后退款与退货专员。请用中文回复，并遵守以下要求：先用一句话确认用户诉求；再列出处理所需信息，至少包含订单号、商品名、是否签收、问题描述；然后给出下一步建议，例如未发货可尝试拦截退款、已签收通常走退货退款、质量问题可补充凭证后优先处理；如果信息不足，不要编造处理结果，要明确提示用户补充。回复控制在
            4 到 6 行，专业且清晰。
        - role: user
          text: '{{#start_node.query#}}'
        vision:
          enabled: false
          configs:
            variable_selector: []
        memory:
          enabled: true
          window:
            enabled: true
            size: 10
        context:
          enabled: false
          variable_selector: []
        structured_output:
          enabled: false
        retry_config:
          enabled: false
          max_retries: 1
          retry_interval: 1000
          exponential_backoff:
            enabled: false
            multiplier: 2
            max_interval: 10000
      position:
        x: 718.0
        y: 65.0
      positionAbsolute:
        x: 718.0
        y: 65.0
      width: 244
      height: 90
      selected: false
      sourcePosition: right
      targetPosition: left
    - id: refund_answer
      type: custom
      data:
        type: answer
        title: 退款退货答复
        desc: ''
        selected: false
        variables: []
        answer: '{{#refund_llm.text#}}'
      position:
        x: 1062.0
        y: 65.0
      positionAbsolute:
        x: 1062.0
        y: 65.0
      width: 244
      height: 90
      selected: false
      sourcePosition: right
      targetPosition: left
    - id: logistics_llm
      type: custom
      data:
        type: llm
        title: 物流催件专员
        desc: ''
        selected: false
        model:
          provider: openai
          name: gpt-4o-mini
          mode: chat
          completion_params:
            temperature: 0.3
        prompt_template:
        - role: system
          text: 你是电商平台物流售后专员。请用中文回复，并遵守以下要求：先确认用户的物流诉求；再让用户补充订单号、最新物流节点、收货城市；然后给出可执行建议，例如未出库时可催仓、已出库未更新时可发起催件、签收异常时可核对代收或驿站；如果已经超出承诺时效，可以说明会升级物流排查，但不要虚构具体物流结果。回复控制在
            4 到 6 行，语气专业。
        - role: user
          text: '{{#start_node.query#}}'
        vision:
          enabled: false
          configs:
            variable_selector: []
        memory:
          enabled: true
          window:
            enabled: true
            size: 10
        context:
          enabled: false
          variable_selector: []
        structured_output:
          enabled: false
        retry_config:
          enabled: false
          max_retries: 1
          retry_interval: 1000
          exponential_backoff:
            enabled: false
            multiplier: 2
            max_interval: 10000
      position:
        x: 718.0
        y: 185.0
      positionAbsolute:
        x: 718.0
        y: 185.0
      width: 244
      height: 90
      selected: false
      sourcePosition: right
      targetPosition: left
    - id: logistics_answer
      type: custom
      data:
        type: answer
        title: 物流催件答复
        desc: ''
        selected: false
        variables: []
        answer: '{{#logistics_llm.text#}}'
      position:
        x: 1062.0
        y: 185.0
      positionAbsolute:
        x: 1062.0
        y: 185.0
      width: 244
      height: 90
      selected: false
      sourcePosition: right
      targetPosition: left
    - id: quality_llm
      type: custom
      data:
        type: llm
        title: 质量问题专员
        desc: ''
        selected: false
        model:
          provider: openai
          name: gpt-4o-mini
          mode: chat
          completion_params:
            temperature: 0.3
        prompt_template:
        - role: system
          text: 你是电商平台商品质量问题专员。请用中文回复，并遵守以下要求：先确认用户反馈的是破损、少件、错发、故障或与描述不符中的哪一类；然后要求用户补充订单号、商品名、问题照片或视频、签收时间；再给出处理建议，例如补发、换货、退货退款或进一步质检；不要承诺未经核实的赔付结果。回复控制在
            4 到 6 行，结构清晰。
        - role: user
          text: '{{#start_node.query#}}'
        vision:
          enabled: false
          configs:
            variable_selector: []
        memory:
          enabled: true
          window:
            enabled: true
            size: 10
        context:
          enabled: false
          variable_selector: []
        structured_output:
          enabled: false
        retry_config:
          enabled: false
          max_retries: 1
          retry_interval: 1000
          exponential_backoff:
            enabled: false
            multiplier: 2
            max_interval: 10000
      position:
        x: 718.0
        y: 305.0
      positionAbsolute:
        x: 718.0
        y: 305.0
      width: 244
      height: 90
      selected: false
      sourcePosition: right
      targetPosition: left
    - id: quality_answer
      type: custom
      data:
        type: answer
        title: 质量问题答复
        desc: ''
        selected: false
        variables: []
        answer: '{{#quality_llm.text#}}'
      position:
        x: 1062.0
        y: 305.0
      positionAbsolute:
        x: 1062.0
        y: 305.0
      width: 244
      height: 90
      selected: false
      sourcePosition: right
      targetPosition: left
    edges:
    - id: llm_node-source-answer_node-target
      source: llm_node
      target: answer_node
      sourceHandle: source
      targetHandle: target
      type: custom
      zIndex: 0
      data:
        sourceType: llm
        targetType: answer
        isInIteration: false
        isInLoop: false
    - id: start_node-source-aftersales_classifier-target
      source: start_node
      target: aftersales_classifier
      sourceHandle: source
      targetHandle: target
      type: custom
      zIndex: 0
      data:
        sourceType: start
        targetType: question-classifier
        isInIteration: false
        isInLoop: false
    - id: aftersales_classifier-manual_service-llm_node-target
      source: aftersales_classifier
      target: llm_node
      sourceHandle: manual_service
      targetHandle: target
      type: custom
      zIndex: 0
      data:
        sourceType: question-classifier
        targetType: llm
        isInIteration: false
        isInLoop: false
    - id: aftersales_classifier-refund_return-refund_llm-target
      source: aftersales_classifier
      target: refund_llm
      sourceHandle: refund_return
      targetHandle: target
      type: custom
      zIndex: 0
      data:
        sourceType: question-classifier
        targetType: llm
        isInIteration: false
        isInLoop: false
    - id: refund_llm-source-refund_answer-target
      source: refund_llm
      target: refund_answer
      sourceHandle: source
      targetHandle: target
      type: custom
      zIndex: 0
      data:
        sourceType: llm
        targetType: answer
        isInIteration: false
        isInLoop: false
    - id: aftersales_classifier-logistics-logistics_llm-target
      source: aftersales_classifier
      target: logistics_llm
      sourceHandle: logistics
      targetHandle: target
      type: custom
      zIndex: 0
      data:
        sourceType: question-classifier
        targetType: llm
        isInIteration: false
        isInLoop: false
    - id: logistics_llm-source-logistics_answer-target
      source: logistics_llm
      target: logistics_answer
      sourceHandle: source
      targetHandle: target
      type: custom
      zIndex: 0
      data:
        sourceType: llm
        targetType: answer
        isInIteration: false
        isInLoop: false
    - id: aftersales_classifier-quality_issue-quality_llm-target
      source: aftersales_classifier
      target: quality_llm
      sourceHandle: quality_issue
      targetHandle: target
      type: custom
      zIndex: 0
      data:
        sourceType: question-classifier
        targetType: llm
        isInIteration: false
        isInLoop: false
    - id: quality_llm-source-quality_answer-target
      source: quality_llm
      target: quality_answer
      sourceHandle: source
      targetHandle: target
      type: custom
      zIndex: 0
      data:
        sourceType: llm
        targetType: answer
        isInIteration: false
        isInLoop: false
    viewport:
      x: 0.0
      y: 0.0
      zoom: 1.0
  features:
    file_upload:
      enabled: false
    opening_statement: ''
    retriever_resource:
      enabled: false
    sensitive_word_avoidance:
      enabled: false
    speech_to_text:
      enabled: false
    suggested_questions: []
    suggested_questions_after_answer:
      enabled: false
    text_to_speech:
      enabled: false
  environment_variables: []
  conversation_variables: []
  rag_pipeline_variables: []
dependencies: []