feat: Support parameter extraction nodes (#4208)

Author: shaohuzhang1

apps/application/flow/step_node/__init__.py

@@ -21,6 +21,7 @@
 from .loop_node import *
 from .loop_start_node import *
 from .mcp_node import BaseMcpNode
+from .parameter_extraction_node import BaseParameterExtractionNode
 from .question_node import *
 from .reranker_node import *
 from .search_document_node import BaseSearchDocumentNode
@@ -44,7 +45,7 @@
              BaseVideoUnderstandNode,
              BaseIntentNode, BaseLoopNode, BaseLoopStartStepNode,
              BaseLoopContinueNode,
-             BaseLoopBreakNode, BaseVariableSplittingNode]
+             BaseLoopBreakNode, BaseVariableSplittingNode, BaseParameterExtractionNode]
 
 
 def get_node(node_type):

apps/application/flow/step_node/parameter_extraction_node/__init__.py

@@ -0,0 +1,9 @@
+# coding=utf-8
+"""
+    @project: MaxKB
+    @Author：虎虎
+    @file： __init__.py.py
+    @date：2025/10/13 14:56
+    @desc:
+"""
+from .impl import *

apps/application/flow/step_node/parameter_extraction_node/i_parameter_extraction_node.py

@@ -0,0 +1,39 @@
+# coding=utf-8
+
+from typing import Type
+
+from django.utils.translation import gettext_lazy as _
+from rest_framework import serializers
+
+from application.flow.i_step_node import INode, NodeResult
+
+
+class VariableSplittingNodeParamsSerializer(serializers.Serializer):
+    input_variable = serializers.ListField(required=True,
+                                           label=_("input variable"))
+
+    variable_list = serializers.ListField(required=True,
+                                          label=_("Split variables"))
+
+    model_params_setting = serializers.DictField(required=False,
+                                                 label=_("Model parameter settings"))
+
+    model_id = serializers.CharField(required=True, label=_("Model id"))
+
+
+class IParameterExtractionNode(INode):
+    type = 'parameter-extraction-node'
+
+    def get_node_params_serializer_class(self) -> Type[serializers.Serializer]:
+        return VariableSplittingNodeParamsSerializer
+
+    def _run(self):
+        input_variable = self.workflow_manage.get_reference_field(
+            self.node_params_serializer.data.get('input_variable')[0],
+            self.node_params_serializer.data.get('input_variable')[1:])
+        return self.execute(input_variable, self.node_params_serializer.data['variable_list'],
+                            self.node_params_serializer.data['model_params_setting'],
+                            self.node_params_serializer.data['model_id'])
+
+    def execute(self, input_variable, variable_list, model_params_setting, model_id, **kwargs) -> NodeResult:
+        pass

apps/application/flow/step_node/parameter_extraction_node/impl/__init__.py

@@ -0,0 +1,9 @@
+# coding=utf-8
+"""
+    @project: MaxKB
+    @Author：虎虎
+    @file： __init__.py.py
+    @date：2025/10/13 15:01
+    @desc:
+"""
+from .base_parameter_extraction_node import *

apps/application/flow/step_node/parameter_extraction_node/impl/base_parameter_extraction_node.py

@@ -0,0 +1,112 @@
+# coding=utf-8
+"""
+    @project: MaxKB
+    @Author：虎虎
+    @file： base_variable_splitting_node.py
+    @date：2025/10/13 15:02
+    @desc:
+"""
+import json
+import re
+
+from django.db.models import QuerySet
+from langchain_core.messages import HumanMessage
+from langchain_core.prompts import PromptTemplate
+
+from application.flow.i_step_node import NodeResult
+from application.flow.step_node.parameter_extraction_node.i_parameter_extraction_node import IParameterExtractionNode
+from models_provider.models import Model
+from models_provider.tools import get_model_instance_by_model_workspace_id, get_model_credential
+
+prompt = """
+Please strictly process the text according to the following requirements:
+**Task**: 
+Extract specified field information from given text
+
+**Enter text**: 
+{{question}}
+
+**Extract configuration**: 
+{{properties}}
+
+**Rule**:
+- Strictly follow the data and field of Extract configuration
+- If not found, use null value
+- Only return pure JSON without additional text
+- Keep the string format neat
+"""
+
+
+def get_default_model_params_setting(model_id):
+    model = QuerySet(Model).filter(id=model_id).first()
+    credential = get_model_credential(model.provider, model.model_type, model.model_name)
+    model_params_setting = credential.get_model_params_setting_form(
+        model.model_name).get_default_form_data()
+    return model_params_setting
+
+
+def generate_properties(variable_list):
+    return {variable['field']: {'type': variable['parameter_type'], 'description': variable['desc'],
+                                'title': variable['label']} for variable in
+            variable_list}
+
+
+def generate_example(variable_list):
+    return {variable['field']: None for variable in variable_list}
+
+
+def generate_content(input_variable, variable_list):
+    properties = generate_properties(variable_list)
+    prompt_template = PromptTemplate.from_template(prompt, template_format='jinja2')
+    value = prompt_template.format(properties=properties, question=input_variable)
+    return value
+
+
+def json_loads(response, expected_fields):
+    if not response or not isinstance(response, str):
+        return {field: None for field in expected_fields}
+
+    cleaned = response.strip()
+
+    extraction_strategies = [
+        lambda: json.loads(cleaned),
+        lambda: json.loads(re.search(r'```(?:json)?\s*(\{.*?\})\s*```', cleaned, re.DOTALL).group(1)),
+        lambda: json.loads(re.search(r'(\{[\s\S]*\})', cleaned).group(1)),
+    ]
+    for strategy in extraction_strategies:
+        try:
+            result = strategy()
+            return result
+        except:
+            continue
+    return generate_example(expected_fields)
+
+
+class BaseParameterExtractionNode(IParameterExtractionNode):
+
+    def save_context(self, details, workflow_manage):
+        for key, value in details.get('result').items():
+            self.context['key'] = value
+        self.context['result'] = details.get('result')
+
+    def execute(self, input_variable, variable_list, model_params_setting, model_id, **kwargs) -> NodeResult:
+        if model_params_setting is None:
+            model_params_setting = get_default_model_params_setting(model_id)
+        workspace_id = self.workflow_manage.get_body().get('workspace_id')
+        chat_model = get_model_instance_by_model_workspace_id(model_id, workspace_id,
+                                                              **model_params_setting)
+        content = generate_content(input_variable, variable_list)
+        response = chat_model.invoke([HumanMessage(content=content)])
+        result = json_loads(response.content, variable_list)
+        return NodeResult({'result': result, **result}, {})
+
+    def get_details(self, index: int, **kwargs):
+        return {
+            'name': self.node.properties.get('stepName'),
+            "index": index,
+            'run_time': self.context.get('run_time'),
+            'type': self.node.type,
+            'result': self.context.get('result'),
+            'status': self.status,
+            'err_message': self.err_message
+        }

ui/src/assets/workflow/icon_parameter_extraction.svg

@@ -35,6 +35,7 @@ export enum WorkflowType {
   LoopBreakNode = 'loop-break-node',
   VariableSplittingNode = 'variable-splitting-node',
   VideoUnderstandNode = 'video-understand-node',
+  ParameterExtractionNode = 'parameter-extraction-node',
 }
 export enum WorkflowMode {
   // 应用工作流

ui/src/enums/application.ts

@@ -429,6 +429,20 @@ You are a master of problem optimization, adept at accurately inferring user int
         placeholder: 'Please enter expression',
       },
     },
+    parameterExtractionNode: {
+      label: '參數提取',
+      text: '利用 AI 模型提取結構化參數',
+      result: '結果',
+      selectVariables: {
+        label: '選擇變數',
+        placeholder: '請選擇變數',
+      },
+      extractParameters: {
+        label: '提取參數',
+        desc: '描述',
+        parameterType: '參數類型',
+      },
+    },
   },
   compare: {
     is_null: 'Is null',

ui/src/locales/lang/en-US/views/application-workflow.ts

@@ -442,6 +442,20 @@ export default {
         placeholder: '请输入表达式',
       },
     },
+    parameterExtractionNode: {
+      label: '参数提取',
+      text: '利用 AI 模型提取结构化参数',
+      result: '结果',
+      selectVariables: {
+        label: '选择变量',
+        placeholder: '请选择变量',
+      },
+      extractParameters: {
+        label: '提取参数',
+        desc: '描述',
+        parameterType: '参数类型',
+      },
+    },
   },
   compare: {
     is_null: '为空',

ui/src/locales/lang/zh-CN/views/application-workflow.ts

@@ -415,6 +415,20 @@ export default {
         placeholder: '請輸入表達式',
       },
     },
+    parameterExtractionNode: {
+      label: 'Parameter Extraction',
+      text: 'Extract structured parameters using AI model',
+      result: 'Result',
+      selectVariables: {
+        label: 'Select Variables',
+        placeholder: 'Please select variables',
+      },
+      extractParameters: {
+        label: 'Extract Parameters',
+        desc: 'Description',
+        parameterType: 'Parameter Type',
+      },
+    },
   },
   compare: {
     is_null: '為空',