feat: application workflow (#3415)

apps/application/flow/step_node/__init__.py

@@ -10,26 +10,24 @@ from .ai_chat_step_node import *
 from .application_node import BaseApplicationNode
 from .condition_node import *
 from .direct_reply_node import *
+from .document_extract_node import *
 from .form_node import *
-from .function_lib_node import *
+from .image_generate_step_node import *
-from .function_node import *
+from .image_understand_step_node import *
+from .mcp_node import BaseMcpNode
 from .question_node import *
 from .reranker_node import *
-
+from .search_knowledge_node import *
-from .document_extract_node import *
-from .image_understand_step_node import *
-from .image_generate_step_node import *
-
-from .search_dataset_node import *
 from .speech_to_text_step_node import BaseSpeechToTextNode
 from .start_node import *
 from .text_to_speech_step_node.impl.base_text_to_speech_node import BaseTextToSpeechNode
+from .tool_lib_node import *
+from .tool_node import *
 from .variable_assign_node import BaseVariableAssignNode
-from .mcp_node import BaseMcpNode
 
-node_list = [BaseStartStepNode, BaseChatNode, BaseSearchDatasetNode, BaseQuestionNode,
+node_list = [BaseStartStepNode, BaseChatNode, BaseSearchKnowledgeNode, BaseQuestionNode,
              BaseConditionNode, BaseReplyNode,
-             BaseFunctionNodeNode, BaseFunctionLibNodeNode, BaseRerankerNode, BaseApplicationNode,
+             BaseToolNodeNode, BaseToolLibNodeNode, BaseRerankerNode, BaseApplicationNode,
              BaseDocumentExtractNode,
              BaseImageUnderstandNode, BaseFormNode, BaseSpeechToTextNode, BaseTextToSpeechNode,
              BaseImageGenerateNode, BaseVariableAssignNode, BaseMcpNode]

apps/application/flow/step_node/search_knowledge_node/i_search_knowledge_node.py

@@ -35,9 +35,9 @@ class DatasetSettingSerializer(serializers.Serializer):
 
 class SearchDatasetStepNodeSerializer(serializers.Serializer):
     # 需要查询的数据集id列表
-    dataset_id_list = serializers.ListField(required=True, child=serializers.UUIDField(required=True),
+    knowledge_id_list = serializers.ListField(required=True, child=serializers.UUIDField(required=True),
-                                            label=_("Dataset id list"))
+                                              label=_("Dataset id list"))
-    dataset_setting = DatasetSettingSerializer(required=True)
+    knowledge_setting = DatasetSettingSerializer(required=True)
 
     question_reference_address = serializers.ListField(required=True)
 
@@ -52,8 +52,8 @@ def get_paragraph_list(chat_record, node_id):
                          'paragraph_list', []) is not None and key == node_id])
 
 
-class ISearchDatasetStepNode(INode):
+class ISearchKnowledgeStepNode(INode):
-    type = 'search-dataset-node'
+    type = 'search-knowledge-node'
 
     def get_node_params_serializer_class(self) -> Type[serializers.Serializer]:
         return SearchDatasetStepNodeSerializer

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

@@ -6,4 +6,4 @@
     @date：2024/6/11 15:35
     @desc:
 """
-from .base_search_dataset_node import BaseSearchDatasetNode
+from .base_search_knowledge_node import BaseSearchKnowledgeNode

apps/application/flow/step_node/search_knowledge_node/impl/base_search_knowledge_node.py

@@ -13,7 +13,7 @@ from django.db import connection
 from django.db.models import QuerySet
 
 from application.flow.i_step_node import NodeResult
-from application.flow.step_node.search_dataset_node.i_search_dataset_node import ISearchDatasetStepNode
+from application.flow.step_node.search_knowledge_node.i_search_knowledge_node import ISearchKnowledgeStepNode
 from common.config.embedding_config import VectorStore
 from common.db.search import native_search
 from common.utils.common import get_file_content
@@ -44,7 +44,7 @@ def reset_title(title):
         return f"#### {title}\n"
 
 
-class BaseSearchDatasetNode(ISearchDatasetStepNode):
+class BaseSearchKnowledgeNode(ISearchKnowledgeStepNode):
     def save_context(self, details, workflow_manage):
         result = details.get('paragraph_list', [])
         dataset_setting = self.node_params_serializer.data.get('dataset_setting')
@@ -60,24 +60,25 @@ class BaseSearchDatasetNode(ISearchDatasetStepNode):
              result])[0:dataset_setting.get('max_paragraph_char_number', 5000)]
         self.context['directly_return'] = directly_return
 
-    def execute(self, dataset_id_list, dataset_setting, question,
+    def execute(self, knowledge_id_list, knowledge_setting, question,
                 exclude_paragraph_id_list=None,
                 **kwargs) -> NodeResult:
         self.context['question'] = question
-        if len(dataset_id_list) == 0:
+        if len(knowledge_id_list) == 0:
             return get_none_result(question)
-        model_id = get_embedding_id(dataset_id_list)
+        model_id = get_embedding_id(knowledge_id_list)
         workspace_id = self.workflow_manage.get_body().get('workspace_id')
         embedding_model = get_model_instance_by_model_workspace_id(model_id, workspace_id)
         embedding_value = embedding_model.embed_query(question)
         vector = VectorStore.get_embedding_vector()
         exclude_document_id_list = [str(document.id) for document in
                                     QuerySet(Document).filter(
-                                        dataset_id__in=dataset_id_list,
+                                        knowledge_id__in=knowledge_id_list,
                                         is_active=False)]
-        embedding_list = vector.query(question, embedding_value, dataset_id_list, exclude_document_id_list,
+        embedding_list = vector.query(question, embedding_value, knowledge_id_list, exclude_document_id_list,
-                                      exclude_paragraph_id_list, True, dataset_setting.get('top_n'),
+                                      exclude_paragraph_id_list, True, knowledge_setting.get('top_n'),
-                                      dataset_setting.get('similarity'), SearchMode(dataset_setting.get('search_mode')))
+                                      knowledge_setting.get('similarity'),
+                                      SearchMode(knowledge_setting.get('search_mode')))
         # 手动关闭数据库连接
         connection.close()
         if embedding_list is None:
@@ -89,7 +90,7 @@ class BaseSearchDatasetNode(ISearchDatasetStepNode):
                            'is_hit_handling_method_list': [row for row in result if row.get('is_hit_handling_method')],
                            'data': '\n'.join(
                                [f"{reset_title(paragraph.get('title', ''))}{paragraph.get('content')}" for paragraph in
-                                result])[0:dataset_setting.get('max_paragraph_char_number', 5000)],
+                                result])[0:knowledge_setting.get('max_paragraph_char_number', 5000)],
                            'directly_return': '\n'.join(
                                [paragraph.get('content') for paragraph in
                                 result if
@@ -112,7 +113,7 @@ class BaseSearchDatasetNode(ISearchDatasetStepNode):
                 'update_time': paragraph.get('update_time').strftime("%Y-%m-%d %H:%M:%S"),
                 'create_time': paragraph.get('create_time').strftime("%Y-%m-%d %H:%M:%S"),
                 'id': str(paragraph.get('id')),
-                'dataset_id': str(paragraph.get('dataset_id')),
+                'knowledge_id': str(paragraph.get('knowledge_id')),
                 'document_id': str(paragraph.get('document_id'))
             }
 
@@ -124,7 +125,7 @@ class BaseSearchDatasetNode(ISearchDatasetStepNode):
         paragraph_list = native_search(QuerySet(Paragraph).filter(id__in=paragraph_id_list),
                                        get_file_content(
                                            os.path.join(PROJECT_DIR, "apps", "application", 'sql',
-                                                        'list_dataset_paragraph_by_paragraph_id.sql')),
+                                                        'list_knowledge_paragraph_by_paragraph_id.sql')),
                                        with_table_name=True)
         # 如果向量库中存在脏数据 直接删除
         if len(paragraph_list) != len(paragraph_id_list):

apps/application/flow/step_node/tool_lib_node/i_tool_lib_node.py

@@ -9,13 +9,12 @@
 from typing import Type
 
 from django.db.models import QuerySet
+from django.utils.translation import gettext_lazy as _
 from rest_framework import serializers
 
 from application.flow.i_step_node import INode, NodeResult
 from common.field.common import ObjectField
-
 from tools.models.tool import Tool
-from django.utils.translation import gettext_lazy as _
 
 
 class InputField(serializers.Serializer):
@@ -36,8 +35,8 @@ class FunctionLibNodeParamsSerializer(serializers.Serializer):
             raise Exception(_('The function has been deleted'))
 
 
-class IFunctionLibNode(INode):
+class IToolLibNode(INode):
-    type = 'function-lib-node'
+    type = 'tool-lib-node'
 
     def get_node_params_serializer_class(self) -> Type[serializers.Serializer]:
         return FunctionLibNodeParamsSerializer

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

@@ -6,4 +6,4 @@
     @date：2024/8/8 17:48
     @desc:
 """
-from .base_function_lib_node import BaseFunctionLibNodeNode
+from .base_tool_lib_node import BaseToolLibNodeNode

apps/application/flow/step_node/tool_lib_node/impl/base_tool_lib_node.py

@@ -14,7 +14,7 @@ from django.db.models import QuerySet
 from django.utils.translation import gettext as _
 
 from application.flow.i_step_node import NodeResult
-from application.flow.step_node.function_lib_node.i_function_lib_node import IFunctionLibNode
+from application.flow.step_node.tool_lib_node.i_tool_lib_node import IToolLibNode
 from common.exception.app_exception import AppApiException
 from common.utils.function_code import FunctionExecutor
 from common.utils.rsa_util import rsa_long_decrypt
@@ -110,7 +110,7 @@ def valid_function(function_lib, user_id):
         raise Exception(_('Function {name} is unavailable').format(name=function_lib.name))
 
 
-class BaseFunctionLibNodeNode(IFunctionLibNode):
+class BaseToolLibNodeNode(IToolLibNode):
     def save_context(self, details, workflow_manage):
         self.context['result'] = details.get('result')
         if self.node_params.get('is_result'):

apps/application/flow/step_node/tool_node/i_tool_node.py

@@ -51,8 +51,8 @@ class FunctionNodeParamsSerializer(serializers.Serializer):
         super().is_valid(raise_exception=True)
 
 
-class IFunctionNode(INode):
+class IToolNode(INode):
-    type = 'function-node'
+    type = 'tool-node'
 
     def get_node_params_serializer_class(self) -> Type[serializers.Serializer]:
         return FunctionNodeParamsSerializer

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

@@ -6,4 +6,4 @@
     @date：2024/8/13 11:19
     @desc:
 """
-from .base_function_node import BaseFunctionNodeNode
+from .base_tool_node import BaseToolNodeNode

apps/application/flow/step_node/tool_node/impl/base_tool_node.py

@@ -11,7 +11,7 @@ import time
 from typing import Dict
 
 from application.flow.i_step_node import NodeResult
-from application.flow.step_node.function_node.i_function_node import IFunctionNode
+from application.flow.step_node.tool_node.i_tool_node import IToolNode
 from common.utils.function_code import FunctionExecutor
 from maxkb.const import CONFIG
 
@@ -79,7 +79,7 @@ def convert_value(name: str, value, _type, is_required, source, node):
         raise Exception(f'字段:{name}类型:{_type}值:{value}类型错误')
 
 
-class BaseFunctionNodeNode(IFunctionNode):
+class BaseToolNodeNode(IToolNode):
     def save_context(self, details, workflow_manage):
         self.context['result'] = details.get('result')
         if self.node_params.get('is_result', False):

apps/application/serializers/application.py

@@ -720,14 +720,48 @@ class ApplicationOperateSerializer(serializers.Serializer):
             self.is_valid()
         application_id = self.data.get("application_id")
         application = QuerySet(Application).get(id=application_id)
-        knowledge_list = self.list_knowledge(with_valid=False)
+        available_knowledge_list = self.list_knowledge(with_valid=False)
-        mapping_knowledge_id_list = [str(akm.knowledge_id) for akm in
+        available_knowledge_dict = {knowledge.get('id'): knowledge for knowledge in available_knowledge_list}
-                                     QuerySet(ApplicationKnowledgeMapping).filter(application_id=application_id)]
+        knowledge_list = []
-        knowledge_id_list = [d.get('id') for d in
+        knowledge_id_list = []
-                             list(filter(lambda row: mapping_knowledge_id_list.__contains__(row.get('id')),
+        if application.type == 'SIMPLE':
-                                         knowledge_list))]
+            mapping_knowledge_list = QuerySet(ApplicationKnowledgeMapping).filter(application_id=application_id)
+            knowledge_list = [available_knowledge_dict.get(str(km.knowledge_id)) for km in mapping_knowledge_list if
+                              available_knowledge_dict.__contains__(str(km.knowledge_id))]
+            knowledge_id_list = [k.get('id') for k in knowledge_list]
+        else:
+            self.update_knowledge_node(application.work_flow, available_knowledge_dict)
+
         return {**ApplicationSerializerModel(application).data,
-                'knowledge_id_list': knowledge_id_list}
+                'knowledge_id_list': knowledge_id_list,
+                'knowledge_list': knowledge_list}
+
+    @staticmethod
+    def get_search_node(work_flow):
+        if work_flow is None:
+            return []
+        return [node for node in work_flow.get('nodes', []) if node.get('type', '') == 'search-knowledge-node']
+
+    def update_knowledge_node(self, workflow, available_knowledge_dict):
+        """
+        修改知识库检索节点 数据
+        定义 all_knowledge_id_list:    所有的关联知识库
+            dataset_id_list:          当前用户可看到的关联知识库列表
+            knowledge_list:           用户
+        @param workflow:              知识库
+        @param available_knowledge_dict:   当前用户可用的知识库
+        @return:
+        """
+        knowledge_node_list = self.get_search_node(workflow)
+        for search_node in knowledge_node_list:
+            node_data = search_node.get('properties', {}).get('node_data', {})
+            # 当前知识库关联的所有知识库
+            knowledge_id_list = node_data.get('knowledge_id_list', [])
+            knowledge_list = [available_knowledge_dict.get(knowledge_id) for knowledge_id in knowledge_id_list if
+                              available_knowledge_dict.__contains__(knowledge_id)]
+            node_data['all_knowledge_id_list'] = knowledge_id_list
+            node_data['knowledge_id_list'] = [knowledge.get('id') for knowledge in knowledge_list]
+            node_data['knowledge_list'] = knowledge_list
 
     def list_knowledge(self, with_valid=True):
         if with_valid:
@@ -744,8 +778,9 @@ class ApplicationOperateSerializer(serializers.Serializer):
             # 组合查询
             query = white_list_condition | default_condition
             inner = QuerySet(knowledge_workspace_authorization_model).filter(query)
-            share_knowledge_list = [KnowledgeModelSerializer(k).data for k in QuerySet(Knowledge).filter(id__in=inner)]
+            share_knowledge_list = [{**KnowledgeModelSerializer(k).data, 'scope': 'SHARED'} for k in
-        workspace_knowledge_list = [k for k in KnowledgeSerializer.Query(
+                                    QuerySet(Knowledge).filter(id__in=inner)]
+        workspace_knowledge_list = [{**k, 'scope': 'WORKSPACE'} for k in KnowledgeSerializer.Query(
             data={
                 'folder_id': 'default',
                 'workspace_id': workspace_id,

apps/application/sql/list_knowledge_paragraph_by_paragraph_id.sql

@@ -0,0 +1,11 @@
+SELECT
+	paragraph.*,
+	knowledge."name" AS "knowledge_name",
+	"document"."name" AS "document_name",
+	"document"."meta" AS "meta",
+	"document"."hit_handling_method" AS "hit_handling_method",
+	"document"."directly_return_similarity" as "directly_return_similarity"
+FROM
+	paragraph paragraph
+	LEFT JOIN knowledge knowledge ON knowledge."id" = paragraph.knowledge_id
+	LEFT JOIN "document" "document" ON "document"."id" =paragraph.document_id

ui/src/views/application-workflow/component/DropdownMenu.vue

@@ -139,13 +139,13 @@ const search_text = ref<string>('')
 const props = defineProps({
   show: {
     type: Boolean,
-    default: false
+    default: false,
   },
   id: {
     type: String,
-    default: ''
+    default: '',
   },
-  workflowRef: Object
+  workflowRef: Object,
 })
 
 const emit = defineEmits(['clickNodes', 'onmousedown'])
@@ -156,19 +156,19 @@ const activeName = ref('base')
 const toolList = ref<any[]>([])
 const filter_tool_lib_list = computed(() => {
   return toolList.value.filter((item: any) =>
-    item.name.toLocaleLowerCase().includes(search_text.value.toLocaleLowerCase())
+    item.name.toLocaleLowerCase().includes(search_text.value.toLocaleLowerCase()),
   )
 })
 const applicationList = ref<any[]>([])
 const filter_application_list = computed(() => {
   return applicationList.value.filter((item: any) =>
-    item.name.toLocaleLowerCase().includes(search_text.value.toLocaleLowerCase())
+    item.name.toLocaleLowerCase().includes(search_text.value.toLocaleLowerCase()),
   )
 })
 
 const filter_menu_nodes = computed(() => {
   return menuNodes.filter((item) =>
-    item.label.toLocaleLowerCase().includes(search_text.value.toLocaleLowerCase())
+    item.label.toLocaleLowerCase().includes(search_text.value.toLocaleLowerCase()),
   )
 })
 function clickNodes(item: any, data?: any, type?: string) {
@@ -180,8 +180,8 @@ function clickNodes(item: any, data?: any, type?: string) {
         tool_lib_id: data.id,
         input_field_list: data.input_field_list.map((field: any) => ({
           ...field,
-          value: field.source == 'reference' ? [] : ''
+          value: field.source == 'reference' ? [] : '',
-        }))
+        })),
       }
     }
     if (type == 'application') {
@@ -199,14 +199,14 @@ function clickNodes(item: any, data?: any, type?: string) {
             : {
                 ...(fileUploadSetting.document ? { document_list: [] } : {}),
                 ...(fileUploadSetting.image ? { image_list: [] } : {}),
-                ...(fileUploadSetting.audio ? { audio_list: [] } : {})
+                ...(fileUploadSetting.audio ? { audio_list: [] } : {}),
-              })
+              }),
         }
       } else {
         item['properties']['node_data'] = {
           name: data.name,
           icon: data.icon,
-          application_id: data.id
+          application_id: data.id,
         }
       }
     }
@@ -225,8 +225,8 @@ function onmousedown(item: any, data?: any, type?: string) {
         tool_lib_id: data.id,
         input_field_list: data.input_field_list.map((field: any) => ({
           ...field,
-          value: field.source == 'reference' ? [] : ''
+          value: field.source == 'reference' ? [] : '',
-        }))
+        })),
       }
     }
     if (type == 'application') {
@@ -244,14 +244,14 @@ function onmousedown(item: any, data?: any, type?: string) {
             : {
                 ...(fileUploadSetting.document ? { document_list: [] } : {}),
                 ...(fileUploadSetting.image ? { image_list: [] } : {}),
-                ...(fileUploadSetting.audio ? { audio_list: [] } : {})
+                ...(fileUploadSetting.audio ? { audio_list: [] } : {}),
-              })
+              }),
         }
       } else {
         item['properties']['node_data'] = {
           name: data.name,
           icon: data.icon,
-          application_id: data.id
+          application_id: data.id,
         }
       }
     }

ui/src/views/application/ApplicationSetting.vue

@@ -527,10 +527,9 @@ const rules = reactive<FormRules<ApplicationFormType>>({
   ],
 })
 const modelOptions = ref<any>(null)
-const knowledgeList = ref([])
+const knowledgeList = ref<Array<any>>([])
 const sttModelOptions = ref<any>(null)
 const ttsModelOptions = ref<any>(null)
-const showEditIcon = ref(false)
 
 function submitPrologueDialog(val: string) {
   applicationForm.value.prologue = val
@@ -618,8 +617,9 @@ function removeKnowledge(id: any) {
   }
 }
 
-function addKnowledge(val: Array<string>) {
+function addKnowledge(val: Array<any>) {
-  applicationForm.value.knowledge_id_list = val
+  knowledgeList.value = val
+  applicationForm.value.knowledge_id_list = val.map((item) => item.id)
 }
 
 function openKnowledgeDialog() {
@@ -633,17 +633,9 @@ function getDetail() {
     applicationForm.value.stt_model_id = res.data.stt_model
     applicationForm.value.tts_model_id = res.data.tts_model
     applicationForm.value.tts_type = res.data.tts_type
+    knowledgeList.value = res.data.knowledge_list
     applicationForm.value.model_setting.no_references_prompt =
       res.data.model_setting.no_references_prompt || '{question}'
-    application.asyncGetAccessToken(id, loading).then((res: any) => {
-      applicationForm.value = { ...applicationForm.value, ...res.data }
-    })
-  })
-}
-
-function getKnowledge() {
-  knowledge.asyncGetFolderKnowledge('', knowledgeLoading).then((res: any) => {
-    knowledgeList.value = res.data
   })
 }
 
@@ -715,12 +707,11 @@ function refreshIcon() {
 }
 
 function refresh() {
-  getKnowledge()
+  // getDetail()
 }
 
 onMounted(() => {
   getSelectModel()
-  getKnowledge()
   getDetail()
   getSTTModel()
   getTTSModel()

ui/src/views/application/component/AddKnowledgeDialog.vue

@@ -111,7 +111,7 @@ const emit = defineEmits(['addData', 'refresh'])
 const { folder, user, knowledge } = useStore()
 
 const dialogVisible = ref<boolean>(false)
-const checkList = ref([])
+const checkList = ref<Array<string>>([])
 const currentEmbedding = ref('')
 const searchValue = ref('')
 const searchDate = ref<any[]>([])
@@ -166,7 +166,10 @@ const open = (checked: any) => {
 }
 
 const submitHandle = () => {
-  emit('addData', checkList.value)
+  emit(
+    'addData',
+    searchDate.value.filter((item: any) => checkList.value.includes(item.id)),
+  )
   dialogVisible.value = false
 }
 

ui/src/workflow/nodes/search-knowledge-node/index.vue

@@ -109,7 +109,6 @@
       ref="AddknowledgeDialogRef"
       @addData="addKnowledge"
       :data="knowledgeList"
-      @refresh="refresh"
       :loading="knowledgeLoading"
     />
   </NodeContainer>
@@ -174,8 +173,13 @@ function removeknowledge(id: any) {
   set(props.nodeModel.properties.node_data, 'knowledge_id_list', list)
 }
 
-function addKnowledge(val: Array<string>) {
+function addKnowledge(val: Array<any>) {
-  set(props.nodeModel.properties.node_data, 'knowledge_id_list', val)
+  set(
+    props.nodeModel.properties.node_data,
+    'knowledge_id_list',
+    val.map((item) => item.id),
+  )
+  knowledgeList.value = val
 }
 
 function openknowledgeDialog() {
@@ -184,21 +188,6 @@ function openknowledgeDialog() {
   }
 }
 
-function getknowledge() {
-  // if (id) {
-  //   application.asyncGetApplicationKnowledge(id, knowledgeLoading).then((res: any) => {
-  //     knowledgeList.value = res.data
-  //   })
-  // } else {
-  knowledge.asyncGetFolderKnowledge('',knowledgeLoading).then((res: any) => {
-    knowledgeList.value = res.data?.filter((v: any) => v.user_id === user.userInfo?.id)
-  })
-  // }
-}
-function refresh() {
-  getknowledge()
-}
-
 const validate = () => {
   return Promise.all([
     nodeCascaderRef.value.validate(),
@@ -209,7 +198,8 @@ const validate = () => {
 }
 
 onMounted(() => {
-  getknowledge()
+  console.log(props.nodeModel.properties.node_data)
+  knowledgeList.value = props.nodeModel.properties.node_data.knowledge_list
   set(props.nodeModel, 'validate', validate)
 })
 </script>