Merge branch 'main' into aliyun/dev_0.5

# Conflicts:
#	api/core/ops/entities/config_entity.py

.devcontainer/post_create_command.sh

@@ -1,6 +1,6 @@
 #!/bin/bash
 
-npm add -g pnpm@10.8.0
+npm add -g pnpm@10.11.1
 cd web && pnpm install
 pipx install uv
 

.github/actions/setup-uv/action.yml

@@ -8,7 +8,7 @@ inputs:
   uv-version:
     description: UV version to set up
     required: true
-    default: '0.6.14'
+    default: '~=0.7.11'
   uv-lockfile:
     description: Path to the UV lockfile to restore cache from
     required: true

.gitignore

@@ -192,12 +192,12 @@ sdks/python-client/dist
 sdks/python-client/dify_client.egg-info
 
 .vscode/*
-!.vscode/launch.json
+!.vscode/launch.json.template
+!.vscode/README.md
 pyrightconfig.json
 api/.vscode
 
 .idea/
-.vscode
 
 # pnpm
 /.pnpm-store
@@ -207,3 +207,6 @@ plugins.jsonl
 
 # mise
 mise.toml
+
+# Next.js build output
+.next/

.vscode/README.md

@@ -0,0 +1,14 @@
+# Debugging with VS Code
+
+This `launch.json.template` file provides various debug configurations for the Dify project within VS Code / Cursor. To use these configurations, you should copy the contents of this file into a new file named `launch.json` in the same `.vscode` directory.
+
+## How to Use
+
+1.  **Create `launch.json`**: If you don't have one, create a file named `launch.json` inside the `.vscode` directory.
+2.  **Copy Content**: Copy the entire content from `launch.json.template` into your newly created `launch.json` file.
+3.  **Select Debug Configuration**: Go to the Run and Debug view in VS Code / Cursor (Ctrl+Shift+D or Cmd+Shift+D).
+4.  **Start Debugging**: Select the desired configuration from the dropdown menu and click the green play button.
+
+## Tips
+
+- If you need to debug with Edge browser instead of Chrome, modify the `serverReadyAction` configuration in the "Next.js: debug full stack" section, change `"debugWithChrome"` to `"debugWithEdge"` to use Microsoft Edge for debugging.

.vscode/launch.json.template

@@ -0,0 +1,68 @@
+{
+    "version": "0.2.0",
+    "configurations": [
+        {
+            "name": "Python: Flask API",
+            "type": "debugpy",
+            "request": "launch",
+            "module": "flask",
+            "env": {
+                "FLASK_APP": "app.py",
+                "FLASK_ENV": "development",
+                "GEVENT_SUPPORT": "True"
+            },
+            "args": [
+                "run",
+                "--host=0.0.0.0",
+                "--port=5001",
+                "--no-debugger",
+                "--no-reload"
+            ],
+            "jinja": true,
+            "justMyCode": true,
+            "cwd": "${workspaceFolder}/api",
+            "python": "${workspaceFolder}/api/.venv/bin/python"
+        },
+        {
+            "name": "Python: Celery Worker (Solo)",
+            "type": "debugpy",
+            "request": "launch",
+            "module": "celery",
+            "env": {
+                "GEVENT_SUPPORT": "True"
+            },
+            "args": [
+                "-A",
+                "app.celery",
+                "worker",
+                "-P",
+                "solo",
+                "-c",
+                "1",
+                "-Q",
+                "dataset,generation,mail,ops_trace",
+                "--loglevel",
+                "INFO"
+            ],
+            "justMyCode": false,
+            "cwd": "${workspaceFolder}/api",
+            "python": "${workspaceFolder}/api/.venv/bin/python"
+        },
+        {
+            "name": "Next.js: debug full stack",
+            "type": "node",
+            "request": "launch",
+            "program": "${workspaceFolder}/web/node_modules/next/dist/bin/next",
+            "runtimeArgs": ["--inspect"],
+            "skipFiles": ["<node_internals>/**"],
+            "serverReadyAction": {
+              "action": "debugWithChrome",
+              "killOnServerStop": true,
+              "pattern": "- Local:.+(https?://.+)",
+              "uriFormat": "%s",
+              "webRoot": "${workspaceFolder}/web"
+            },
+            "cwd": "${workspaceFolder}/web"
+          }
+    ]
+}

api/.env.example

@@ -491,3 +491,10 @@ OTEL_METRIC_EXPORT_TIMEOUT=30000
 
 # Prevent Clickjacking
 ALLOW_EMBED=false
+
+# Dataset queue monitor configuration
+QUEUE_MONITOR_THRESHOLD=200
+# You can configure multiple ones, separated by commas. eg: test1@dify.ai,test2@dify.ai
+QUEUE_MONITOR_ALERT_EMAILS=
+# Monitor interval in minutes, default is 30 minutes
+QUEUE_MONITOR_INTERVAL=30

api/.ruff.toml

@@ -43,6 +43,7 @@ select = [
     "S307", # suspicious-eval-usage, disallow use of `eval` and `ast.literal_eval`
     "S301", # suspicious-pickle-usage, disallow use of `pickle` and its wrappers.
     "S302", # suspicious-marshal-usage, disallow use of `marshal` module
+    "S311", # suspicious-non-cryptographic-random-usage
 ]
 
 ignore = [

api/Dockerfile

@@ -4,7 +4,7 @@ FROM python:3.12-slim-bookworm AS base
 WORKDIR /app/api
 
 # Install uv
-ENV UV_VERSION=0.6.14
+ENV UV_VERSION=0.7.11
 
 RUN pip install --no-cache-dir uv==${UV_VERSION}
 

api/commands.py

@@ -846,6 +846,9 @@ def clear_orphaned_file_records(force: bool):
         {"type": "text", "table": "workflow_node_executions", "column": "outputs"},
         {"type": "text", "table": "conversations", "column": "introduction"},
         {"type": "text", "table": "conversations", "column": "system_instruction"},
+        {"type": "text", "table": "accounts", "column": "avatar"},
+        {"type": "text", "table": "apps", "column": "icon"},
+        {"type": "text", "table": "sites", "column": "icon"},
         {"type": "json", "table": "messages", "column": "inputs"},
         {"type": "json", "table": "messages", "column": "message"},
     ]

api/configs/middleware/__init__.py

@@ -2,7 +2,7 @@ import os
 from typing import Any, Literal, Optional
 from urllib.parse import parse_qsl, quote_plus
 
-from pydantic import Field, NonNegativeInt, PositiveFloat, PositiveInt, computed_field
+from pydantic import Field, NonNegativeFloat, NonNegativeInt, PositiveFloat, PositiveInt, computed_field
 from pydantic_settings import BaseSettings
 
 from .cache.redis_config import RedisConfig
@@ -256,6 +256,25 @@ class InternalTestConfig(BaseSettings):
     )
 
 
+class DatasetQueueMonitorConfig(BaseSettings):
+    """
+    Configuration settings for Dataset Queue Monitor
+    """
+
+    QUEUE_MONITOR_THRESHOLD: Optional[NonNegativeInt] = Field(
+        description="Threshold for dataset queue monitor",
+        default=200,
+    )
+    QUEUE_MONITOR_ALERT_EMAILS: Optional[str] = Field(
+        description="Emails for dataset queue monitor alert, separated by commas",
+        default=None,
+    )
+    QUEUE_MONITOR_INTERVAL: Optional[NonNegativeFloat] = Field(
+        description="Interval for dataset queue monitor in minutes",
+        default=30,
+    )
+
+
 class MiddlewareConfig(
     # place the configs in alphabet order
     CeleryConfig,
@@ -303,5 +322,6 @@ class MiddlewareConfig(
     BaiduVectorDBConfig,
     OpenGaussConfig,
     TableStoreConfig,
+    DatasetQueueMonitorConfig,
 ):
     pass

api/configs/packaging/__init__.py

@@ -9,7 +9,7 @@ class PackagingInfo(BaseSettings):
 
     CURRENT_VERSION: str = Field(
         description="Dify version",
-        default="1.4.1",
+        default="1.4.2",
     )
 
     COMMIT_SHA: str = Field(

api/configs/remote_settings_sources/nacos/http_request.py

@@ -60,8 +60,7 @@ class NacosHttpClient:
             sign_str = tenant + "+"
         if group:
             sign_str = sign_str + group + "+"
-        if sign_str:
-            sign_str += ts
+        sign_str += ts  # Directly concatenate ts without conditional checks, because the nacos auth header forced it.
         return sign_str
 
     def get_access_token(self, force_refresh=False):

api/controllers/console/app/annotation.py

@@ -208,7 +208,7 @@ class AnnotationBatchImportApi(Resource):
         if len(request.files) > 1:
             raise TooManyFilesError()
         # check file type
-        if not file.filename.endswith(".csv"):
+        if not file.filename or not file.filename.endswith(".csv"):
             raise ValueError("Invalid file type. Only CSV files are allowed")
         return AppAnnotationService.batch_import_app_annotations(app_id, file)
 

api/controllers/console/auth/forgot_password.py

@@ -119,9 +119,6 @@ class ForgotPasswordResetApi(Resource):
         if not reset_data:
             raise InvalidTokenError()
         # Must use token in reset phase
-        if reset_data.get("phase", "") != "reset":
-            raise InvalidTokenError()
-        # Must use token in reset phase
         if reset_data.get("phase", "") != "reset":
             raise InvalidTokenError()
 

api/controllers/console/datasets/datasets_segments.py

@@ -374,7 +374,7 @@ class DatasetDocumentSegmentBatchImportApi(Resource):
         if len(request.files) > 1:
             raise TooManyFilesError()
         # check file type
-        if not file.filename.endswith(".csv"):
+        if not file.filename or not file.filename.endswith(".csv"):
             raise ValueError("Invalid file type. Only CSV files are allowed")
 
         try:

api/controllers/console/explore/installed_app.py

@@ -59,7 +59,14 @@ class InstalledAppsListApi(Resource):
         if FeatureService.get_system_features().webapp_auth.enabled:
             user_id = current_user.id
             res = []
+            app_ids = [installed_app["app"].id for installed_app in installed_app_list]
+            webapp_settings = EnterpriseService.WebAppAuth.batch_get_app_access_mode_by_id(app_ids)
             for installed_app in installed_app_list:
+                webapp_setting = webapp_settings.get(installed_app["app"].id)
+                if not webapp_setting:
+                    continue
+                if webapp_setting.access_mode == "sso_verified":
+                    continue
                 app_code = AppService.get_app_code_by_id(str(installed_app["app"].id))
                 if EnterpriseService.WebAppAuth.is_user_allowed_to_access_webapp(
                     user_id=user_id,

api/controllers/console/wraps.py

@@ -44,6 +44,17 @@ def only_edition_cloud(view):
     return decorated
 
 
+def only_edition_enterprise(view):
+    @wraps(view)
+    def decorated(*args, **kwargs):
+        if not dify_config.ENTERPRISE_ENABLED:
+            abort(404)
+
+        return view(*args, **kwargs)
+
+    return decorated
+
+
 def only_edition_self_hosted(view):
     @wraps(view)
     def decorated(*args, **kwargs):

api/controllers/inner_api/plugin/plugin.py

@@ -29,7 +29,7 @@ from core.plugin.entities.request import (
     RequestRequestUploadFile,
 )
 from core.tools.entities.tool_entities import ToolProviderType
-from libs.helper import compact_generate_response
+from libs.helper import length_prefixed_response
 from models.account import Account, Tenant
 from models.model import EndUser
 
@@ -44,7 +44,7 @@ class PluginInvokeLLMApi(Resource):
             response = PluginModelBackwardsInvocation.invoke_llm(user_model.id, tenant_model, payload)
             return PluginModelBackwardsInvocation.convert_to_event_stream(response)
 
-        return compact_generate_response(generator())
+        return length_prefixed_response(0xF, generator())
 
 
 class PluginInvokeTextEmbeddingApi(Resource):
@@ -101,7 +101,7 @@ class PluginInvokeTTSApi(Resource):
             )
             return PluginModelBackwardsInvocation.convert_to_event_stream(response)
 
-        return compact_generate_response(generator())
+        return length_prefixed_response(0xF, generator())
 
 
 class PluginInvokeSpeech2TextApi(Resource):
@@ -162,7 +162,7 @@ class PluginInvokeToolApi(Resource):
                 ),
             )
 
-        return compact_generate_response(generator())
+        return length_prefixed_response(0xF, generator())
 
 
 class PluginInvokeParameterExtractorNodeApi(Resource):
@@ -228,7 +228,7 @@ class PluginInvokeAppApi(Resource):
             files=payload.files,
         )
 
-        return compact_generate_response(PluginAppBackwardsInvocation.convert_to_event_stream(response))
+        return length_prefixed_response(0xF, PluginAppBackwardsInvocation.convert_to_event_stream(response))
 
 
 class PluginInvokeEncryptApi(Resource):

api/controllers/inner_api/plugin/wraps.py

@@ -32,6 +32,7 @@ def get_user(tenant_id: str, user_id: str | None) -> Account | EndUser:
                     )
                     session.add(user_model)
                     session.commit()
+                    session.refresh(user_model)
             else:
                 user_model = AccountService.load_user(user_id)
                 if not user_model:

api/controllers/service_api/dataset/dataset.py

@@ -1,19 +1,21 @@
 from flask import request
-from flask_restful import marshal, reqparse
+from flask_restful import marshal, marshal_with, reqparse
 from werkzeug.exceptions import Forbidden, NotFound
 
 import services.dataset_service
 from controllers.service_api import api
 from controllers.service_api.dataset.error import DatasetInUseError, DatasetNameDuplicateError
-from controllers.service_api.wraps import DatasetApiResource
+from controllers.service_api.wraps import DatasetApiResource, validate_dataset_token
 from core.model_runtime.entities.model_entities import ModelType
 from core.plugin.entities.plugin import ModelProviderID
 from core.provider_manager import ProviderManager
 from fields.dataset_fields import dataset_detail_fields
+from fields.tag_fields import tag_fields
 from libs.login import current_user
 from models.dataset import Dataset, DatasetPermissionEnum
 from services.dataset_service import DatasetPermissionService, DatasetService
 from services.entities.knowledge_entities.knowledge_entities import RetrievalModel
+from services.tag_service import TagService
 
 
 def _validate_name(name):
@@ -320,5 +322,135 @@ class DatasetApi(DatasetApiResource):
             raise DatasetInUseError()
 
 
+class DatasetTagsApi(DatasetApiResource):
+    @validate_dataset_token
+    @marshal_with(tag_fields)
+    def get(self, _, dataset_id):
+        """Get all knowledge type tags."""
+        tags = TagService.get_tags("knowledge", current_user.current_tenant_id)
+
+        return tags, 200
+
+    @validate_dataset_token
+    def post(self, _, dataset_id):
+        """Add a knowledge type tag."""
+        if not (current_user.is_editor or current_user.is_dataset_editor):
+            raise Forbidden()
+
+        parser = reqparse.RequestParser()
+        parser.add_argument(
+            "name",
+            nullable=False,
+            required=True,
+            help="Name must be between 1 to 50 characters.",
+            type=DatasetTagsApi._validate_tag_name,
+        )
+
+        args = parser.parse_args()
+        args["type"] = "knowledge"
+        tag = TagService.save_tags(args)
+
+        response = {"id": tag.id, "name": tag.name, "type": tag.type, "binding_count": 0}
+
+        return response, 200
+
+    @validate_dataset_token
+    def patch(self, _, dataset_id):
+        if not (current_user.is_editor or current_user.is_dataset_editor):
+            raise Forbidden()
+
+        parser = reqparse.RequestParser()
+        parser.add_argument(
+            "name",
+            nullable=False,
+            required=True,
+            help="Name must be between 1 to 50 characters.",
+            type=DatasetTagsApi._validate_tag_name,
+        )
+        parser.add_argument("tag_id", nullable=False, required=True, help="Id of a tag.", type=str)
+        args = parser.parse_args()
+        args["type"] = "knowledge"
+        tag = TagService.update_tags(args, args.get("tag_id"))
+
+        binding_count = TagService.get_tag_binding_count(args.get("tag_id"))
+
+        response = {"id": tag.id, "name": tag.name, "type": tag.type, "binding_count": binding_count}
+
+        return response, 200
+
+    @validate_dataset_token
+    def delete(self, _, dataset_id):
+        """Delete a knowledge type tag."""
+        if not current_user.is_editor:
+            raise Forbidden()
+        parser = reqparse.RequestParser()
+        parser.add_argument("tag_id", nullable=False, required=True, help="Id of a tag.", type=str)
+        args = parser.parse_args()
+        TagService.delete_tag(args.get("tag_id"))
+
+        return 204
+
+    @staticmethod
+    def _validate_tag_name(name):
+        if not name or len(name) < 1 or len(name) > 50:
+            raise ValueError("Name must be between 1 to 50 characters.")
+        return name
+
+
+class DatasetTagBindingApi(DatasetApiResource):
+    @validate_dataset_token
+    def post(self, _, dataset_id):
+        # The role of the current user in the ta table must be admin, owner, editor, or dataset_operator
+        if not (current_user.is_editor or current_user.is_dataset_editor):
+            raise Forbidden()
+
+        parser = reqparse.RequestParser()
+        parser.add_argument(
+            "tag_ids", type=list, nullable=False, required=True, location="json", help="Tag IDs is required."
+        )
+        parser.add_argument(
+            "target_id", type=str, nullable=False, required=True, location="json", help="Target Dataset ID is required."
+        )
+
+        args = parser.parse_args()
+        args["type"] = "knowledge"
+        TagService.save_tag_binding(args)
+
+        return 204
+
+
+class DatasetTagUnbindingApi(DatasetApiResource):
+    @validate_dataset_token
+    def post(self, _, dataset_id):
+        # The role of the current user in the ta table must be admin, owner, editor, or dataset_operator
+        if not (current_user.is_editor or current_user.is_dataset_editor):
+            raise Forbidden()
+
+        parser = reqparse.RequestParser()
+        parser.add_argument("tag_id", type=str, nullable=False, required=True, help="Tag ID is required.")
+        parser.add_argument("target_id", type=str, nullable=False, required=True, help="Target ID is required.")
+
+        args = parser.parse_args()
+        args["type"] = "knowledge"
+        TagService.delete_tag_binding(args)
+
+        return 204
+
+
+class DatasetTagsBindingStatusApi(DatasetApiResource):
+    @validate_dataset_token
+    def get(self, _, *args, **kwargs):
+        """Get all knowledge type tags."""
+        dataset_id = kwargs.get("dataset_id")
+        tags = TagService.get_tags_by_target_id("knowledge", current_user.current_tenant_id, str(dataset_id))
+        tags_list = [{"id": tag.id, "name": tag.name} for tag in tags]
+        response = {"data": tags_list, "total": len(tags)}
+        return response, 200
+
+
 api.add_resource(DatasetListApi, "/datasets")
 api.add_resource(DatasetApi, "/datasets/<uuid:dataset_id>")
+api.add_resource(DatasetTagsApi, "/datasets/tags")
+api.add_resource(DatasetTagBindingApi, "/datasets/tags/binding")
+api.add_resource(DatasetTagUnbindingApi, "/datasets/tags/unbinding")
+api.add_resource(DatasetTagsBindingStatusApi, "/datasets/<uuid:dataset_id>/tags")

api/controllers/service_api/dataset/document.py

@@ -175,8 +175,11 @@ class DocumentAddByFileApi(DatasetApiResource):
 
         if not dataset:
             raise ValueError("Dataset does not exist.")
-        if not dataset.indexing_technique and not args.get("indexing_technique"):
+
+        indexing_technique = args.get("indexing_technique") or dataset.indexing_technique
+        if not indexing_technique:
             raise ValueError("indexing_technique is required.")
+        args["indexing_technique"] = indexing_technique
 
         # save file info
         file = request.files["file"]
@@ -206,12 +209,16 @@ class DocumentAddByFileApi(DatasetApiResource):
         knowledge_config = KnowledgeConfig(**args)
         DocumentService.document_create_args_validate(knowledge_config)
 
+        dataset_process_rule = dataset.latest_process_rule if "process_rule" not in args else None
+        if not knowledge_config.original_document_id and not dataset_process_rule and not knowledge_config.process_rule:
+            raise ValueError("process_rule is required.")
+
         try:
             documents, batch = DocumentService.save_document_with_dataset_id(
                 dataset=dataset,
                 knowledge_config=knowledge_config,
                 account=dataset.created_by_account,
-                dataset_process_rule=dataset.latest_process_rule if "process_rule" not in args else None,
+                dataset_process_rule=dataset_process_rule,
                 created_from="api",
             )
         except ProviderTokenNotInitError as ex:

api/controllers/service_api/dataset/segment.py

@@ -208,6 +208,28 @@ class DatasetSegmentApi(DatasetApiResource):
         )
         return {"data": marshal(updated_segment, segment_fields), "doc_form": document.doc_form}, 200
 
+    def get(self, tenant_id, dataset_id, document_id, segment_id):
+        # check dataset
+        dataset_id = str(dataset_id)
+        tenant_id = str(tenant_id)
+        dataset = db.session.query(Dataset).filter(Dataset.tenant_id == tenant_id, Dataset.id == dataset_id).first()
+        if not dataset:
+            raise NotFound("Dataset not found.")
+        # check user's model setting
+        DatasetService.check_dataset_model_setting(dataset)
+        # check document
+        document_id = str(document_id)
+        document = DocumentService.get_document(dataset_id, document_id)
+        if not document:
+            raise NotFound("Document not found.")
+        # check segment
+        segment_id = str(segment_id)
+        segment = SegmentService.get_segment_by_id(segment_id=segment_id, tenant_id=current_user.current_tenant_id)
+        if not segment:
+            raise NotFound("Segment not found.")
+
+        return {"data": marshal(segment, segment_fields), "doc_form": document.doc_form}, 200
+
 
 class ChildChunkApi(DatasetApiResource):
     """Resource for child chunks."""

api/controllers/web/__init__.py

@@ -15,4 +15,17 @@ api.add_resource(FileApi, "/files/upload")
 api.add_resource(RemoteFileInfoApi, "/remote-files/<path:url>")
 api.add_resource(RemoteFileUploadApi, "/remote-files/upload")
 
-from . import app, audio, completion, conversation, feature, message, passport, saved_message, site, workflow
+from . import (
+    app,
+    audio,
+    completion,
+    conversation,
+    feature,
+    forgot_password,
+    login,
+    message,
+    passport,
+    saved_message,
+    site,
+    workflow,
+)

api/controllers/web/app.py

@@ -10,6 +10,8 @@ from libs.passport import PassportService
 from models.model import App, AppMode
 from services.app_service import AppService
 from services.enterprise.enterprise_service import EnterpriseService
+from services.feature_service import FeatureService
+from services.webapp_auth_service import WebAppAuthService
 
 
 class AppParameterApi(WebApiResource):
@@ -46,10 +48,22 @@ class AppMeta(WebApiResource):
 class AppAccessMode(Resource):
     def get(self):
         parser = reqparse.RequestParser()
-        parser.add_argument("appId", type=str, required=True, location="args")
+        parser.add_argument("appId", type=str, required=False, location="args")
+        parser.add_argument("appCode", type=str, required=False, location="args")
         args = parser.parse_args()
 
-        app_id = args["appId"]
+        features = FeatureService.get_system_features()
+        if not features.webapp_auth.enabled:
+            return {"accessMode": "public"}
+
+        app_id = args.get("appId")
+        if args.get("appCode"):
+            app_code = args["appCode"]
+            app_id = AppService.get_app_id_by_code(app_code)
+
+        if not app_id:
+            raise ValueError("appId or appCode must be provided")
+
         res = EnterpriseService.WebAppAuth.get_app_access_mode_by_id(app_id)
 
         return {"accessMode": res.access_mode}
@@ -75,6 +89,10 @@ class AppWebAuthPermission(Resource):
         except Exception as e:
             pass
 
+        features = FeatureService.get_system_features()
+        if not features.webapp_auth.enabled:
+            return {"result": True}
+
         parser = reqparse.RequestParser()
         parser.add_argument("appId", type=str, required=True, location="args")
         args = parser.parse_args()
@@ -82,7 +100,9 @@ class AppWebAuthPermission(Resource):
         app_id = args["appId"]
         app_code = AppService.get_app_code_by_id(app_id)
 
-        res = EnterpriseService.WebAppAuth.is_user_allowed_to_access_webapp(str(user_id), app_code)
+        res = True
+        if WebAppAuthService.is_app_require_permission_check(app_id=app_id):
+            res = EnterpriseService.WebAppAuth.is_user_allowed_to_access_webapp(str(user_id), app_code)
         return {"result": res}
 
 

api/controllers/web/forgot_password.py

@@ -0,0 +1,147 @@
+import base64
+import secrets
+
+from flask import request
+from flask_restful import Resource, reqparse
+from sqlalchemy import select
+from sqlalchemy.orm import Session
+
+from controllers.console.auth.error import (
+    EmailCodeError,
+    EmailPasswordResetLimitError,
+    InvalidEmailError,
+    InvalidTokenError,
+    PasswordMismatchError,
+)
+from controllers.console.error import AccountNotFound, EmailSendIpLimitError
+from controllers.console.wraps import email_password_login_enabled, only_edition_enterprise, setup_required
+from controllers.web import api
+from extensions.ext_database import db
+from libs.helper import email, extract_remote_ip
+from libs.password import hash_password, valid_password
+from models.account import Account
+from services.account_service import AccountService
+
+
+class ForgotPasswordSendEmailApi(Resource):
+    @only_edition_enterprise
+    @setup_required
+    @email_password_login_enabled
+    def post(self):
+        parser = reqparse.RequestParser()
+        parser.add_argument("email", type=email, required=True, location="json")
+        parser.add_argument("language", type=str, required=False, location="json")
+        args = parser.parse_args()
+
+        ip_address = extract_remote_ip(request)
+        if AccountService.is_email_send_ip_limit(ip_address):
+            raise EmailSendIpLimitError()
+
+        if args["language"] is not None and args["language"] == "zh-Hans":
+            language = "zh-Hans"
+        else:
+            language = "en-US"
+
+        with Session(db.engine) as session:
+            account = session.execute(select(Account).filter_by(email=args["email"])).scalar_one_or_none()
+        token = None
+        if account is None:
+            raise AccountNotFound()
+        else:
+            token = AccountService.send_reset_password_email(account=account, email=args["email"], language=language)
+
+        return {"result": "success", "data": token}
+
+
+class ForgotPasswordCheckApi(Resource):
+    @only_edition_enterprise
+    @setup_required
+    @email_password_login_enabled
+    def post(self):
+        parser = reqparse.RequestParser()
+        parser.add_argument("email", type=str, required=True, location="json")
+        parser.add_argument("code", type=str, required=True, location="json")
+        parser.add_argument("token", type=str, required=True, nullable=False, location="json")
+        args = parser.parse_args()
+
+        user_email = args["email"]
+
+        is_forgot_password_error_rate_limit = AccountService.is_forgot_password_error_rate_limit(args["email"])
+        if is_forgot_password_error_rate_limit:
+            raise EmailPasswordResetLimitError()
+
+        token_data = AccountService.get_reset_password_data(args["token"])
+        if token_data is None:
+            raise InvalidTokenError()
+
+        if user_email != token_data.get("email"):
+            raise InvalidEmailError()
+
+        if args["code"] != token_data.get("code"):
+            AccountService.add_forgot_password_error_rate_limit(args["email"])
+            raise EmailCodeError()
+
+        # Verified, revoke the first token
+        AccountService.revoke_reset_password_token(args["token"])
+
+        # Refresh token data by generating a new token
+        _, new_token = AccountService.generate_reset_password_token(
+            user_email, code=args["code"], additional_data={"phase": "reset"}
+        )
+
+        AccountService.reset_forgot_password_error_rate_limit(args["email"])
+        return {"is_valid": True, "email": token_data.get("email"), "token": new_token}
+
+
+class ForgotPasswordResetApi(Resource):
+    @only_edition_enterprise
+    @setup_required
+    @email_password_login_enabled
+    def post(self):
+        parser = reqparse.RequestParser()
+        parser.add_argument("token", type=str, required=True, nullable=False, location="json")
+        parser.add_argument("new_password", type=valid_password, required=True, nullable=False, location="json")
+        parser.add_argument("password_confirm", type=valid_password, required=True, nullable=False, location="json")
+        args = parser.parse_args()
+
+        # Validate passwords match
+        if args["new_password"] != args["password_confirm"]:
+            raise PasswordMismatchError()
+
+        # Validate token and get reset data
+        reset_data = AccountService.get_reset_password_data(args["token"])
+        if not reset_data:
+            raise InvalidTokenError()
+        # Must use token in reset phase
+        if reset_data.get("phase", "") != "reset":
+            raise InvalidTokenError()
+
+        # Revoke token to prevent reuse
+        AccountService.revoke_reset_password_token(args["token"])
+
+        # Generate secure salt and hash password
+        salt = secrets.token_bytes(16)
+        password_hashed = hash_password(args["new_password"], salt)
+
+        email = reset_data.get("email", "")
+
+        with Session(db.engine) as session:
+            account = session.execute(select(Account).filter_by(email=email)).scalar_one_or_none()
+
+            if account:
+                self._update_existing_account(account, password_hashed, salt, session)
+            else:
+                raise AccountNotFound()
+
+        return {"result": "success"}
+
+    def _update_existing_account(self, account, password_hashed, salt, session):
+        # Update existing account credentials
+        account.password = base64.b64encode(password_hashed).decode()
+        account.password_salt = base64.b64encode(salt).decode()
+        session.commit()
+
+
+api.add_resource(ForgotPasswordSendEmailApi, "/forgot-password")
+api.add_resource(ForgotPasswordCheckApi, "/forgot-password/validity")
+api.add_resource(ForgotPasswordResetApi, "/forgot-password/resets")

api/controllers/web/login.py

@@ -1,12 +1,11 @@
-from flask import request
 from flask_restful import Resource, reqparse
 from jwt import InvalidTokenError  # type: ignore
-from werkzeug.exceptions import BadRequest
 
 import services
 from controllers.console.auth.error import EmailCodeError, EmailOrPasswordMismatchError, InvalidEmailError
 from controllers.console.error import AccountBannedError, AccountNotFound
-from controllers.console.wraps import setup_required
+from controllers.console.wraps import only_edition_enterprise, setup_required
+from controllers.web import api
 from libs.helper import email
 from libs.password import valid_password
 from services.account_service import AccountService
@@ -16,6 +15,8 @@ from services.webapp_auth_service import WebAppAuthService
 class LoginApi(Resource):
     """Resource for web app email/password login."""
 
+    @setup_required
+    @only_edition_enterprise
     def post(self):
         """Authenticate user and login."""
         parser = reqparse.RequestParser()
@@ -23,10 +24,6 @@ class LoginApi(Resource):
         parser.add_argument("password", type=valid_password, required=True, location="json")
         args = parser.parse_args()
 
-        app_code = request.headers.get("X-App-Code")
-        if app_code is None:
-            raise BadRequest("X-App-Code header is missing.")
-
         try:
             account = WebAppAuthService.authenticate(args["email"], args["password"])
         except services.errors.account.AccountLoginError:
@@ -36,12 +33,8 @@ class LoginApi(Resource):
         except services.errors.account.AccountNotFoundError:
             raise AccountNotFound()
 
-        WebAppAuthService._validate_user_accessibility(account=account, app_code=app_code)
-
-        end_user = WebAppAuthService.create_end_user(email=args["email"], app_code=app_code)
-
-        token = WebAppAuthService.login(account=account, app_code=app_code, end_user_id=end_user.id)
-        return {"result": "success", "token": token}
+        token = WebAppAuthService.login(account=account)
+        return {"result": "success", "data": {"access_token": token}}
 
 
 # class LogoutApi(Resource):
@@ -56,6 +49,7 @@ class LoginApi(Resource):
 
 class EmailCodeLoginSendEmailApi(Resource):
     @setup_required
+    @only_edition_enterprise
     def post(self):
         parser = reqparse.RequestParser()
         parser.add_argument("email", type=email, required=True, location="json")
@@ -78,6 +72,7 @@ class EmailCodeLoginSendEmailApi(Resource):
 
 class EmailCodeLoginApi(Resource):
     @setup_required
+    @only_edition_enterprise
     def post(self):
         parser = reqparse.RequestParser()
         parser.add_argument("email", type=str, required=True, location="json")
@@ -86,9 +81,6 @@ class EmailCodeLoginApi(Resource):
         args = parser.parse_args()
 
         user_email = args["email"]
-        app_code = request.headers.get("X-App-Code")
-        if app_code is None:
-            raise BadRequest("X-App-Code header is missing.")
 
         token_data = WebAppAuthService.get_email_code_login_data(args["token"])
         if token_data is None:
@@ -105,16 +97,12 @@ class EmailCodeLoginApi(Resource):
         if not account:
             raise AccountNotFound()
 
-        WebAppAuthService._validate_user_accessibility(account=account, app_code=app_code)
-
-        end_user = WebAppAuthService.create_end_user(email=user_email, app_code=app_code)
-
-        token = WebAppAuthService.login(account=account, app_code=app_code, end_user_id=end_user.id)
+        token = WebAppAuthService.login(account=account)
         AccountService.reset_login_error_rate_limit(args["email"])
-        return {"result": "success", "token": token}
+        return {"result": "success", "data": {"access_token": token}}
 
 
-# api.add_resource(LoginApi, "/login")
+api.add_resource(LoginApi, "/login")
 # api.add_resource(LogoutApi, "/logout")
-# api.add_resource(EmailCodeLoginSendEmailApi, "/email-code-login")
-# api.add_resource(EmailCodeLoginApi, "/email-code-login/validity")
+api.add_resource(EmailCodeLoginSendEmailApi, "/email-code-login")
+api.add_resource(EmailCodeLoginApi, "/email-code-login/validity")

api/controllers/web/passport.py

@@ -1,9 +1,11 @@
 import uuid
+from datetime import UTC, datetime, timedelta
 
 from flask import request
 from flask_restful import Resource
 from werkzeug.exceptions import NotFound, Unauthorized
 
+from configs import dify_config
 from controllers.web import api
 from controllers.web.error import WebAppAuthRequiredError
 from extensions.ext_database import db
@@ -11,6 +13,7 @@ from libs.passport import PassportService
 from models.model import App, EndUser, Site
 from services.enterprise.enterprise_service import EnterpriseService
 from services.feature_service import FeatureService
+from services.webapp_auth_service import WebAppAuthService, WebAppAuthType
 
 
 class PassportResource(Resource):
@@ -20,10 +23,19 @@ class PassportResource(Resource):
         system_features = FeatureService.get_system_features()
         app_code = request.headers.get("X-App-Code")
         user_id = request.args.get("user_id")
+        web_app_access_token = request.args.get("web_app_access_token")
 
         if app_code is None:
             raise Unauthorized("X-App-Code header is missing.")
 
+        # exchange token for enterprise logined web user
+        enterprise_user_decoded = decode_enterprise_webapp_user_id(web_app_access_token)
+        if enterprise_user_decoded:
+            # a web user has already logged in, exchange a token for this app without redirecting to the login page
+            return exchange_token_for_existing_web_user(
+                app_code=app_code, enterprise_user_decoded=enterprise_user_decoded
+            )
+
         if system_features.webapp_auth.enabled:
             app_settings = EnterpriseService.WebAppAuth.get_app_access_mode_by_code(app_code=app_code)
             if not app_settings or not app_settings.access_mode == "public":
@@ -84,6 +96,128 @@ class PassportResource(Resource):
 api.add_resource(PassportResource, "/passport")
 
 
+def decode_enterprise_webapp_user_id(jwt_token: str | None):
+    """
+    Decode the enterprise user session from the Authorization header.
+    """
+    if not jwt_token:
+        return None
+
+    decoded = PassportService().verify(jwt_token)
+    source = decoded.get("token_source")
+    if not source or source != "webapp_login_token":
+        raise Unauthorized("Invalid token source. Expected 'webapp_login_token'.")
+    return decoded
+
+
+def exchange_token_for_existing_web_user(app_code: str, enterprise_user_decoded: dict):
+    """
+    Exchange a token for an existing web user session.
+    """
+    user_id = enterprise_user_decoded.get("user_id")
+    end_user_id = enterprise_user_decoded.get("end_user_id")
+    session_id = enterprise_user_decoded.get("session_id")
+    user_auth_type = enterprise_user_decoded.get("auth_type")
+    if not user_auth_type:
+        raise Unauthorized("Missing auth_type in the token.")
+
+    site = db.session.query(Site).filter(Site.code == app_code, Site.status == "normal").first()
+    if not site:
+        raise NotFound()
+
+    app_model = db.session.query(App).filter(App.id == site.app_id).first()
+    if not app_model or app_model.status != "normal" or not app_model.enable_site:
+        raise NotFound()
+
+    app_auth_type = WebAppAuthService.get_app_auth_type(app_code=app_code)
+
+    if app_auth_type == WebAppAuthType.PUBLIC:
+        return _exchange_for_public_app_token(app_model, site, enterprise_user_decoded)
+    elif app_auth_type == WebAppAuthType.EXTERNAL and user_auth_type != "external":
+        raise WebAppAuthRequiredError("Please login as external user.")
+    elif app_auth_type == WebAppAuthType.INTERNAL and user_auth_type != "internal":
+        raise WebAppAuthRequiredError("Please login as internal user.")
+
+    end_user = None
+    if end_user_id:
+        end_user = db.session.query(EndUser).filter(EndUser.id == end_user_id).first()
+    if session_id:
+        end_user = (
+            db.session.query(EndUser)
+            .filter(
+                EndUser.session_id == session_id,
+                EndUser.tenant_id == app_model.tenant_id,
+                EndUser.app_id == app_model.id,
+            )
+            .first()
+        )
+    if not end_user:
+        if not session_id:
+            raise NotFound("Missing session_id for existing web user.")
+        end_user = EndUser(
+            tenant_id=app_model.tenant_id,
+            app_id=app_model.id,
+            type="browser",
+            is_anonymous=True,
+            session_id=session_id,
+        )
+        db.session.add(end_user)
+        db.session.commit()
+    exp_dt = datetime.now(UTC) + timedelta(hours=dify_config.ACCESS_TOKEN_EXPIRE_MINUTES * 24)
+    exp = int(exp_dt.timestamp())
+    payload = {
+        "iss": site.id,
+        "sub": "Web API Passport",
+        "app_id": site.app_id,
+        "app_code": site.code,
+        "user_id": user_id,
+        "end_user_id": end_user.id,
+        "auth_type": user_auth_type,
+        "granted_at": int(datetime.now(UTC).timestamp()),
+        "token_source": "webapp",
+        "exp": exp,
+    }
+    token: str = PassportService().issue(payload)
+    return {
+        "access_token": token,
+    }
+
+
+def _exchange_for_public_app_token(app_model, site, token_decoded):
+    user_id = token_decoded.get("user_id")
+    end_user = None
+    if user_id:
+        end_user = (
+            db.session.query(EndUser).filter(EndUser.app_id == app_model.id, EndUser.session_id == user_id).first()
+        )
+
+    if not end_user:
+        end_user = EndUser(
+            tenant_id=app_model.tenant_id,
+            app_id=app_model.id,
+            type="browser",
+            is_anonymous=True,
+            session_id=generate_session_id(),
+        )
+
+        db.session.add(end_user)
+        db.session.commit()
+
+    payload = {
+        "iss": site.app_id,
+        "sub": "Web API Passport",
+        "app_id": site.app_id,
+        "app_code": site.code,
+        "end_user_id": end_user.id,
+    }
+
+    tk = PassportService().issue(payload)
+
+    return {
+        "access_token": tk,
+    }
+
+
 def generate_session_id():
     """
     Generate a unique session ID.

api/controllers/web/wraps.py

@@ -1,3 +1,4 @@
+from datetime import UTC, datetime
 from functools import wraps
 
 from flask import request
@@ -8,8 +9,9 @@ from controllers.web.error import WebAppAuthAccessDeniedError, WebAppAuthRequire
 from extensions.ext_database import db
 from libs.passport import PassportService
 from models.model import App, EndUser, Site
-from services.enterprise.enterprise_service import EnterpriseService
+from services.enterprise.enterprise_service import EnterpriseService, WebAppSettings
 from services.feature_service import FeatureService
+from services.webapp_auth_service import WebAppAuthService
 
 
 def validate_jwt_token(view=None):
@@ -45,7 +47,8 @@ def decode_jwt_token():
             raise Unauthorized("Invalid Authorization header format. Expected 'Bearer <api-key>' format.")
         decoded = PassportService().verify(tk)
         app_code = decoded.get("app_code")
-        app_model = db.session.query(App).filter(App.id == decoded["app_id"]).first()
+        app_id = decoded.get("app_id")
+        app_model = db.session.query(App).filter(App.id == app_id).first()
         site = db.session.query(Site).filter(Site.code == app_code).first()
         if not app_model:
             raise NotFound()
@@ -53,23 +56,30 @@ def decode_jwt_token():
             raise BadRequest("Site URL is no longer valid.")
         if app_model.enable_site is False:
             raise BadRequest("Site is disabled.")
-        end_user = db.session.query(EndUser).filter(EndUser.id == decoded["end_user_id"]).first()
+        end_user_id = decoded.get("end_user_id")
+        end_user = db.session.query(EndUser).filter(EndUser.id == end_user_id).first()
         if not end_user:
             raise NotFound()
 
         # for enterprise webapp auth
         app_web_auth_enabled = False
+        webapp_settings = None
         if system_features.webapp_auth.enabled:
-            app_web_auth_enabled = (
-                EnterpriseService.WebAppAuth.get_app_access_mode_by_code(app_code=app_code).access_mode != "public"
-            )
+            webapp_settings = EnterpriseService.WebAppAuth.get_app_access_mode_by_code(app_code=app_code)
+            if not webapp_settings:
+                raise NotFound("Web app settings not found.")
+            app_web_auth_enabled = webapp_settings.access_mode != "public"
 
         _validate_webapp_token(decoded, app_web_auth_enabled, system_features.webapp_auth.enabled)
-        _validate_user_accessibility(decoded, app_code, app_web_auth_enabled, system_features.webapp_auth.enabled)
+        _validate_user_accessibility(
+            decoded, app_code, app_web_auth_enabled, system_features.webapp_auth.enabled, webapp_settings
+        )
 
         return app_model, end_user
     except Unauthorized as e:
         if system_features.webapp_auth.enabled:
+            if not app_code:
+                raise Unauthorized("Please re-login to access the web app.")
             app_web_auth_enabled = (
                 EnterpriseService.WebAppAuth.get_app_access_mode_by_code(app_code=str(app_code)).access_mode != "public"
             )
@@ -95,15 +105,41 @@ def _validate_webapp_token(decoded, app_web_auth_enabled: bool, system_webapp_au
             raise Unauthorized("webapp token expired.")
 
 
-def _validate_user_accessibility(decoded, app_code, app_web_auth_enabled: bool, system_webapp_auth_enabled: bool):
+def _validate_user_accessibility(
+    decoded,
+    app_code,
+    app_web_auth_enabled: bool,
+    system_webapp_auth_enabled: bool,
+    webapp_settings: WebAppSettings | None,
+):
     if system_webapp_auth_enabled and app_web_auth_enabled:
         # Check if the user is allowed to access the web app
         user_id = decoded.get("user_id")
         if not user_id:
             raise WebAppAuthRequiredError()
 
-        if not EnterpriseService.WebAppAuth.is_user_allowed_to_access_webapp(user_id, app_code=app_code):
-            raise WebAppAuthAccessDeniedError()
+        if not webapp_settings:
+            raise WebAppAuthRequiredError("Web app settings not found.")
+
+        if WebAppAuthService.is_app_require_permission_check(access_mode=webapp_settings.access_mode):
+            if not EnterpriseService.WebAppAuth.is_user_allowed_to_access_webapp(user_id, app_code=app_code):
+                raise WebAppAuthAccessDeniedError()
+
+        auth_type = decoded.get("auth_type")
+        granted_at = decoded.get("granted_at")
+        if not auth_type:
+            raise WebAppAuthAccessDeniedError("Missing auth_type in the token.")
+        if not granted_at:
+            raise WebAppAuthAccessDeniedError("Missing granted_at in the token.")
+        # check if sso has been updated
+        if auth_type == "external":
+            last_update_time = EnterpriseService.get_app_sso_settings_last_update_time()
+            if granted_at and datetime.fromtimestamp(granted_at, tz=UTC) < last_update_time:
+                raise WebAppAuthAccessDeniedError("SSO settings have been updated. Please re-login.")
+        elif auth_type == "internal":
+            last_update_time = EnterpriseService.get_workspace_sso_settings_last_update_time()
+            if granted_at and datetime.fromtimestamp(granted_at, tz=UTC) < last_update_time:
+                raise WebAppAuthAccessDeniedError("SSO settings have been updated. Please re-login.")
 
 
 class WebApiResource(Resource):

api/core/app/app_config/easy_ui_based_app/model_config/converter.py

@@ -70,7 +70,7 @@ class ModelConfigConverter:
         if not model_mode:
             model_mode = LLMMode.CHAT.value
             if model_schema and model_schema.model_properties.get(ModelPropertyKey.MODE):
-                model_mode = LLMMode.value_of(model_schema.model_properties[ModelPropertyKey.MODE]).value
+                model_mode = LLMMode(model_schema.model_properties[ModelPropertyKey.MODE]).value
 
         if not model_schema:
             raise ValueError(f"Model {model_name} not exist.")

api/core/app/apps/advanced_chat/generate_task_pipeline.py

@@ -1,4 +1,3 @@
-import json
 import logging
 import time
 from collections.abc import Generator, Mapping
@@ -57,10 +56,9 @@ from core.app.entities.task_entities import (
     WorkflowTaskState,
 )
 from core.app.task_pipeline.based_generate_task_pipeline import BasedGenerateTaskPipeline
-from core.app.task_pipeline.message_cycle_manage import MessageCycleManage
+from core.app.task_pipeline.message_cycle_manager import MessageCycleManager
 from core.base.tts import AppGeneratorTTSPublisher, AudioTrunk
 from core.model_runtime.entities.llm_entities import LLMUsage
-from core.model_runtime.utils.encoders import jsonable_encoder
 from core.ops.ops_trace_manager import TraceQueueManager
 from core.workflow.entities.workflow_execution import WorkflowExecutionStatus, WorkflowType
 from core.workflow.enums import SystemVariableKey
@@ -141,7 +139,7 @@ class AdvancedChatAppGenerateTaskPipeline:
         )
 
         self._task_state = WorkflowTaskState()
-        self._message_cycle_manager = MessageCycleManage(
+        self._message_cycle_manager = MessageCycleManager(
             application_generate_entity=application_generate_entity, task_state=self._task_state
         )
 
@@ -162,7 +160,7 @@ class AdvancedChatAppGenerateTaskPipeline:
         :return:
         """
         # start generate conversation name thread
-        self._conversation_name_generate_thread = self._message_cycle_manager._generate_conversation_name(
+        self._conversation_name_generate_thread = self._message_cycle_manager.generate_conversation_name(
             conversation_id=self._conversation_id, query=self._application_generate_entity.query
         )
 
@@ -605,22 +603,18 @@ class AdvancedChatAppGenerateTaskPipeline:
                 yield self._message_end_to_stream_response()
                 break
             elif isinstance(event, QueueRetrieverResourcesEvent):
-                self._message_cycle_manager._handle_retriever_resources(event)
+                self._message_cycle_manager.handle_retriever_resources(event)
 
                 with Session(db.engine, expire_on_commit=False) as session:
                     message = self._get_message(session=session)
-                    message.message_metadata = (
-                        json.dumps(jsonable_encoder(self._task_state.metadata)) if self._task_state.metadata else None
-                    )
+                    message.message_metadata = self._task_state.metadata.model_dump_json()
                     session.commit()
             elif isinstance(event, QueueAnnotationReplyEvent):
-                self._message_cycle_manager._handle_annotation_reply(event)
+                self._message_cycle_manager.handle_annotation_reply(event)
 
                 with Session(db.engine, expire_on_commit=False) as session:
                     message = self._get_message(session=session)
-                    message.message_metadata = (
-                        json.dumps(jsonable_encoder(self._task_state.metadata)) if self._task_state.metadata else None
-                    )
+                    message.message_metadata = self._task_state.metadata.model_dump_json()
                     session.commit()
             elif isinstance(event, QueueTextChunkEvent):
                 delta_text = event.text
@@ -637,12 +631,12 @@ class AdvancedChatAppGenerateTaskPipeline:
                     tts_publisher.publish(queue_message)
 
                 self._task_state.answer += delta_text
-                yield self._message_cycle_manager._message_to_stream_response(
+                yield self._message_cycle_manager.message_to_stream_response(
                     answer=delta_text, message_id=self._message_id, from_variable_selector=event.from_variable_selector
                 )
             elif isinstance(event, QueueMessageReplaceEvent):
                 # published by moderation
-                yield self._message_cycle_manager._message_replace_to_stream_response(
+                yield self._message_cycle_manager.message_replace_to_stream_response(
                     answer=event.text, reason=event.reason
                 )
             elif isinstance(event, QueueAdvancedChatMessageEndEvent):
@@ -654,7 +648,7 @@ class AdvancedChatAppGenerateTaskPipeline:
                 )
                 if output_moderation_answer:
                     self._task_state.answer = output_moderation_answer
-                    yield self._message_cycle_manager._message_replace_to_stream_response(
+                    yield self._message_cycle_manager.message_replace_to_stream_response(
                         answer=output_moderation_answer,
                         reason=QueueMessageReplaceEvent.MessageReplaceReason.OUTPUT_MODERATION,
                     )
@@ -683,9 +677,7 @@ class AdvancedChatAppGenerateTaskPipeline:
         message = self._get_message(session=session)
         message.answer = self._task_state.answer
         message.provider_response_latency = time.perf_counter() - self._base_task_pipeline._start_at
-        message.message_metadata = (
-            json.dumps(jsonable_encoder(self._task_state.metadata)) if self._task_state.metadata else None
-        )
+        message.message_metadata = self._task_state.metadata.model_dump_json()
         message_files = [
             MessageFile(
                 message_id=message.id,
@@ -713,9 +705,9 @@ class AdvancedChatAppGenerateTaskPipeline:
             message.answer_price_unit = usage.completion_price_unit
             message.total_price = usage.total_price
             message.currency = usage.currency
-            self._task_state.metadata["usage"] = jsonable_encoder(usage)
+            self._task_state.metadata.usage = usage
         else:
-            self._task_state.metadata["usage"] = jsonable_encoder(LLMUsage.empty_usage())
+            self._task_state.metadata.usage = LLMUsage.empty_usage()
         message_was_created.send(
             message,
             application_generate_entity=self._application_generate_entity,
@@ -726,18 +718,16 @@ class AdvancedChatAppGenerateTaskPipeline:
         Message end to stream response.
         :return:
         """
-        extras = {}
-        if self._task_state.metadata:
-            extras["metadata"] = self._task_state.metadata.copy()
+        extras = self._task_state.metadata.model_dump()
 
-            if "annotation_reply" in extras["metadata"]:
-                del extras["metadata"]["annotation_reply"]
+        if self._task_state.metadata.annotation_reply:
+            del extras["annotation_reply"]
 
         return MessageEndStreamResponse(
             task_id=self._application_generate_entity.task_id,
             id=self._message_id,
             files=self._recorded_files,
-            metadata=extras.get("metadata", {}),
+            metadata=extras,
         )
 
     def _handle_output_moderation_chunk(self, text: str) -> bool:

api/core/app/apps/base_app_runner.py

@@ -1,3 +1,4 @@
+import logging
 import time
 from collections.abc import Generator, Mapping, Sequence
 from typing import TYPE_CHECKING, Any, Optional, Union
@@ -33,6 +34,8 @@ from models.model import App, AppMode, Message, MessageAnnotation
 if TYPE_CHECKING:
     from core.file.models import File
 
+_logger = logging.getLogger(__name__)
+
 
 class AppRunner:
     def get_pre_calculate_rest_tokens(
@@ -298,7 +301,7 @@ class AppRunner:
         )
 
     def _handle_invoke_result_stream(
-        self, invoke_result: Generator, queue_manager: AppQueueManager, agent: bool
+        self, invoke_result: Generator[LLMResultChunk, None, None], queue_manager: AppQueueManager, agent: bool
     ) -> None:
         """
         Handle invoke result
@@ -317,18 +320,28 @@ class AppRunner:
             else:
                 queue_manager.publish(QueueAgentMessageEvent(chunk=result), PublishFrom.APPLICATION_MANAGER)
 
-            text += result.delta.message.content
+            message = result.delta.message
+            if isinstance(message.content, str):
+                text += message.content
+            elif isinstance(message.content, list):
+                for content in message.content:
+                    if not isinstance(content, str):
+                        # TODO(QuantumGhost): Add multimodal output support for easy ui.
+                        _logger.warning("received multimodal output, type=%s", type(content))
+                        text += content.data
+                    else:
+                        text += content  # failback to str
 
             if not model:
                 model = result.model
 
             if not prompt_messages:
-                prompt_messages = result.prompt_messages
+                prompt_messages = list(result.prompt_messages)
 
             if result.delta.usage:
                 usage = result.delta.usage
 
-        if not usage:
+        if usage is None:
             usage = LLMUsage.empty_usage()
 
         llm_result = LLMResult(

api/core/app/apps/workflow/generate_task_pipeline.py

@@ -50,7 +50,6 @@ from core.app.entities.task_entities import (
     WorkflowAppStreamResponse,
     WorkflowFinishStreamResponse,
     WorkflowStartStreamResponse,
-    WorkflowTaskState,
 )
 from core.app.task_pipeline.based_generate_task_pipeline import BasedGenerateTaskPipeline
 from core.base.tts import AppGeneratorTTSPublisher, AudioTrunk
@@ -130,9 +129,7 @@ class WorkflowAppGenerateTaskPipeline:
         )
 
         self._application_generate_entity = application_generate_entity
-        self._workflow_id = workflow.id
         self._workflow_features_dict = workflow.features_dict
-        self._task_state = WorkflowTaskState()
         self._workflow_run_id = ""
 
     def process(self) -> Union[WorkflowAppBlockingResponse, Generator[WorkflowAppStreamResponse, None, None]]:
@@ -543,7 +540,6 @@ class WorkflowAppGenerateTaskPipeline:
                 if tts_publisher:
                     tts_publisher.publish(queue_message)
 
-                self._task_state.answer += delta_text
                 yield self._text_chunk_to_stream_response(
                     delta_text, from_variable_selector=event.from_variable_selector
                 )

api/core/app/entities/queue_entities.py

@@ -1,4 +1,4 @@
-from collections.abc import Mapping
+from collections.abc import Mapping, Sequence
 from datetime import datetime
 from enum import Enum, StrEnum
 from typing import Any, Optional
@@ -6,6 +6,7 @@ from typing import Any, Optional
 from pydantic import BaseModel
 
 from core.model_runtime.entities.llm_entities import LLMResult, LLMResultChunk
+from core.rag.entities.citation_metadata import RetrievalSourceMetadata
 from core.workflow.entities.node_entities import AgentNodeStrategyInit
 from core.workflow.entities.workflow_node_execution import WorkflowNodeExecutionMetadataKey
 from core.workflow.graph_engine.entities.graph_runtime_state import GraphRuntimeState
@@ -283,7 +284,7 @@ class QueueRetrieverResourcesEvent(AppQueueEvent):
     """
 
     event: QueueEvent = QueueEvent.RETRIEVER_RESOURCES
-    retriever_resources: list[dict]
+    retriever_resources: Sequence[RetrievalSourceMetadata]
     in_iteration_id: Optional[str] = None
     """iteration id if node is in iteration"""
     in_loop_id: Optional[str] = None

api/core/app/entities/task_entities.py

@@ -2,20 +2,37 @@ from collections.abc import Mapping, Sequence
 from enum import Enum
 from typing import Any, Optional
 
-from pydantic import BaseModel, ConfigDict
+from pydantic import BaseModel, ConfigDict, Field
 
-from core.model_runtime.entities.llm_entities import LLMResult
+from core.model_runtime.entities.llm_entities import LLMResult, LLMUsage
 from core.model_runtime.utils.encoders import jsonable_encoder
+from core.rag.entities.citation_metadata import RetrievalSourceMetadata
 from core.workflow.entities.node_entities import AgentNodeStrategyInit
 from core.workflow.entities.workflow_node_execution import WorkflowNodeExecutionMetadataKey, WorkflowNodeExecutionStatus
 
 
+class AnnotationReplyAccount(BaseModel):
+    id: str
+    name: str
+
+
+class AnnotationReply(BaseModel):
+    id: str
+    account: AnnotationReplyAccount
+
+
+class TaskStateMetadata(BaseModel):
+    annotation_reply: AnnotationReply | None = None
+    retriever_resources: Sequence[RetrievalSourceMetadata] = Field(default_factory=list)
+    usage: LLMUsage | None = None
+
+
 class TaskState(BaseModel):
     """
     TaskState entity
     """
 
-    metadata: dict = {}
+    metadata: TaskStateMetadata = Field(default_factory=TaskStateMetadata)
 
 
 class EasyUITaskState(TaskState):

api/core/app/task_pipeline/easy_ui_based_generate_task_pipeline.py

@@ -1,4 +1,3 @@
-import json
 import logging
 import time
 from collections.abc import Generator
@@ -43,15 +42,15 @@ from core.app.entities.task_entities import (
     StreamResponse,
 )
 from core.app.task_pipeline.based_generate_task_pipeline import BasedGenerateTaskPipeline
-from core.app.task_pipeline.message_cycle_manage import MessageCycleManage
+from core.app.task_pipeline.message_cycle_manager import MessageCycleManager
 from core.base.tts import AppGeneratorTTSPublisher, AudioTrunk
 from core.model_manager import ModelInstance
 from core.model_runtime.entities.llm_entities import LLMResult, LLMResultChunk, LLMResultChunkDelta, LLMUsage
 from core.model_runtime.entities.message_entities import (
     AssistantPromptMessage,
+    TextPromptMessageContent,
 )
 from core.model_runtime.model_providers.__base.large_language_model import LargeLanguageModel
-from core.model_runtime.utils.encoders import jsonable_encoder
 from core.ops.entities.trace_entity import TraceTaskName
 from core.ops.ops_trace_manager import TraceQueueManager, TraceTask
 from core.prompt.utils.prompt_message_util import PromptMessageUtil
@@ -63,7 +62,7 @@ from models.model import AppMode, Conversation, Message, MessageAgentThought
 logger = logging.getLogger(__name__)
 
 
-class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline, MessageCycleManage):
+class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline):
     """
     EasyUIBasedGenerateTaskPipeline is a class that generate stream output and state management for Application.
     """
@@ -104,6 +103,11 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline, MessageCycleMan
             )
         )
 
+        self._message_cycle_manager = MessageCycleManager(
+            application_generate_entity=application_generate_entity,
+            task_state=self._task_state,
+        )
+
         self._conversation_name_generate_thread: Optional[Thread] = None
 
     def process(
@@ -115,7 +119,7 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline, MessageCycleMan
     ]:
         if self._application_generate_entity.app_config.app_mode != AppMode.COMPLETION:
             # start generate conversation name thread
-            self._conversation_name_generate_thread = self._generate_conversation_name(
+            self._conversation_name_generate_thread = self._message_cycle_manager.generate_conversation_name(
                 conversation_id=self._conversation_id, query=self._application_generate_entity.query or ""
             )
 
@@ -136,9 +140,9 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline, MessageCycleMan
             if isinstance(stream_response, ErrorStreamResponse):
                 raise stream_response.err
             elif isinstance(stream_response, MessageEndStreamResponse):
-                extras = {"usage": jsonable_encoder(self._task_state.llm_result.usage)}
+                extras = {"usage": self._task_state.llm_result.usage.model_dump()}
                 if self._task_state.metadata:
-                    extras["metadata"] = self._task_state.metadata
+                    extras["metadata"] = self._task_state.metadata.model_dump()
                 response: Union[ChatbotAppBlockingResponse, CompletionAppBlockingResponse]
                 if self._conversation_mode == AppMode.COMPLETION.value:
                     response = CompletionAppBlockingResponse(
@@ -277,7 +281,9 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline, MessageCycleMan
                 )
                 if output_moderation_answer:
                     self._task_state.llm_result.message.content = output_moderation_answer
-                    yield self._message_replace_to_stream_response(answer=output_moderation_answer)
+                    yield self._message_cycle_manager.message_replace_to_stream_response(
+                        answer=output_moderation_answer
+                    )
 
                 with Session(db.engine) as session:
                     # Save message
@@ -286,9 +292,9 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline, MessageCycleMan
                 message_end_resp = self._message_end_to_stream_response()
                 yield message_end_resp
             elif isinstance(event, QueueRetrieverResourcesEvent):
-                self._handle_retriever_resources(event)
+                self._message_cycle_manager.handle_retriever_resources(event)
             elif isinstance(event, QueueAnnotationReplyEvent):
-                annotation = self._handle_annotation_reply(event)
+                annotation = self._message_cycle_manager.handle_annotation_reply(event)
                 if annotation:
                     self._task_state.llm_result.message.content = annotation.content
             elif isinstance(event, QueueAgentThoughtEvent):
@@ -296,7 +302,7 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline, MessageCycleMan
                 if agent_thought_response is not None:
                     yield agent_thought_response
             elif isinstance(event, QueueMessageFileEvent):
-                response = self._message_file_to_stream_response(event)
+                response = self._message_cycle_manager.message_file_to_stream_response(event)
                 if response:
                     yield response
             elif isinstance(event, QueueLLMChunkEvent | QueueAgentMessageEvent):
@@ -304,6 +310,23 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline, MessageCycleMan
                 delta_text = chunk.delta.message.content
                 if delta_text is None:
                     continue
+                if isinstance(chunk.delta.message.content, list):
+                    delta_text = ""
+                    for content in chunk.delta.message.content:
+                        logger.debug(
+                            "The content type %s in LLM chunk delta message content.: %r", type(content), content
+                        )
+                        if isinstance(content, TextPromptMessageContent):
+                            delta_text += content.data
+                        elif isinstance(content, str):
+                            delta_text += content  # failback to str
+                        else:
+                            logger.warning(
+                                "Unsupported content type %s in LLM chunk delta message content.: %r",
+                                type(content),
+                                content,
+                            )
+                            continue
 
                 if not self._task_state.llm_result.prompt_messages:
                     self._task_state.llm_result.prompt_messages = chunk.prompt_messages
@@ -318,7 +341,7 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline, MessageCycleMan
                 self._task_state.llm_result.message.content = current_content
 
                 if isinstance(event, QueueLLMChunkEvent):
-                    yield self._message_to_stream_response(
+                    yield self._message_cycle_manager.message_to_stream_response(
                         answer=cast(str, delta_text),
                         message_id=self._message_id,
                     )
@@ -328,7 +351,7 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline, MessageCycleMan
                         message_id=self._message_id,
                     )
             elif isinstance(event, QueueMessageReplaceEvent):
-                yield self._message_replace_to_stream_response(answer=event.text)
+                yield self._message_cycle_manager.message_replace_to_stream_response(answer=event.text)
             elif isinstance(event, QueuePingEvent):
                 yield self._ping_stream_response()
             else:
@@ -372,9 +395,7 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline, MessageCycleMan
         message.provider_response_latency = time.perf_counter() - self._start_at
         message.total_price = usage.total_price
         message.currency = usage.currency
-        message.message_metadata = (
-            json.dumps(jsonable_encoder(self._task_state.metadata)) if self._task_state.metadata else None
-        )
+        message.message_metadata = self._task_state.metadata.model_dump_json()
 
         if trace_manager:
             trace_manager.add_trace_task(
@@ -423,16 +444,12 @@ class EasyUIBasedGenerateTaskPipeline(BasedGenerateTaskPipeline, MessageCycleMan
         Message end to stream response.
         :return:
         """
-        self._task_state.metadata["usage"] = jsonable_encoder(self._task_state.llm_result.usage)
-
-        extras = {}
-        if self._task_state.metadata:
-            extras["metadata"] = self._task_state.metadata
-
+        self._task_state.metadata.usage = self._task_state.llm_result.usage
+        metadata_dict = self._task_state.metadata.model_dump()
         return MessageEndStreamResponse(
             task_id=self._application_generate_entity.task_id,
             id=self._message_id,
-            metadata=extras.get("metadata", {}),
+            metadata=metadata_dict,
         )
 
     def _agent_message_to_stream_response(self, answer: str, message_id: str) -> AgentMessageStreamResponse:

api/core/app/task_pipeline/message_cycle_manager.py

@@ -17,6 +17,8 @@ from core.app.entities.queue_entities import (
     QueueRetrieverResourcesEvent,
 )
 from core.app.entities.task_entities import (
+    AnnotationReply,
+    AnnotationReplyAccount,
     EasyUITaskState,
     MessageFileStreamResponse,
     MessageReplaceStreamResponse,
@@ -30,7 +32,7 @@ from models.model import AppMode, Conversation, MessageAnnotation, MessageFile
 from services.annotation_service import AppAnnotationService
 
 
-class MessageCycleManage:
+class MessageCycleManager:
     def __init__(
         self,
         *,
@@ -45,7 +47,7 @@ class MessageCycleManage:
         self._application_generate_entity = application_generate_entity
         self._task_state = task_state
 
-    def _generate_conversation_name(self, *, conversation_id: str, query: str) -> Optional[Thread]:
+    def generate_conversation_name(self, *, conversation_id: str, query: str) -> Optional[Thread]:
         """
         Generate conversation name.
         :param conversation_id: conversation id
@@ -102,7 +104,7 @@ class MessageCycleManage:
                 db.session.commit()
                 db.session.close()
 
-    def _handle_annotation_reply(self, event: QueueAnnotationReplyEvent) -> Optional[MessageAnnotation]:
+    def handle_annotation_reply(self, event: QueueAnnotationReplyEvent) -> Optional[MessageAnnotation]:
         """
         Handle annotation reply.
         :param event: event
@@ -111,25 +113,28 @@ class MessageCycleManage:
         annotation = AppAnnotationService.get_annotation_by_id(event.message_annotation_id)
         if annotation:
             account = annotation.account
-            self._task_state.metadata["annotation_reply"] = {
-                "id": annotation.id,
-                "account": {"id": annotation.account_id, "name": account.name if account else "Dify user"},
-            }
+            self._task_state.metadata.annotation_reply = AnnotationReply(
+                id=annotation.id,
+                account=AnnotationReplyAccount(
+                    id=annotation.account_id,
+                    name=account.name if account else "Dify user",
+                ),
+            )
 
             return annotation
 
         return None
 
-    def _handle_retriever_resources(self, event: QueueRetrieverResourcesEvent) -> None:
+    def handle_retriever_resources(self, event: QueueRetrieverResourcesEvent) -> None:
         """
         Handle retriever resources.
         :param event: event
         :return:
         """
         if self._application_generate_entity.app_config.additional_features.show_retrieve_source:
-            self._task_state.metadata["retriever_resources"] = event.retriever_resources
+            self._task_state.metadata.retriever_resources = event.retriever_resources
 
-    def _message_file_to_stream_response(self, event: QueueMessageFileEvent) -> Optional[MessageFileStreamResponse]:
+    def message_file_to_stream_response(self, event: QueueMessageFileEvent) -> Optional[MessageFileStreamResponse]:
         """
         Message file to stream response.
         :param event: event
@@ -166,7 +171,7 @@ class MessageCycleManage:
 
         return None
 
-    def _message_to_stream_response(
+    def message_to_stream_response(
         self, answer: str, message_id: str, from_variable_selector: Optional[list[str]] = None
     ) -> MessageStreamResponse:
         """
@@ -182,7 +187,7 @@ class MessageCycleManage:
             from_variable_selector=from_variable_selector,
         )
 
-    def _message_replace_to_stream_response(self, answer: str, reason: str = "") -> MessageReplaceStreamResponse:
+    def message_replace_to_stream_response(self, answer: str, reason: str = "") -> MessageReplaceStreamResponse:
         """
         Message replace to stream response.
         :param answer: answer

api/core/callback_handler/index_tool_callback_handler.py

@@ -1,8 +1,10 @@
 import logging
+from collections.abc import Sequence
 
 from core.app.apps.base_app_queue_manager import AppQueueManager, PublishFrom
 from core.app.entities.app_invoke_entities import InvokeFrom
 from core.app.entities.queue_entities import QueueRetrieverResourcesEvent
+from core.rag.entities.citation_metadata import RetrievalSourceMetadata
 from core.rag.index_processor.constant.index_type import IndexType
 from core.rag.models.document import Document
 from extensions.ext_database import db
@@ -85,7 +87,8 @@ class DatasetIndexToolCallbackHandler:
 
                 db.session.commit()
 
-    def return_retriever_resource_info(self, resource: list):
+    # TODO(-LAN-): Improve type check
+    def return_retriever_resource_info(self, resource: Sequence[RetrievalSourceMetadata]):
         """Handle return_retriever_resource_info."""
         self._queue_manager.publish(
             QueueRetrieverResourcesEvent(retriever_resources=resource), PublishFrom.APPLICATION_MANAGER

api/core/entities/model_entities.py

@@ -55,6 +55,25 @@ class ProviderModelWithStatusEntity(ProviderModel):
     status: ModelStatus
     load_balancing_enabled: bool = False
 
+    def raise_for_status(self) -> None:
+        """
+        Check model status and raise ValueError if not active.
+
+        :raises ValueError: When model status is not active, with a descriptive message
+        """
+        if self.status == ModelStatus.ACTIVE:
+            return
+
+        error_messages = {
+            ModelStatus.NO_CONFIGURE: "Model is not configured",
+            ModelStatus.QUOTA_EXCEEDED: "Model quota has been exceeded",
+            ModelStatus.NO_PERMISSION: "No permission to use this model",
+            ModelStatus.DISABLED: "Model is disabled",
+        }
+
+        if self.status in error_messages:
+            raise ValueError(error_messages[self.status])
+
 
 class ModelWithProviderEntity(ProviderModelWithStatusEntity):
     """

api/core/extension/extensible.py

@@ -41,45 +41,53 @@ class Extensible:
         extensions = []
         position_map: dict[str, int] = {}
 
-        # get the path of the current class
-        current_path = os.path.abspath(cls.__module__.replace(".", os.path.sep) + ".py")
-        current_dir_path = os.path.dirname(current_path)
-
-        # traverse subdirectories
-        for subdir_name in os.listdir(current_dir_path):
-            if subdir_name.startswith("__"):
-                continue
-
-            subdir_path = os.path.join(current_dir_path, subdir_name)
-            extension_name = subdir_name
-            if os.path.isdir(subdir_path):
+        # Get the package name from the module path
+        package_name = ".".join(cls.__module__.split(".")[:-1])
+
+        try:
+            # Get package directory path
+            package_spec = importlib.util.find_spec(package_name)
+            if not package_spec or not package_spec.origin:
+                raise ImportError(f"Could not find package {package_name}")
+
+            package_dir = os.path.dirname(package_spec.origin)
+
+            # Traverse subdirectories
+            for subdir_name in os.listdir(package_dir):
+                if subdir_name.startswith("__"):
+                    continue
+
+                subdir_path = os.path.join(package_dir, subdir_name)
+                if not os.path.isdir(subdir_path):
+                    continue
+
+                extension_name = subdir_name
                 file_names = os.listdir(subdir_path)
 
-                # is builtin extension, builtin extension
-                # in the front-end page and business logic, there are special treatments.
+                # Check for extension module file
+                if (extension_name + ".py") not in file_names:
+                    logging.warning(f"Missing {extension_name}.py file in {subdir_path}, Skip.")
+                    continue
+
+                # Check for builtin flag and position
                 builtin = False
-                # default position is 0 can not be None for sort_to_dict_by_position_map
                 position = 0
                 if "__builtin__" in file_names:
                     builtin = True
-
                     builtin_file_path = os.path.join(subdir_path, "__builtin__")
                     if os.path.exists(builtin_file_path):
                         position = int(Path(builtin_file_path).read_text(encoding="utf-8").strip())
                     position_map[extension_name] = position
 
-                if (extension_name + ".py") not in file_names:
-                    logging.warning(f"Missing {extension_name}.py file in {subdir_path}, Skip.")
-                    continue
-
-                # Dynamic loading {subdir_name}.py file and find the subclass of Extensible
-                py_path = os.path.join(subdir_path, extension_name + ".py")
-                spec = importlib.util.spec_from_file_location(extension_name, py_path)
+                # Import the extension module
+                module_name = f"{package_name}.{extension_name}.{extension_name}"
+                spec = importlib.util.find_spec(module_name)
                 if not spec or not spec.loader:
-                    raise Exception(f"Failed to load module {extension_name} from {py_path}")
+                    raise ImportError(f"Failed to load module {module_name}")
                 mod = importlib.util.module_from_spec(spec)
                 spec.loader.exec_module(mod)
 
+                # Find extension class
                 extension_class = None
                 for name, obj in vars(mod).items():
                     if isinstance(obj, type) and issubclass(obj, cls) and obj != cls:
@@ -87,21 +95,21 @@ class Extensible:
                         break
 
                 if not extension_class:
-                    logging.warning(f"Missing subclass of {cls.__name__} in {py_path}, Skip.")
+                    logging.warning(f"Missing subclass of {cls.__name__} in {module_name}, Skip.")
                     continue
 
+                # Load schema if not builtin
                 json_data: dict[str, Any] = {}
                 if not builtin:
-                    if "schema.json" not in file_names:
+                    json_path = os.path.join(subdir_path, "schema.json")
+                    if not os.path.exists(json_path):
                         logging.warning(f"Missing schema.json file in {subdir_path}, Skip.")
                         continue
 
-                    json_path = os.path.join(subdir_path, "schema.json")
-                    json_data = {}
-                    if os.path.exists(json_path):
-                        with open(json_path, encoding="utf-8") as f:
-                            json_data = json.load(f)
+                    with open(json_path, encoding="utf-8") as f:
+                        json_data = json.load(f)
 
+                # Create extension
                 extensions.append(
                     ModuleExtension(
                         extension_class=extension_class,
@@ -113,6 +121,11 @@ class Extensible:
                     )
                 )
 
+        except Exception as e:
+            logging.exception("Error scanning extensions")
+            raise
+
+        # Sort extensions by position
         sorted_extensions = sort_to_dict_by_position_map(
             position_map=position_map, data=extensions, name_func=lambda x: x.name
         )

api/core/helper/code_executor/code_executor.py

@@ -15,6 +15,7 @@ from core.helper.code_executor.python3.python3_transformer import Python3Templat
 from core.helper.code_executor.template_transformer import TemplateTransformer
 
 logger = logging.getLogger(__name__)
+code_execution_endpoint_url = URL(str(dify_config.CODE_EXECUTION_ENDPOINT))
 
 
 class CodeExecutionError(Exception):
@@ -64,7 +65,7 @@ class CodeExecutor:
         :param code: code
         :return:
         """
-        url = URL(str(dify_config.CODE_EXECUTION_ENDPOINT)) / "v1" / "sandbox" / "run"
+        url = code_execution_endpoint_url / "v1" / "sandbox" / "run"
 
         headers = {"X-Api-Key": dify_config.CODE_EXECUTION_API_KEY}
 

api/core/helper/marketplace.py

@@ -7,29 +7,28 @@ from configs import dify_config
 from core.helper.download import download_with_size_limit
 from core.plugin.entities.marketplace import MarketplacePluginDeclaration
 
+marketplace_api_url = URL(str(dify_config.MARKETPLACE_API_URL))
 
-def get_plugin_pkg_url(plugin_unique_identifier: str):
-    return (URL(str(dify_config.MARKETPLACE_API_URL)) / "api/v1/plugins/download").with_query(
-        unique_identifier=plugin_unique_identifier
-    )
+
+def get_plugin_pkg_url(plugin_unique_identifier: str) -> str:
+    return str((marketplace_api_url / "api/v1/plugins/download").with_query(unique_identifier=plugin_unique_identifier))
 
 
 def download_plugin_pkg(plugin_unique_identifier: str):
-    url = str(get_plugin_pkg_url(plugin_unique_identifier))
-    return download_with_size_limit(url, dify_config.PLUGIN_MAX_PACKAGE_SIZE)
+    return download_with_size_limit(get_plugin_pkg_url(plugin_unique_identifier), dify_config.PLUGIN_MAX_PACKAGE_SIZE)
 
 
 def batch_fetch_plugin_manifests(plugin_ids: list[str]) -> Sequence[MarketplacePluginDeclaration]:
     if len(plugin_ids) == 0:
         return []
 
-    url = str(URL(str(dify_config.MARKETPLACE_API_URL)) / "api/v1/plugins/batch")
+    url = str(marketplace_api_url / "api/v1/plugins/batch")
     response = requests.post(url, json={"plugin_ids": plugin_ids})
     response.raise_for_status()
     return [MarketplacePluginDeclaration(**plugin) for plugin in response.json()["data"]["plugins"]]
 
 
 def record_install_plugin_event(plugin_unique_identifier: str):
-    url = str(URL(str(dify_config.MARKETPLACE_API_URL)) / "api/v1/stats/plugins/install_count")
+    url = str(marketplace_api_url / "api/v1/stats/plugins/install_count")
     response = requests.post(url, json={"unique_identifier": plugin_unique_identifier})
     response.raise_for_status()

api/core/helper/moderation.py

@@ -1,5 +1,5 @@
 import logging
-import random
+import secrets
 from typing import cast
 
 from core.app.entities.app_invoke_entities import ModelConfigWithCredentialsEntity
@@ -38,7 +38,7 @@ def check_moderation(tenant_id: str, model_config: ModelConfigWithCredentialsEnt
             if len(text_chunks) == 0:
                 return True
 
-            text_chunk = random.choice(text_chunks)
+            text_chunk = secrets.choice(text_chunks)
 
             try:
                 model_provider_factory = ModelProviderFactory(tenant_id)

api/core/llm_generator/prompts.py

@@ -1,61 +1,20 @@
-# Written by YORKI MINAKO🤡, Edited by Xiaoyi
-CONVERSATION_TITLE_PROMPT = """You need to decompose the user's input into "subject" and "intention" in order to accurately figure out what the user's input language actually is.
-Notice: the language type user uses could be diverse, which can be English, Chinese, Italian, Español, Arabic, Japanese, French, and etc.
-ENSURE your output is in the SAME language as the user's input!
-Your output is restricted only to: (Input language) Intention + Subject(short as possible)
-Your output MUST be a valid JSON.
+# Written by YORKI MINAKO🤡, Edited by Xiaoyi, Edited by yasu-oh
+CONVERSATION_TITLE_PROMPT = """You are asked to generate a concise chat title by decomposing the user’s input into two parts: “Intention” and “Subject”.
 
-Tip: When the user's question is directed at you (the language model), you can add an emoji to make it more fun.
+1. Detect Input Language
+Automatically identify the language of the user’s input (e.g. English, Chinese, Italian, Español, Arabic, Japanese, French, and etc.).
 
+2. Generate Title
+- Combine Intention + Subject into a single, as-short-as-possible phrase.
+- The title must be natural, friendly, and in the same language as the input.
+- If the input is a direct question to the model, you may add an emoji at the end.
 
-example 1:
-User Input: hi, yesterday i had some burgers.
+3. Output Format
+Return **only** a valid JSON object with these exact keys and no additional text:
 {
-  "Language Type": "The user's input is pure English",
-  "Your Reasoning": "The language of my output must be pure English.",
-  "Your Output": "sharing yesterday's food"
-}
-
-example 2:
-User Input: hello
-{
-  "Language Type": "The user's input is pure English",
-  "Your Reasoning": "The language of my output must be pure English.",
-  "Your Output": "Greeting myself☺️"
-}
-
-
-example 3:
-User Input: why mmap file: oom
-{
-  "Language Type": "The user's input is written in pure English",
-  "Your Reasoning": "The language of my output must be pure English.",
-  "Your Output": "Asking about the reason for mmap file: oom"
-}
-
-
-example 4:
-User Input: www.convinceme.yesterday-you-ate-seafood.tv讲了什么？
-{
-  "Language Type": "The user's input English-Chinese mixed",
-  "Your Reasoning": "The English-part is an URL, the main intention is still written in Chinese, so the language of my output must be using Chinese.",
-  "Your Output": "询问网站www.convinceme.yesterday-you-ate-seafood.tv"
-}
-
-example 5:
-User Input: why小红的年龄is老than小明？
-{
-  "Language Type": "The user's input is English-Chinese mixed",
-  "Your Reasoning": "The English parts are filler words, the main intention is written in Chinese, besides, Chinese occupies a greater \"actual meaning\" than English, so the language of my output must be using Chinese.",
-  "Your Output": "询问小红和小明的年龄"
-}
-
-example 6:
-User Input: yo, 你今天咋样？
-{
-  "Language Type": "The user's input is English-Chinese mixed",
-  "Your Reasoning": "The English-part is a subjective particle, the main intention is written in Chinese, so the language of my output must be using Chinese.",
-  "Your Output": "查询今日我的状态☺️"
+  "Language Type": "<Detected language>",
+  "Your Reasoning": "<Brief explanation in that language>",
+  "Your Output": "<Intention + Subject>"
 }
 
 User Input:

api/core/model_runtime/entities/llm_entities.py

@@ -17,19 +17,6 @@ class LLMMode(StrEnum):
     COMPLETION = "completion"
     CHAT = "chat"
 
-    @classmethod
-    def value_of(cls, value: str) -> "LLMMode":
-        """
-        Get value of given mode.
-
-        :param value: mode value
-        :return: mode
-        """
-        for mode in cls:
-            if mode.value == value:
-                return mode
-        raise ValueError(f"invalid mode value {value}")
-
 
 class LLMUsage(ModelUsage):
     """

api/core/model_runtime/entities/model_entities.py

@@ -160,6 +160,10 @@ class ProviderModel(BaseModel):
     deprecated: bool = False
     model_config = ConfigDict(protected_namespaces=())
 
+    @property
+    def support_structure_output(self) -> bool:
+        return self.features is not None and ModelFeature.STRUCTURED_OUTPUT in self.features
+
 
 class ParameterRule(BaseModel):
     """

api/core/model_runtime/utils/encoders.py

@@ -129,17 +129,18 @@ def jsonable_encoder(
             sqlalchemy_safe=sqlalchemy_safe,
         )
     if dataclasses.is_dataclass(obj):
-        # FIXME: mypy error, try to fix it instead of using type: ignore
-        obj_dict = dataclasses.asdict(obj)  # type: ignore
-        return jsonable_encoder(
-            obj_dict,
-            by_alias=by_alias,
-            exclude_unset=exclude_unset,
-            exclude_defaults=exclude_defaults,
-            exclude_none=exclude_none,
-            custom_encoder=custom_encoder,
-            sqlalchemy_safe=sqlalchemy_safe,
-        )
+        # Ensure obj is a dataclass instance, not a dataclass type
+        if not isinstance(obj, type):
+            obj_dict = dataclasses.asdict(obj)
+            return jsonable_encoder(
+                obj_dict,
+                by_alias=by_alias,
+                exclude_unset=exclude_unset,
+                exclude_defaults=exclude_defaults,
+                exclude_none=exclude_none,
+                custom_encoder=custom_encoder,
+                sqlalchemy_safe=sqlalchemy_safe,
+            )
     if isinstance(obj, Enum):
         return obj.value
     if isinstance(obj, PurePath):

api/core/ops/base_trace_instance.py

@@ -1,7 +1,11 @@
 from abc import ABC, abstractmethod
 
+from sqlalchemy.orm import Session
+
 from core.ops.entities.config_entity import BaseTracingConfig
 from core.ops.entities.trace_entity import BaseTraceInfo
+from extensions.ext_database import db
+from models import Account, App, TenantAccountJoin
 
 
 class BaseTraceInstance(ABC):
@@ -24,3 +28,38 @@ class BaseTraceInstance(ABC):
         Subclasses must implement specific tracing logic for activities.
         """
         ...
+
+    def get_service_account_with_tenant(self, app_id: str) -> Account:
+        """
+        Get service account for an app and set up its tenant.
+
+        Args:
+            app_id: The ID of the app
+
+        Returns:
+            Account: The service account with tenant set up
+
+        Raises:
+            ValueError: If app, creator account or tenant cannot be found
+        """
+        with Session(db.engine, expire_on_commit=False) as session:
+            # Get the app to find its creator
+            app = session.query(App).filter(App.id == app_id).first()
+            if not app:
+                raise ValueError(f"App with id {app_id} not found")
+
+            if not app.created_by:
+                raise ValueError(f"App with id {app_id} has no creator (created_by is None)")
+
+            service_account = session.query(Account).filter(Account.id == app.created_by).first()
+            if not service_account:
+                raise ValueError(f"Creator account with id {app.created_by} not found for app {app_id}")
+
+            current_tenant = (
+                session.query(TenantAccountJoin).filter_by(account_id=service_account.id, current=True).first()
+            )
+            if not current_tenant:
+                raise ValueError(f"Current tenant not found for account {service_account.id}")
+            service_account.set_tenant_id(current_tenant.tenant_id)
+
+            return service_account

api/core/ops/entities/config_entity.py

@@ -99,6 +99,7 @@ class WeaveConfig(BaseTracingConfig):
     entity: str | None = None
     project: str
     endpoint: str = "https://trace.wandb.ai"
+    host: str | None = None
 
     @field_validator("endpoint")
     @classmethod
@@ -110,6 +111,14 @@ class WeaveConfig(BaseTracingConfig):
 
         return v
 
+    @field_validator("host")
+    @classmethod
+    def validate_host(cls, v, info: ValidationInfo):
+        if v is not None and v != "":
+            if not v.startswith(("https://", "http://")):
+                raise ValueError("host must start with https:// or http://")
+        return v
+
 class AliyunConfig(BaseTracingConfig):
     """
     Model class for Aliyun tracing config.

api/core/ops/langfuse_trace/langfuse_trace.py

@@ -4,7 +4,7 @@ from datetime import datetime, timedelta
 from typing import Optional
 
 from langfuse import Langfuse  # type: ignore
-from sqlalchemy.orm import Session, sessionmaker
+from sqlalchemy.orm import sessionmaker
 
 from core.ops.base_trace_instance import BaseTraceInstance
 from core.ops.entities.config_entity import LangfuseConfig
@@ -31,8 +31,7 @@ from core.ops.utils import filter_none_values
 from core.repositories import SQLAlchemyWorkflowNodeExecutionRepository
 from core.workflow.nodes.enums import NodeType
 from extensions.ext_database import db
-from models import Account, App, EndUser, WorkflowNodeExecutionTriggeredFrom
-from models.account import TenantAccountJoin
+from models import EndUser, WorkflowNodeExecutionTriggeredFrom
 
 logger = logging.getLogger(__name__)
 
@@ -115,28 +114,11 @@ class LangFuseDataTrace(BaseTraceInstance):
         # through workflow_run_id get all_nodes_execution using repository
         session_factory = sessionmaker(bind=db.engine)
         # Find the app's creator account
-        with Session(db.engine, expire_on_commit=False) as session:
-            # Get the app to find its creator
-            app_id = trace_info.metadata.get("app_id")
-            if not app_id:
-                raise ValueError("No app_id found in trace_info metadata")
-
-            app = session.query(App).filter(App.id == app_id).first()
-            if not app:
-                raise ValueError(f"App with id {app_id} not found")
-
-            if not app.created_by:
-                raise ValueError(f"App with id {app_id} has no creator (created_by is None)")
-
-            service_account = session.query(Account).filter(Account.id == app.created_by).first()
-            if not service_account:
-                raise ValueError(f"Creator account with id {app.created_by} not found for app {app_id}")
-            current_tenant = (
-                session.query(TenantAccountJoin).filter_by(account_id=service_account.id, current=True).first()
-            )
-            if not current_tenant:
-                raise ValueError(f"Current tenant not found for account {service_account.id}")
-            service_account.set_tenant_id(current_tenant.tenant_id)
+        app_id = trace_info.metadata.get("app_id")
+        if not app_id:
+            raise ValueError("No app_id found in trace_info metadata")
+
+        service_account = self.get_service_account_with_tenant(app_id)
 
         workflow_node_execution_repository = SQLAlchemyWorkflowNodeExecutionRepository(
             session_factory=session_factory,

api/core/ops/langsmith_trace/langsmith_trace.py

@@ -6,7 +6,7 @@ from typing import Optional, cast
 
 from langsmith import Client
 from langsmith.schemas import RunBase
-from sqlalchemy.orm import Session, sessionmaker
+from sqlalchemy.orm import sessionmaker
 
 from core.ops.base_trace_instance import BaseTraceInstance
 from core.ops.entities.config_entity import LangSmithConfig
@@ -31,7 +31,7 @@ from core.repositories import SQLAlchemyWorkflowNodeExecutionRepository
 from core.workflow.entities.workflow_node_execution import WorkflowNodeExecutionMetadataKey
 from core.workflow.nodes.enums import NodeType
 from extensions.ext_database import db
-from models import Account, App, EndUser, MessageFile, WorkflowNodeExecutionTriggeredFrom
+from models import EndUser, MessageFile, WorkflowNodeExecutionTriggeredFrom
 
 logger = logging.getLogger(__name__)
 
@@ -139,22 +139,11 @@ class LangSmithDataTrace(BaseTraceInstance):
         # through workflow_run_id get all_nodes_execution using repository
         session_factory = sessionmaker(bind=db.engine)
         # Find the app's creator account
-        with Session(db.engine, expire_on_commit=False) as session:
-            # Get the app to find its creator
-            app_id = trace_info.metadata.get("app_id")
-            if not app_id:
-                raise ValueError("No app_id found in trace_info metadata")
-
-            app = session.query(App).filter(App.id == app_id).first()
-            if not app:
-                raise ValueError(f"App with id {app_id} not found")
-
-            if not app.created_by:
-                raise ValueError(f"App with id {app_id} has no creator (created_by is None)")
-
-            service_account = session.query(Account).filter(Account.id == app.created_by).first()
-            if not service_account:
-                raise ValueError(f"Creator account with id {app.created_by} not found for app {app_id}")
+        app_id = trace_info.metadata.get("app_id")
+        if not app_id:
+            raise ValueError("No app_id found in trace_info metadata")
+
+        service_account = self.get_service_account_with_tenant(app_id)
 
         workflow_node_execution_repository = SQLAlchemyWorkflowNodeExecutionRepository(
             session_factory=session_factory,

api/core/ops/opik_trace/opik_trace.py

@@ -6,7 +6,7 @@ from typing import Optional, cast
 
 from opik import Opik, Trace
 from opik.id_helpers import uuid4_to_uuid7
-from sqlalchemy.orm import Session, sessionmaker
+from sqlalchemy.orm import sessionmaker
 
 from core.ops.base_trace_instance import BaseTraceInstance
 from core.ops.entities.config_entity import OpikConfig
@@ -25,7 +25,7 @@ from core.repositories import SQLAlchemyWorkflowNodeExecutionRepository
 from core.workflow.entities.workflow_node_execution import WorkflowNodeExecutionMetadataKey
 from core.workflow.nodes.enums import NodeType
 from extensions.ext_database import db
-from models import Account, App, EndUser, MessageFile, WorkflowNodeExecutionTriggeredFrom
+from models import EndUser, MessageFile, WorkflowNodeExecutionTriggeredFrom
 
 logger = logging.getLogger(__name__)
 
@@ -154,22 +154,11 @@ class OpikDataTrace(BaseTraceInstance):
         # through workflow_run_id get all_nodes_execution using repository
         session_factory = sessionmaker(bind=db.engine)
         # Find the app's creator account
-        with Session(db.engine, expire_on_commit=False) as session:
-            # Get the app to find its creator
-            app_id = trace_info.metadata.get("app_id")
-            if not app_id:
-                raise ValueError("No app_id found in trace_info metadata")
-
-            app = session.query(App).filter(App.id == app_id).first()
-            if not app:
-                raise ValueError(f"App with id {app_id} not found")
-
-            if not app.created_by:
-                raise ValueError(f"App with id {app_id} has no creator (created_by is None)")
-
-            service_account = session.query(Account).filter(Account.id == app.created_by).first()
-            if not service_account:
-                raise ValueError(f"Creator account with id {app.created_by} not found for app {app_id}")
+        app_id = trace_info.metadata.get("app_id")
+        if not app_id:
+            raise ValueError("No app_id found in trace_info metadata")
+
+        service_account = self.get_service_account_with_tenant(app_id)
 
         workflow_node_execution_repository = SQLAlchemyWorkflowNodeExecutionRepository(
             session_factory=session_factory,

api/core/ops/ops_trace_manager.py

@@ -81,7 +81,7 @@ class OpsTraceProviderConfigMap(dict[str, dict[str, Any]]):
                 return {
                     "config_class": WeaveConfig,
                     "secret_keys": ["api_key"],
-                    "other_keys": ["project", "entity", "endpoint"],
+                    "other_keys": ["project", "entity", "endpoint", "host"],
                     "trace_instance": WeaveDataTrace,
                 }
 

api/core/ops/weave_trace/weave_trace.py

@@ -6,7 +6,7 @@ from typing import Any, Optional, cast
 
 import wandb
 import weave
-from sqlalchemy.orm import Session, sessionmaker
+from sqlalchemy.orm import sessionmaker
 
 from core.ops.base_trace_instance import BaseTraceInstance
 from core.ops.entities.config_entity import WeaveConfig
@@ -26,7 +26,7 @@ from core.repositories import SQLAlchemyWorkflowNodeExecutionRepository
 from core.workflow.entities.workflow_node_execution import WorkflowNodeExecutionMetadataKey
 from core.workflow.nodes.enums import NodeType
 from extensions.ext_database import db
-from models import Account, App, EndUser, MessageFile, WorkflowNodeExecutionTriggeredFrom
+from models import EndUser, MessageFile, WorkflowNodeExecutionTriggeredFrom
 
 logger = logging.getLogger(__name__)
 
@@ -40,9 +40,14 @@ class WeaveDataTrace(BaseTraceInstance):
         self.weave_api_key = weave_config.api_key
         self.project_name = weave_config.project
         self.entity = weave_config.entity
+        self.host = weave_config.host
+
+        # Login with API key first, including host if provided
+        if self.host:
+            login_status = wandb.login(key=self.weave_api_key, verify=True, relogin=True, host=self.host)
+        else:
+            login_status = wandb.login(key=self.weave_api_key, verify=True, relogin=True)
 
-        # Login with API key first
-        login_status = wandb.login(key=self.weave_api_key, verify=True, relogin=True)
         if not login_status:
             logger.error("Failed to login to Weights & Biases with the provided API key")
             raise ValueError("Weave login failed")
@@ -133,22 +138,11 @@ class WeaveDataTrace(BaseTraceInstance):
         # through workflow_run_id get all_nodes_execution using repository
         session_factory = sessionmaker(bind=db.engine)
         # Find the app's creator account
-        with Session(db.engine, expire_on_commit=False) as session:
-            # Get the app to find its creator
-            app_id = trace_info.metadata.get("app_id")
-            if not app_id:
-                raise ValueError("No app_id found in trace_info metadata")
-
-            app = session.query(App).filter(App.id == app_id).first()
-            if not app:
-                raise ValueError(f"App with id {app_id} not found")
+        app_id = trace_info.metadata.get("app_id")
+        if not app_id:
+            raise ValueError("No app_id found in trace_info metadata")
 
-            if not app.created_by:
-                raise ValueError(f"App with id {app_id} has no creator (created_by is None)")
-
-            service_account = session.query(Account).filter(Account.id == app.created_by).first()
-            if not service_account:
-                raise ValueError(f"Creator account with id {app.created_by} not found for app {app_id}")
+        service_account = self.get_service_account_with_tenant(app_id)
 
         workflow_node_execution_repository = SQLAlchemyWorkflowNodeExecutionRepository(
             session_factory=session_factory,
@@ -397,7 +391,11 @@ class WeaveDataTrace(BaseTraceInstance):
 
     def api_check(self):
         try:
-            login_status = wandb.login(key=self.weave_api_key, verify=True, relogin=True)
+            if self.host:
+                login_status = wandb.login(key=self.weave_api_key, verify=True, relogin=True, host=self.host)
+            else:
+                login_status = wandb.login(key=self.weave_api_key, verify=True, relogin=True)
+
             if not login_status:
                 raise ValueError("Weave login failed")
             else:

api/core/plugin/backwards_invocation/base.py

@@ -11,14 +11,12 @@ class BaseBackwardsInvocation:
             try:
                 for chunk in response:
                     if isinstance(chunk, BaseModel | dict):
-                        yield BaseBackwardsInvocationResponse(data=chunk).model_dump_json().encode() + b"\n\n"
-                    elif isinstance(chunk, str):
-                        yield f"event: {chunk}\n\n".encode()
+                        yield BaseBackwardsInvocationResponse(data=chunk).model_dump_json().encode()
             except Exception as e:
                 error_message = BaseBackwardsInvocationResponse(error=str(e)).model_dump_json()
-                yield f"{error_message}\n\n".encode()
+                yield error_message.encode()
         else:
-            yield BaseBackwardsInvocationResponse(data=response).model_dump_json().encode() + b"\n\n"
+            yield BaseBackwardsInvocationResponse(data=response).model_dump_json().encode()
 
 
 T = TypeVar("T", bound=dict | Mapping | str | bool | int | BaseModel)

api/core/plugin/backwards_invocation/model.py

@@ -21,7 +21,7 @@ from core.plugin.entities.request import (
 )
 from core.tools.entities.tool_entities import ToolProviderType
 from core.tools.utils.model_invocation_utils import ModelInvocationUtils
-from core.workflow.nodes.llm.node import LLMNode
+from core.workflow.nodes.llm import llm_utils
 from models.account import Tenant
 
 
@@ -55,7 +55,7 @@ class PluginModelBackwardsInvocation(BaseBackwardsInvocation):
             def handle() -> Generator[LLMResultChunk, None, None]:
                 for chunk in response:
                     if chunk.delta.usage:
-                        LLMNode.deduct_llm_quota(
+                        llm_utils.deduct_llm_quota(
                             tenant_id=tenant.id, model_instance=model_instance, usage=chunk.delta.usage
                         )
                     chunk.prompt_messages = []
@@ -64,7 +64,7 @@ class PluginModelBackwardsInvocation(BaseBackwardsInvocation):
             return handle()
         else:
             if response.usage:
-                LLMNode.deduct_llm_quota(tenant_id=tenant.id, model_instance=model_instance, usage=response.usage)
+                llm_utils.deduct_llm_quota(tenant_id=tenant.id, model_instance=model_instance, usage=response.usage)
 
             def handle_non_streaming(response: LLMResult) -> Generator[LLMResultChunk, None, None]:
                 yield LLMResultChunk(

api/core/plugin/impl/base.py

@@ -31,8 +31,7 @@ from core.plugin.impl.exc import (
     PluginUniqueIdentifierError,
 )
 
-plugin_daemon_inner_api_baseurl = dify_config.PLUGIN_DAEMON_URL
-plugin_daemon_inner_api_key = dify_config.PLUGIN_DAEMON_KEY
+plugin_daemon_inner_api_baseurl = URL(str(dify_config.PLUGIN_DAEMON_URL))
 
 T = TypeVar("T", bound=(BaseModel | dict | list | bool | str))
 
@@ -53,9 +52,9 @@ class BasePluginClient:
         """
         Make a request to the plugin daemon inner API.
         """
-        url = URL(str(plugin_daemon_inner_api_baseurl)) / path
+        url = plugin_daemon_inner_api_baseurl / path
         headers = headers or {}
-        headers["X-Api-Key"] = plugin_daemon_inner_api_key
+        headers["X-Api-Key"] = dify_config.PLUGIN_DAEMON_KEY
         headers["Accept-Encoding"] = "gzip, deflate, br"
 
         if headers.get("Content-Type") == "application/json" and isinstance(data, dict):

api/core/provider_manager.py

@@ -3,7 +3,9 @@ from collections import defaultdict
 from json import JSONDecodeError
 from typing import Any, Optional, cast
 
+from sqlalchemy import select
 from sqlalchemy.exc import IntegrityError
+from sqlalchemy.orm import Session
 
 from configs import dify_config
 from core.entities.model_entities import DefaultModelEntity, DefaultModelProviderEntity
@@ -393,19 +395,13 @@ class ProviderManager:
 
     @staticmethod
     def _get_all_providers(tenant_id: str) -> dict[str, list[Provider]]:
-        """
-        Get all provider records of the workspace.
-
-        :param tenant_id: workspace id
-        :return:
-        """
-        providers = db.session.query(Provider).filter(Provider.tenant_id == tenant_id, Provider.is_valid == True).all()
-
         provider_name_to_provider_records_dict = defaultdict(list)
-        for provider in providers:
-            # TODO: Use provider name with prefix after the data migration
-            provider_name_to_provider_records_dict[str(ModelProviderID(provider.provider_name))].append(provider)
-
+        with Session(db.engine, expire_on_commit=False) as session:
+            stmt = select(Provider).where(Provider.tenant_id == tenant_id, Provider.is_valid == True)
+            providers = session.scalars(stmt)
+            for provider in providers:
+                # Use provider name with prefix after the data migration
+                provider_name_to_provider_records_dict[str(ModelProviderID(provider.provider_name))].append(provider)
         return provider_name_to_provider_records_dict
 
     @staticmethod
@@ -416,17 +412,12 @@ class ProviderManager:
         :param tenant_id: workspace id
         :return:
         """
-        # Get all provider model records of the workspace
-        provider_models = (
-            db.session.query(ProviderModel)
-            .filter(ProviderModel.tenant_id == tenant_id, ProviderModel.is_valid == True)
-            .all()
-        )
-
         provider_name_to_provider_model_records_dict = defaultdict(list)
-        for provider_model in provider_models:
-            provider_name_to_provider_model_records_dict[provider_model.provider_name].append(provider_model)
-
+        with Session(db.engine, expire_on_commit=False) as session:
+            stmt = select(ProviderModel).where(ProviderModel.tenant_id == tenant_id, ProviderModel.is_valid == True)
+            provider_models = session.scalars(stmt)
+            for provider_model in provider_models:
+                provider_name_to_provider_model_records_dict[provider_model.provider_name].append(provider_model)
         return provider_name_to_provider_model_records_dict
 
     @staticmethod
@@ -437,17 +428,14 @@ class ProviderManager:
         :param tenant_id: workspace id
         :return:
         """
-        preferred_provider_types = (
-            db.session.query(TenantPreferredModelProvider)
-            .filter(TenantPreferredModelProvider.tenant_id == tenant_id)
-            .all()
-        )
-
-        provider_name_to_preferred_provider_type_records_dict = {
-            preferred_provider_type.provider_name: preferred_provider_type
-            for preferred_provider_type in preferred_provider_types
-        }
-
+        provider_name_to_preferred_provider_type_records_dict = {}
+        with Session(db.engine, expire_on_commit=False) as session:
+            stmt = select(TenantPreferredModelProvider).where(TenantPreferredModelProvider.tenant_id == tenant_id)
+            preferred_provider_types = session.scalars(stmt)
+            provider_name_to_preferred_provider_type_records_dict = {
+                preferred_provider_type.provider_name: preferred_provider_type
+                for preferred_provider_type in preferred_provider_types
+            }
         return provider_name_to_preferred_provider_type_records_dict
 
     @staticmethod
@@ -458,18 +446,14 @@ class ProviderManager:
         :param tenant_id: workspace id
         :return:
         """
-        provider_model_settings = (
-            db.session.query(ProviderModelSetting).filter(ProviderModelSetting.tenant_id == tenant_id).all()
-        )
-
         provider_name_to_provider_model_settings_dict = defaultdict(list)
-        for provider_model_setting in provider_model_settings:
-            (
+        with Session(db.engine, expire_on_commit=False) as session:
+            stmt = select(ProviderModelSetting).where(ProviderModelSetting.tenant_id == tenant_id)
+            provider_model_settings = session.scalars(stmt)
+            for provider_model_setting in provider_model_settings:
                 provider_name_to_provider_model_settings_dict[provider_model_setting.provider_name].append(
                     provider_model_setting
                 )
-            )
-
         return provider_name_to_provider_model_settings_dict
 
     @staticmethod
@@ -492,15 +476,14 @@ class ProviderManager:
         if not model_load_balancing_enabled:
             return {}
 
-        provider_load_balancing_configs = (
-            db.session.query(LoadBalancingModelConfig).filter(LoadBalancingModelConfig.tenant_id == tenant_id).all()
-        )
-
         provider_name_to_provider_load_balancing_model_configs_dict = defaultdict(list)
-        for provider_load_balancing_config in provider_load_balancing_configs:
-            provider_name_to_provider_load_balancing_model_configs_dict[
-                provider_load_balancing_config.provider_name
-            ].append(provider_load_balancing_config)
+        with Session(db.engine, expire_on_commit=False) as session:
+            stmt = select(LoadBalancingModelConfig).where(LoadBalancingModelConfig.tenant_id == tenant_id)
+            provider_load_balancing_configs = session.scalars(stmt)
+            for provider_load_balancing_config in provider_load_balancing_configs:
+                provider_name_to_provider_load_balancing_model_configs_dict[
+                    provider_load_balancing_config.provider_name
+                ].append(provider_load_balancing_config)
 
         return provider_name_to_provider_load_balancing_model_configs_dict
 
@@ -626,10 +609,9 @@ class ProviderManager:
             if not cached_provider_credentials:
                 try:
                     # fix origin data
-                    if (
-                        custom_provider_record.encrypted_config
-                        and not custom_provider_record.encrypted_config.startswith("{")
-                    ):
+                    if custom_provider_record.encrypted_config is None:
+                        raise ValueError("No credentials found")
+                    if not custom_provider_record.encrypted_config.startswith("{"):
                         provider_credentials = {"openai_api_key": custom_provider_record.encrypted_config}
                     else:
                         provider_credentials = json.loads(custom_provider_record.encrypted_config)
@@ -733,7 +715,7 @@ class ProviderManager:
             return SystemConfiguration(enabled=False)
 
         # Convert provider_records to dict
-        quota_type_to_provider_records_dict = {}
+        quota_type_to_provider_records_dict: dict[ProviderQuotaType, Provider] = {}
         for provider_record in provider_records:
             if provider_record.provider_type != ProviderType.SYSTEM.value:
                 continue
@@ -758,6 +740,11 @@ class ProviderManager:
             else:
                 provider_record = quota_type_to_provider_records_dict[provider_quota.quota_type]
 
+                if provider_record.quota_used is None:
+                    raise ValueError("quota_used is None")
+                if provider_record.quota_limit is None:
+                    raise ValueError("quota_limit is None")
+
                 quota_configuration = QuotaConfiguration(
                     quota_type=provider_quota.quota_type,
                     quota_unit=provider_hosting_configuration.quota_unit or QuotaUnit.TOKENS,
@@ -791,10 +778,9 @@ class ProviderManager:
                 cached_provider_credentials = provider_credentials_cache.get()
 
                 if not cached_provider_credentials:
-                    try:
-                        provider_credentials: dict[str, Any] = json.loads(provider_record.encrypted_config)
-                    except JSONDecodeError:
-                        provider_credentials = {}
+                    provider_credentials: dict[str, Any] = {}
+                    if provider_records and provider_records[0].encrypted_config:
+                        provider_credentials = json.loads(provider_records[0].encrypted_config)
 
                     # Get provider credential secret variables
                     provider_credential_secret_variables = self._extract_secret_variables(

api/core/rag/datasource/keyword/jieba/stopwords.py

@@ -720,7 +720,7 @@ STOPWORDS = {
     "〉",
     "〈",
     "…",
-    "　",
+    " ",
     "0",
     "1",
     "2",
@@ -731,16 +731,6 @@ STOPWORDS = {
     "7",
     "8",
     "9",
-    "０",
-    "１",
-    "２",
-    "３",
-    "４",
-    "５",
-    "６",
-    "７",
-    "８",
-    "９",
     "二",
     "三",
     "四",

api/core/rag/datasource/vdb/baidu/baidu_vector.py

@@ -85,7 +85,6 @@ class BaiduVector(BaseVector):
             end = min(start + batch_size, total_count)
             rows = []
             assert len(metadatas) == total_count, "metadatas length should be equal to total_count"
-            # FIXME do you need this assert?
             for i in range(start, end, 1):
                 row = Row(
                     id=metadatas[i].get("doc_id", str(uuid.uuid4())),

api/core/rag/datasource/vdb/elasticsearch/elasticsearch_vector.py

@@ -142,7 +142,7 @@ class ElasticSearchVector(BaseVector):
             if score > score_threshold:
                 if doc.metadata is not None:
                     doc.metadata["score"] = score
-            docs.append(doc)
+                    docs.append(doc)
 
         return docs
 

api/core/rag/datasource/vdb/milvus/milvus_vector.py

@@ -97,6 +97,10 @@ class MilvusVector(BaseVector):
 
         try:
             milvus_version = self._client.get_server_version()
+            # Check if it's Zilliz Cloud - it supports full-text search with Milvus 2.5 compatibility
+            if "Zilliz Cloud" in milvus_version:
+                return True
+            # For standard Milvus installations, check version number
             return version.parse(milvus_version).base_version >= version.parse("2.5.0").base_version
         except Exception as e:
             logger.warning(f"Failed to check Milvus version: {str(e)}. Disabling hybrid search.")

api/core/rag/datasource/vdb/opensearch/opensearch_vector.py

@@ -184,7 +184,16 @@ class OpenSearchVector(BaseVector):
         }
         document_ids_filter = kwargs.get("document_ids_filter")
         if document_ids_filter:
-            query["query"] = {"terms": {"metadata.document_id": document_ids_filter}}
+            query["query"] = {
+                "script_score": {
+                    "query": {"bool": {"filter": [{"terms": {Field.DOCUMENT_ID.value: document_ids_filter}}]}},
+                    "script": {
+                        "source": "knn_score",
+                        "lang": "knn",
+                        "params": {"field": Field.VECTOR.value, "query_value": query_vector, "space_type": "l2"},
+                    },
+                }
+            }
 
         try:
             response = self._client.search(index=self._collection_name.lower(), body=query)
@@ -209,10 +218,10 @@ class OpenSearchVector(BaseVector):
         return docs
 
     def search_by_full_text(self, query: str, **kwargs: Any) -> list[Document]:
-        full_text_query = {"query": {"match": {Field.CONTENT_KEY.value: query}}}
+        full_text_query = {"query": {"bool": {"must": [{"match": {Field.CONTENT_KEY.value: query}}]}}}
         document_ids_filter = kwargs.get("document_ids_filter")
         if document_ids_filter:
-            full_text_query["query"]["terms"] = {"metadata.document_id": document_ids_filter}
+            full_text_query["query"]["bool"]["filter"] = [{"terms": {"metadata.document_id": document_ids_filter}}]
 
         response = self._client.search(index=self._collection_name.lower(), body=full_text_query)
 
@@ -255,7 +264,8 @@ class OpenSearchVector(BaseVector):
                             Field.METADATA_KEY.value: {
                                 "type": "object",
                                 "properties": {
-                                    "doc_id": {"type": "keyword"}  # Map doc_id to keyword type
+                                    "doc_id": {"type": "keyword"},  # Map doc_id to keyword type
+                                    "document_id": {"type": "keyword"},
                                 },
                             },
                         }

api/core/rag/datasource/vdb/oracle/oraclevector.py

@@ -261,7 +261,7 @@ class OracleVector(BaseVector):
                 words = pseg.cut(query)
                 current_entity = ""
                 for word, pos in words:
-                    if pos in {"nr", "Ng", "eng", "nz", "n", "ORG", "v"}:  # nr: 人名, ns: 地名, nt: 机构名
+                    if pos in {"nr", "Ng", "eng", "nz", "n", "ORG", "v"}:  # nr: 人名，ns: 地名，nt: 机构名
                         current_entity += word
                     else:
                         if current_entity:
@@ -303,7 +303,6 @@ class OracleVector(BaseVector):
             return docs
         else:
             return [Document(page_content="", metadata={})]
-        return []
 
     def delete(self) -> None:
         with self._get_connection() as conn:

api/core/rag/datasource/vdb/tidb_on_qdrant/tidb_service.py

@@ -245,4 +245,4 @@ class TidbService:
             return cluster_infos
         else:
             response.raise_for_status()
-            return []  # FIXME for mypy, This line will not be reached as raise_for_status() will raise an exception
+            return []

api/core/rag/embedding/cached_embedding.py

@@ -139,4 +139,4 @@ class CacheEmbedding(Embeddings):
                 logging.exception(f"Failed to add embedding to redis for the text '{text[:10]}...({len(text)} chars)'")
             raise ex
 
-        return embedding_results
+        return embedding_results  # type: ignore

api/core/rag/entities/citation_metadata.py

@@ -0,0 +1,23 @@
+from typing import Any, Optional
+
+from pydantic import BaseModel
+
+
+class RetrievalSourceMetadata(BaseModel):
+    position: Optional[int] = None
+    dataset_id: Optional[str] = None
+    dataset_name: Optional[str] = None
+    document_id: Optional[str] = None
+    document_name: Optional[str] = None
+    data_source_type: Optional[str] = None
+    segment_id: Optional[str] = None
+    retriever_from: Optional[str] = None
+    score: Optional[float] = None
+    hit_count: Optional[int] = None
+    word_count: Optional[int] = None
+    segment_position: Optional[int] = None
+    index_node_hash: Optional[str] = None
+    content: Optional[str] = None
+    page: Optional[int] = None
+    doc_metadata: Optional[dict[str, Any]] = None
+    title: Optional[str] = None

api/core/rag/index_processor/processor/qa_index_processor.py

@@ -104,7 +104,7 @@ class QAIndexProcessor(BaseIndexProcessor):
 
     def format_by_template(self, file: FileStorage, **kwargs) -> list[Document]:
         # check file type
-        if not file.filename.endswith(".csv"):
+        if not file.filename or not file.filename.endswith(".csv"):
             raise ValueError("Invalid file type. Only CSV files are allowed")
 
         try:

api/core/rag/retrieval/dataset_retrieval.py

@@ -35,6 +35,7 @@ from core.prompt.simple_prompt_transform import ModelMode
 from core.rag.data_post_processor.data_post_processor import DataPostProcessor
 from core.rag.datasource.keyword.jieba.jieba_keyword_table_handler import JiebaKeywordTableHandler
 from core.rag.datasource.retrieval_service import RetrievalService
+from core.rag.entities.citation_metadata import RetrievalSourceMetadata
 from core.rag.entities.context_entities import DocumentContext
 from core.rag.entities.metadata_entities import Condition, MetadataCondition
 from core.rag.index_processor.constant.index_type import IndexType
@@ -198,21 +199,21 @@ class DatasetRetrieval:
 
         dify_documents = [item for item in all_documents if item.provider == "dify"]
         external_documents = [item for item in all_documents if item.provider == "external"]
-        document_context_list = []
-        retrieval_resource_list = []
+        document_context_list: list[DocumentContext] = []
+        retrieval_resource_list: list[RetrievalSourceMetadata] = []
         # deal with external documents
         for item in external_documents:
             document_context_list.append(DocumentContext(content=item.page_content, score=item.metadata.get("score")))
-            source = {
-                "dataset_id": item.metadata.get("dataset_id"),
-                "dataset_name": item.metadata.get("dataset_name"),
-                "document_id": item.metadata.get("document_id") or item.metadata.get("title"),
-                "document_name": item.metadata.get("title"),
-                "data_source_type": "external",
-                "retriever_from": invoke_from.to_source(),
-                "score": item.metadata.get("score"),
-                "content": item.page_content,
-            }
+            source = RetrievalSourceMetadata(
+                dataset_id=item.metadata.get("dataset_id"),
+                dataset_name=item.metadata.get("dataset_name"),
+                document_id=item.metadata.get("document_id") or item.metadata.get("title"),
+                document_name=item.metadata.get("title"),
+                data_source_type="external",
+                retriever_from=invoke_from.to_source(),
+                score=item.metadata.get("score"),
+                content=item.page_content,
+            )
             retrieval_resource_list.append(source)
         # deal with dify documents
         if dify_documents:
@@ -248,32 +249,32 @@ class DatasetRetrieval:
                             .first()
                         )
                         if dataset and document:
-                            source = {
-                                "dataset_id": dataset.id,
-                                "dataset_name": dataset.name,
-                                "document_id": document.id,
-                                "document_name": document.name,
-                                "data_source_type": document.data_source_type,
-                                "segment_id": segment.id,
-                                "retriever_from": invoke_from.to_source(),
-                                "score": record.score or 0.0,
-                                "doc_metadata": document.doc_metadata,
-                            }
+                            source = RetrievalSourceMetadata(
+                                dataset_id=dataset.id,
+                                dataset_name=dataset.name,
+                                document_id=document.id,
+                                document_name=document.name,
+                                data_source_type=document.data_source_type,
+                                segment_id=segment.id,
+                                retriever_from=invoke_from.to_source(),
+                                score=record.score or 0.0,
+                                doc_metadata=document.doc_metadata,
+                            )
 
                             if invoke_from.to_source() == "dev":
-                                source["hit_count"] = segment.hit_count
-                                source["word_count"] = segment.word_count
-                                source["segment_position"] = segment.position
-                                source["index_node_hash"] = segment.index_node_hash
+                                source.hit_count = segment.hit_count
+                                source.word_count = segment.word_count
+                                source.segment_position = segment.position
+                                source.index_node_hash = segment.index_node_hash
                             if segment.answer:
-                                source["content"] = f"question:{segment.content} \nanswer:{segment.answer}"
+                                source.content = f"question:{segment.content} \nanswer:{segment.answer}"
                             else:
-                                source["content"] = segment.content
+                                source.content = segment.content
                             retrieval_resource_list.append(source)
         if hit_callback and retrieval_resource_list:
-            retrieval_resource_list = sorted(retrieval_resource_list, key=lambda x: x.get("score") or 0.0, reverse=True)
+            retrieval_resource_list = sorted(retrieval_resource_list, key=lambda x: x.score or 0.0, reverse=True)
             for position, item in enumerate(retrieval_resource_list, start=1):
-                item["position"] = position
+                item.position = position
             hit_callback.return_retriever_resource_info(retrieval_resource_list)
         if document_context_list:
             document_context_list = sorted(document_context_list, key=lambda x: x.score or 0.0, reverse=True)
@@ -936,6 +937,9 @@ class DatasetRetrieval:
         return metadata_filter_document_ids, metadata_condition
 
     def _replace_metadata_filter_value(self, text: str, inputs: dict) -> str:
+        if not inputs:
+            return text
+
         def replacer(match):
             key = match.group(1)
             return str(inputs.get(key, f"{{{{{key}}}}}"))

api/core/rag/retrieval/router/multi_dataset_react_route.py

@@ -9,7 +9,7 @@ from core.prompt.advanced_prompt_transform import AdvancedPromptTransform
 from core.prompt.entities.advanced_prompt_entities import ChatModelMessage, CompletionModelPromptTemplate
 from core.rag.retrieval.output_parser.react_output import ReactAction
 from core.rag.retrieval.output_parser.structured_chat import StructuredChatOutputParser
-from core.workflow.nodes.llm import LLMNode
+from core.workflow.nodes.llm import llm_utils
 
 PREFIX = """Respond to the human as helpfully and accurately as possible. You have access to the following tools:"""
 
@@ -165,7 +165,7 @@ class ReactMultiDatasetRouter:
         text, usage = self._handle_invoke_result(invoke_result=invoke_result)
 
         # deduct quota
-        LLMNode.deduct_llm_quota(tenant_id=tenant_id, model_instance=model_instance, usage=usage)
+        llm_utils.deduct_llm_quota(tenant_id=tenant_id, model_instance=model_instance, usage=usage)
 
         return text, usage
 

api/core/tools/builtin_tool/_position.yaml

@@ -1,3 +1,4 @@
+- audio
 - code
 - time
-- qrcode
+- webscraper

api/core/tools/custom_tool/tool.py

@@ -168,7 +168,7 @@ class ApiTool(Tool):
                 cookies[parameter["name"]] = value
 
             elif parameter["in"] == "header":
-                headers[parameter["name"]] = value
+                headers[parameter["name"]] = str(value)
 
         # check if there is a request body and handle it
         if "requestBody" in self.api_bundle.openapi and self.api_bundle.openapi["requestBody"] is not None:

api/core/tools/entities/tool_entities.py

@@ -279,7 +279,6 @@ class ToolParameter(PluginParameter):
         :param options: the options of the parameter
         """
         # convert options to ToolParameterOption
-        # FIXME fix the type error
         if options:
             option_objs = [
                 PluginParameterOption(value=option, label=I18nObject(en_US=option, zh_Hans=option))

api/core/tools/utils/dataset_retriever/dataset_multi_retriever_tool.py

@@ -8,6 +8,7 @@ from core.callback_handler.index_tool_callback_handler import DatasetIndexToolCa
 from core.model_manager import ModelManager
 from core.model_runtime.entities.model_entities import ModelType
 from core.rag.datasource.retrieval_service import RetrievalService
+from core.rag.entities.citation_metadata import RetrievalSourceMetadata
 from core.rag.models.document import Document as RagDocument
 from core.rag.rerank.rerank_model import RerankModelRunner
 from core.rag.retrieval.retrieval_methods import RetrievalMethod
@@ -107,7 +108,7 @@ class DatasetMultiRetrieverTool(DatasetRetrieverBaseTool):
                 else:
                     document_context_list.append(segment.get_sign_content())
             if self.return_resource:
-                context_list = []
+                context_list: list[RetrievalSourceMetadata] = []
                 resource_number = 1
                 for segment in sorted_segments:
                     dataset = db.session.query(Dataset).filter_by(id=segment.dataset_id).first()
@@ -121,28 +122,28 @@ class DatasetMultiRetrieverTool(DatasetRetrieverBaseTool):
                         .first()
                     )
                     if dataset and document:
-                        source = {
-                            "position": resource_number,
-                            "dataset_id": dataset.id,
-                            "dataset_name": dataset.name,
-                            "document_id": document.id,
-                            "document_name": document.name,
-                            "data_source_type": document.data_source_type,
-                            "segment_id": segment.id,
-                            "retriever_from": self.retriever_from,
-                            "score": document_score_list.get(segment.index_node_id, None),
-                            "doc_metadata": document.doc_metadata,
-                        }
+                        source = RetrievalSourceMetadata(
+                            position=resource_number,
+                            dataset_id=dataset.id,
+                            dataset_name=dataset.name,
+                            document_id=document.id,
+                            document_name=document.name,
+                            data_source_type=document.data_source_type,
+                            segment_id=segment.id,
+                            retriever_from=self.retriever_from,
+                            score=document_score_list.get(segment.index_node_id, None),
+                            doc_metadata=document.doc_metadata,
+                        )
 
                         if self.retriever_from == "dev":
-                            source["hit_count"] = segment.hit_count
-                            source["word_count"] = segment.word_count
-                            source["segment_position"] = segment.position
-                            source["index_node_hash"] = segment.index_node_hash
+                            source.hit_count = segment.hit_count
+                            source.word_count = segment.word_count
+                            source.segment_position = segment.position
+                            source.index_node_hash = segment.index_node_hash
                         if segment.answer:
-                            source["content"] = f"question:{segment.content} \nanswer:{segment.answer}"
+                            source.content = f"question:{segment.content} \nanswer:{segment.answer}"
                         else:
-                            source["content"] = segment.content
+                            source.content = segment.content
                         context_list.append(source)
                     resource_number += 1
 
@@ -152,8 +153,6 @@ class DatasetMultiRetrieverTool(DatasetRetrieverBaseTool):
             return str("\n".join(document_context_list))
         return ""
 
-        raise RuntimeError("not segments found")
-
     def _retriever(
         self,
         flask_app: Flask,

api/core/tools/utils/dataset_retriever/dataset_retriever_tool.py

@@ -4,6 +4,7 @@ from pydantic import BaseModel, Field
 
 from core.app.app_config.entities import DatasetRetrieveConfigEntity, ModelConfig
 from core.rag.datasource.retrieval_service import RetrievalService
+from core.rag.entities.citation_metadata import RetrievalSourceMetadata
 from core.rag.entities.context_entities import DocumentContext
 from core.rag.models.document import Document as RetrievalDocument
 from core.rag.retrieval.dataset_retrieval import DatasetRetrieval
@@ -14,7 +15,7 @@ from models.dataset import Dataset
 from models.dataset import Document as DatasetDocument
 from services.external_knowledge_service import ExternalDatasetService
 
-default_retrieval_model = {
+default_retrieval_model: dict[str, Any] = {
     "search_method": RetrievalMethod.SEMANTIC_SEARCH.value,
     "reranking_enable": False,
     "reranking_model": {"reranking_provider_name": "", "reranking_model_name": ""},
@@ -79,7 +80,7 @@ class DatasetRetrieverTool(DatasetRetrieverBaseTool):
         else:
             document_ids_filter = None
         if dataset.provider == "external":
-            results = []
+            results: list[RetrievalDocument] = []
             external_documents = ExternalDatasetService.fetch_external_knowledge_retrieval(
                 tenant_id=dataset.tenant_id,
                 dataset_id=dataset.id,
@@ -100,21 +101,21 @@ class DatasetRetrieverTool(DatasetRetrieverBaseTool):
                     document.metadata["dataset_name"] = dataset.name
                     results.append(document)
             # deal with external documents
-            context_list = []
+            context_list: list[RetrievalSourceMetadata] = []
             for position, item in enumerate(results, start=1):
                 if item.metadata is not None:
-                    source = {
-                        "position": position,
-                        "dataset_id": item.metadata.get("dataset_id"),
-                        "dataset_name": item.metadata.get("dataset_name"),
-                        "document_id": item.metadata.get("document_id") or item.metadata.get("title"),
-                        "document_name": item.metadata.get("title"),
-                        "data_source_type": "external",
-                        "retriever_from": self.retriever_from,
-                        "score": item.metadata.get("score"),
-                        "title": item.metadata.get("title"),
-                        "content": item.page_content,
-                    }
+                    source = RetrievalSourceMetadata(
+                        position=position,
+                        dataset_id=item.metadata.get("dataset_id"),
+                        dataset_name=item.metadata.get("dataset_name"),
+                        document_id=item.metadata.get("document_id") or item.metadata.get("title"),
+                        document_name=item.metadata.get("title"),
+                        data_source_type="external",
+                        retriever_from=self.retriever_from,
+                        score=item.metadata.get("score"),
+                        title=item.metadata.get("title"),
+                        content=item.page_content,
+                    )
                     context_list.append(source)
             for hit_callback in self.hit_callbacks:
                 hit_callback.return_retriever_resource_info(context_list)
@@ -125,7 +126,7 @@ class DatasetRetrieverTool(DatasetRetrieverBaseTool):
                 return ""
             # get retrieval model , if the model is not setting , using default
             retrieval_model: dict[str, Any] = dataset.retrieval_model or default_retrieval_model
-            retrieval_resource_list = []
+            retrieval_resource_list: list[RetrievalSourceMetadata] = []
             if dataset.indexing_technique == "economy":
                 # use keyword table query
                 documents = RetrievalService.retrieve(
@@ -163,7 +164,7 @@ class DatasetRetrieverTool(DatasetRetrieverBaseTool):
                     for item in documents:
                         if item.metadata is not None and item.metadata.get("score"):
                             document_score_list[item.metadata["doc_id"]] = item.metadata["score"]
-                document_context_list = []
+                document_context_list: list[DocumentContext] = []
                 records = RetrievalService.format_retrieval_documents(documents)
                 if records:
                     for record in records:
@@ -197,37 +198,37 @@ class DatasetRetrieverTool(DatasetRetrieverBaseTool):
                                 .first()
                             )
                             if dataset and document:
-                                source = {
-                                    "dataset_id": dataset.id,
-                                    "dataset_name": dataset.name,
-                                    "document_id": document.id,  # type: ignore
-                                    "document_name": document.name,  # type: ignore
-                                    "data_source_type": document.data_source_type,  # type: ignore
-                                    "segment_id": segment.id,
-                                    "retriever_from": self.retriever_from,
-                                    "score": record.score or 0.0,
-                                    "doc_metadata": document.doc_metadata,  # type: ignore
-                                }
+                                source = RetrievalSourceMetadata(
+                                    dataset_id=dataset.id,
+                                    dataset_name=dataset.name,
+                                    document_id=document.id,  # type: ignore
+                                    document_name=document.name,  # type: ignore
+                                    data_source_type=document.data_source_type,  # type: ignore
+                                    segment_id=segment.id,
+                                    retriever_from=self.retriever_from,
+                                    score=record.score or 0.0,
+                                    doc_metadata=document.doc_metadata,  # type: ignore
+                                )
 
                                 if self.retriever_from == "dev":
-                                    source["hit_count"] = segment.hit_count
-                                    source["word_count"] = segment.word_count
-                                    source["segment_position"] = segment.position
-                                    source["index_node_hash"] = segment.index_node_hash
+                                    source.hit_count = segment.hit_count
+                                    source.word_count = segment.word_count
+                                    source.segment_position = segment.position
+                                    source.index_node_hash = segment.index_node_hash
                                 if segment.answer:
-                                    source["content"] = f"question:{segment.content} \nanswer:{segment.answer}"
+                                    source.content = f"question:{segment.content} \nanswer:{segment.answer}"
                                 else:
-                                    source["content"] = segment.content
+                                    source.content = segment.content
                                 retrieval_resource_list.append(source)
 
             if self.return_resource and retrieval_resource_list:
                 retrieval_resource_list = sorted(
                     retrieval_resource_list,
-                    key=lambda x: x.get("score") or 0.0,
+                    key=lambda x: x.score or 0.0,
                     reverse=True,
                 )
                 for position, item in enumerate(retrieval_resource_list, start=1):  # type: ignore
-                    item["position"] = position  # type: ignore
+                    item.position = position  # type: ignore
                 for hit_callback in self.hit_callbacks:
                     hit_callback.return_retriever_resource_info(retrieval_resource_list)
             if document_context_list:

api/core/tools/utils/message_transformer.py

@@ -32,14 +32,14 @@ class ToolFileMessageTransformer:
                 try:
                     assert isinstance(message.message, ToolInvokeMessage.TextMessage)
                     tool_file_manager = ToolFileManager()
-                    file = tool_file_manager.create_file_by_url(
+                    tool_file = tool_file_manager.create_file_by_url(
                         user_id=user_id,
                         tenant_id=tenant_id,
                         file_url=message.message.text,
                         conversation_id=conversation_id,
                     )
 
-                    url = f"/files/tools/{file.id}{guess_extension(file.mimetype) or '.png'}"
+                    url = f"/files/tools/{tool_file.id}{guess_extension(tool_file.mimetype) or '.png'}"
 
                     yield ToolInvokeMessage(
                         type=ToolInvokeMessage.MessageType.IMAGE_LINK,
@@ -66,10 +66,9 @@ class ToolFileMessageTransformer:
                 if not isinstance(message.message, ToolInvokeMessage.BlobMessage):
                     raise ValueError("unexpected message type")
 
-                # FIXME: should do a type check here.
                 assert isinstance(message.message.blob, bytes)
                 tool_file_manager = ToolFileManager()
-                file = tool_file_manager.create_file_by_raw(
+                tool_file = tool_file_manager.create_file_by_raw(
                     user_id=user_id,
                     tenant_id=tenant_id,
                     conversation_id=conversation_id,
@@ -78,7 +77,7 @@ class ToolFileMessageTransformer:
                     filename=filename,
                 )
 
-                url = cls.get_tool_file_url(tool_file_id=file.id, extension=guess_extension(file.mimetype))
+                url = cls.get_tool_file_url(tool_file_id=tool_file.id, extension=guess_extension(tool_file.mimetype))
 
                 # check if file is image
                 if "image" in mimetype:

api/core/tools/utils/parser.py

@@ -55,6 +55,13 @@ class ApiBasedToolSchemaParser:
             # convert parameters
             parameters = []
             if "parameters" in interface["operation"]:
+                for i, parameter in enumerate(interface["operation"]["parameters"]):
+                    if "$ref" in parameter:
+                        root = openapi
+                        reference = parameter["$ref"].split("/")[1:]
+                        for ref in reference:
+                            root = root[ref]
+                        interface["operation"]["parameters"][i] = root
                 for parameter in interface["operation"]["parameters"]:
                     tool_parameter = ToolParameter(
                         name=parameter["name"],

api/core/workflow/graph_engine/entities/event.py

@@ -1,9 +1,10 @@
-from collections.abc import Mapping
+from collections.abc import Mapping, Sequence
 from datetime import datetime
 from typing import Any, Optional
 
 from pydantic import BaseModel, Field
 
+from core.rag.entities.citation_metadata import RetrievalSourceMetadata
 from core.workflow.entities.node_entities import AgentNodeStrategyInit
 from core.workflow.graph_engine.entities.runtime_route_state import RouteNodeState
 from core.workflow.nodes import NodeType
@@ -82,7 +83,7 @@ class NodeRunStreamChunkEvent(BaseNodeEvent):
 
 
 class NodeRunRetrieverResourceEvent(BaseNodeEvent):
-    retriever_resources: list[dict] = Field(..., description="retriever resources")
+    retriever_resources: Sequence[RetrievalSourceMetadata] = Field(..., description="retriever resources")
     context: str = Field(..., description="context")
 
 

api/core/workflow/graph_engine/graph_engine.py

@@ -53,7 +53,6 @@ from core.workflow.nodes.end.end_stream_processor import EndStreamProcessor
 from core.workflow.nodes.enums import ErrorStrategy, FailBranchSourceHandle
 from core.workflow.nodes.event import RunCompletedEvent, RunRetrieverResourceEvent, RunStreamChunkEvent
 from core.workflow.nodes.node_mapping import NODE_TYPE_CLASSES_MAPPING
-from extensions.ext_database import db
 from models.enums import UserFrom
 from models.workflow import WorkflowType
 
@@ -607,8 +606,6 @@ class GraphEngine:
                         error=str(e),
                     )
                 )
-            finally:
-                db.session.remove()
 
     def _run_node(
         self,
@@ -646,7 +643,6 @@ class GraphEngine:
             agent_strategy=agent_strategy,
         )
 
-        db.session.close()
         max_retries = node_instance.node_data.retry_config.max_retries
         retry_interval = node_instance.node_data.retry_config.retry_interval_seconds
         retries = 0
@@ -863,8 +859,6 @@ class GraphEngine:
             except Exception as e:
                 logger.exception(f"Node {node_instance.node_data.title} run failed")
                 raise e
-            finally:
-                db.session.close()
 
     def _append_variables_recursively(self, node_id: str, variable_key_list: list[str], variable_value: VariableValue):
         """

api/core/workflow/nodes/agent/agent_node.py

@@ -2,6 +2,9 @@ import json
 from collections.abc import Generator, Mapping, Sequence
 from typing import Any, Optional, cast
 
+from sqlalchemy import select
+from sqlalchemy.orm import Session
+
 from core.agent.entities import AgentToolEntity
 from core.agent.plugin_entities import AgentStrategyParameter
 from core.memory.token_buffer_memory import TokenBufferMemory
@@ -320,15 +323,12 @@ class AgentNode(ToolNode):
             return None
         conversation_id = conversation_id_variable.value
 
-        # get conversation
-        conversation = (
-            db.session.query(Conversation)
-            .filter(Conversation.app_id == self.app_id, Conversation.id == conversation_id)
-            .first()
-        )
+        with Session(db.engine, expire_on_commit=False) as session:
+            stmt = select(Conversation).where(Conversation.app_id == self.app_id, Conversation.id == conversation_id)
+            conversation = session.scalar(stmt)
 
-        if not conversation:
-            return None
+            if not conversation:
+                return None
 
         memory = TokenBufferMemory(conversation=conversation, model_instance=model_instance)
 

api/core/workflow/nodes/document_extractor/node.py

@@ -397,19 +397,44 @@ def _extract_text_from_csv(file_content: bytes) -> str:
         if not rows:
             return ""
 
+        # Combine multi-line text in the header row
+        header_row = [cell.replace("\n", " ").replace("\r", "") for cell in rows[0]]
+
         # Create Markdown table
-        markdown_table = "| " + " | ".join(rows[0]) + " |\n"
-        markdown_table += "| " + " | ".join(["---"] * len(rows[0])) + " |\n"
+        markdown_table = "| " + " | ".join(header_row) + " |\n"
+        markdown_table += "| " + " | ".join(["-" * len(col) for col in rows[0]]) + " |\n"
+
+        # Process each data row and combine multi-line text in each cell
         for row in rows[1:]:
-            markdown_table += "| " + " | ".join(row) + " |\n"
+            processed_row = [cell.replace("\n", " ").replace("\r", "") for cell in row]
+            markdown_table += "| " + " | ".join(processed_row) + " |\n"
 
-        return markdown_table.strip()
+        return markdown_table
     except Exception as e:
         raise TextExtractionError(f"Failed to extract text from CSV: {str(e)}") from e
 
 
 def _extract_text_from_excel(file_content: bytes) -> str:
     """Extract text from an Excel file using pandas."""
+
+    def _construct_markdown_table(df: pd.DataFrame) -> str:
+        """Manually construct a Markdown table from a DataFrame."""
+        # Construct the header row
+        header_row = "| " + " | ".join(df.columns) + " |"
+
+        # Construct the separator row
+        separator_row = "| " + " | ".join(["-" * len(col) for col in df.columns]) + " |"
+
+        # Construct the data rows
+        data_rows = []
+        for _, row in df.iterrows():
+            data_row = "| " + " | ".join(map(str, row)) + " |"
+            data_rows.append(data_row)
+
+        # Combine all rows into a single string
+        markdown_table = "\n".join([header_row, separator_row] + data_rows)
+        return markdown_table
+
     try:
         excel_file = pd.ExcelFile(io.BytesIO(file_content))
         markdown_table = ""
@@ -417,8 +442,15 @@ def _extract_text_from_excel(file_content: bytes) -> str:
             try:
                 df = excel_file.parse(sheet_name=sheet_name)
                 df.dropna(how="all", inplace=True)
-                # Create Markdown table two times to separate tables with a newline
-                markdown_table += df.to_markdown(index=False, floatfmt="") + "\n\n"
+
+                # Combine multi-line text in each cell into a single line
+                df = df.applymap(lambda x: " ".join(str(x).splitlines()) if isinstance(x, str) else x)  # type: ignore
+
+                # Combine multi-line text in column names into a single line
+                df.columns = pd.Index([" ".join(col.splitlines()) for col in df.columns])
+
+                # Manually construct the Markdown table
+                markdown_table += _construct_markdown_table(df) + "\n\n"
             except Exception as e:
                 continue
         return markdown_table

api/core/workflow/nodes/event/event.py

@@ -1,8 +1,10 @@
+from collections.abc import Sequence
 from datetime import datetime
 
 from pydantic import BaseModel, Field
 
 from core.model_runtime.entities.llm_entities import LLMUsage
+from core.rag.entities.citation_metadata import RetrievalSourceMetadata
 from core.workflow.entities.node_entities import NodeRunResult
 from core.workflow.entities.workflow_node_execution import WorkflowNodeExecutionStatus
 
@@ -17,7 +19,7 @@ class RunStreamChunkEvent(BaseModel):
 
 
 class RunRetrieverResourceEvent(BaseModel):
-    retriever_resources: list[dict] = Field(..., description="retriever resources")
+    retriever_resources: Sequence[RetrievalSourceMetadata] = Field(..., description="retriever resources")
     context: str = Field(..., description="context")
 
 

api/core/workflow/nodes/http_request/executor.py

@@ -1,8 +1,9 @@
 import base64
 import json
+import secrets
+import string
 from collections.abc import Mapping
 from copy import deepcopy
-from random import randint
 from typing import Any, Literal
 from urllib.parse import urlencode, urlparse
 
@@ -434,4 +435,4 @@ def _generate_random_string(n: int) -> str:
         >>> _generate_random_string(5)
         'abcde'
     """
-    return "".join([chr(randint(97, 122)) for _ in range(n)])
+    return "".join(secrets.choice(string.ascii_lowercase) for _ in range(n))

api/core/workflow/nodes/knowledge_retrieval/entities.py

@@ -132,3 +132,12 @@ class KnowledgeRetrievalNodeData(BaseNodeData):
     metadata_model_config: Optional[ModelConfig] = None
     metadata_filtering_conditions: Optional[MetadataFilteringCondition] = None
     vision: VisionConfig = Field(default_factory=VisionConfig)
+
+    @property
+    def structured_output_enabled(self) -> bool:
+        # NOTE(QuantumGhost): Temporary workaround for issue #20725
+        # (https://github.com/langgenius/dify/issues/20725).
+        #
+        # The proper fix would be to make `KnowledgeRetrievalNode` inherit
+        # from `BaseNode` instead of `LLMNode`.
+        return False

api/core/workflow/nodes/knowledge_retrieval/knowledge_retrieval_node.py

@@ -8,6 +8,7 @@ from typing import Any, Optional, cast
 
 from sqlalchemy import Float, and_, func, or_, text
 from sqlalchemy import cast as sqlalchemy_cast
+from sqlalchemy.orm import Session
 
 from core.app.app_config.entities import DatasetRetrieveConfigEntity
 from core.app.entities.app_invoke_entities import ModelConfigWithCredentialsEntity
@@ -85,30 +86,31 @@ class KnowledgeRetrievalNode(LLMNode):
             return NodeRunResult(
                 status=WorkflowNodeExecutionStatus.FAILED, inputs=variables, error="Query is required."
             )
+        # TODO(-LAN-): Move this check outside.
         # check rate limit
-        if self.tenant_id:
-            knowledge_rate_limit = FeatureService.get_knowledge_rate_limit(self.tenant_id)
-            if knowledge_rate_limit.enabled:
-                current_time = int(time.time() * 1000)
-                key = f"rate_limit_{self.tenant_id}"
-                redis_client.zadd(key, {current_time: current_time})
-                redis_client.zremrangebyscore(key, 0, current_time - 60000)
-                request_count = redis_client.zcard(key)
-                if request_count > knowledge_rate_limit.limit:
+        knowledge_rate_limit = FeatureService.get_knowledge_rate_limit(self.tenant_id)
+        if knowledge_rate_limit.enabled:
+            current_time = int(time.time() * 1000)
+            key = f"rate_limit_{self.tenant_id}"
+            redis_client.zadd(key, {current_time: current_time})
+            redis_client.zremrangebyscore(key, 0, current_time - 60000)
+            request_count = redis_client.zcard(key)
+            if request_count > knowledge_rate_limit.limit:
+                with Session(db.engine) as session:
                     # add ratelimit record
                     rate_limit_log = RateLimitLog(
                         tenant_id=self.tenant_id,
                         subscription_plan=knowledge_rate_limit.subscription_plan,
                         operation="knowledge",
                     )
-                    db.session.add(rate_limit_log)
-                    db.session.commit()
-                    return NodeRunResult(
-                        status=WorkflowNodeExecutionStatus.FAILED,
-                        inputs=variables,
-                        error="Sorry, you have reached the knowledge base request rate limit of your subscription.",
-                        error_type="RateLimitExceeded",
-                    )
+                    session.add(rate_limit_log)
+                    session.commit()
+                return NodeRunResult(
+                    status=WorkflowNodeExecutionStatus.FAILED,
+                    inputs=variables,
+                    error="Sorry, you have reached the knowledge base request rate limit of your subscription.",
+                    error_type="RateLimitExceeded",
+                )
 
         # retrieve knowledge
         try:
@@ -173,7 +175,9 @@ class KnowledgeRetrievalNode(LLMNode):
         dataset_retrieval = DatasetRetrieval()
         if node_data.retrieval_mode == DatasetRetrieveConfigEntity.RetrieveStrategy.SINGLE.value:
             # fetch model config
-            model_instance, model_config = self._fetch_model_config(node_data.single_retrieval_config.model)  # type: ignore
+            if node_data.single_retrieval_config is None:
+                raise ValueError("single_retrieval_config is required")
+            model_instance, model_config = self.get_model_config(node_data.single_retrieval_config.model)
             # check model is support tool calling
             model_type_instance = model_config.provider_model_bundle.model_type_instance
             model_type_instance = cast(LargeLanguageModel, model_type_instance)
@@ -424,7 +428,7 @@ class KnowledgeRetrievalNode(LLMNode):
             raise ValueError("metadata_model_config is required")
         # get metadata model instance
         # fetch model config
-        model_instance, model_config = self._fetch_model_config(node_data.metadata_model_config)  # type: ignore
+        model_instance, model_config = self.get_model_config(metadata_model_config)
         # fetch prompt messages
         prompt_template = self._get_prompt_template(
             node_data=node_data,
@@ -550,14 +554,7 @@ class KnowledgeRetrievalNode(LLMNode):
         variable_mapping[node_id + ".query"] = node_data.query_variable_selector
         return variable_mapping
 
-    def _fetch_model_config(self, model: ModelConfig) -> tuple[ModelInstance, ModelConfigWithCredentialsEntity]:  # type: ignore
-        """
-        Fetch model config
-        :param model: model
-        :return:
-        """
-        if model is None:
-            raise ValueError("model is required")
+    def get_model_config(self, model: ModelConfig) -> tuple[ModelInstance, ModelConfigWithCredentialsEntity]:
         model_name = model.name
         provider_name = model.provider
 

api/core/workflow/nodes/llm/entities.py

@@ -66,7 +66,8 @@ class LLMNodeData(BaseNodeData):
     context: ContextConfig
     vision: VisionConfig = Field(default_factory=VisionConfig)
     structured_output: dict | None = None
-    structured_output_enabled: bool = False
+    # We used 'structured_output_enabled' in the past, but it's not a good name.
+    structured_output_switch_on: bool = Field(False, alias="structured_output_enabled")
 
     @field_validator("prompt_config", mode="before")
     @classmethod
@@ -74,3 +75,7 @@ class LLMNodeData(BaseNodeData):
         if v is None:
             return PromptConfig()
         return v
+
+    @property
+    def structured_output_enabled(self) -> bool:
+        return self.structured_output_switch_on and self.structured_output is not None

api/core/workflow/nodes/llm/llm_utils.py

@@ -0,0 +1,156 @@
+from collections.abc import Sequence
+from datetime import UTC, datetime
+from typing import Optional, cast
+
+from sqlalchemy import select, update
+from sqlalchemy.orm import Session
+
+from configs import dify_config
+from core.app.entities.app_invoke_entities import ModelConfigWithCredentialsEntity
+from core.entities.provider_entities import QuotaUnit
+from core.file.models import File
+from core.memory.token_buffer_memory import TokenBufferMemory
+from core.model_manager import ModelInstance, ModelManager
+from core.model_runtime.entities.llm_entities import LLMUsage
+from core.model_runtime.entities.model_entities import ModelType
+from core.model_runtime.model_providers.__base.large_language_model import LargeLanguageModel
+from core.plugin.entities.plugin import ModelProviderID
+from core.prompt.entities.advanced_prompt_entities import MemoryConfig
+from core.variables.segments import ArrayAnySegment, ArrayFileSegment, FileSegment, NoneSegment, StringSegment
+from core.workflow.entities.variable_pool import VariablePool
+from core.workflow.enums import SystemVariableKey
+from core.workflow.nodes.llm.entities import ModelConfig
+from models import db
+from models.model import Conversation
+from models.provider import Provider, ProviderType
+
+from .exc import InvalidVariableTypeError, LLMModeRequiredError, ModelNotExistError
+
+
+def fetch_model_config(
+    tenant_id: str, node_data_model: ModelConfig
+) -> tuple[ModelInstance, ModelConfigWithCredentialsEntity]:
+    if not node_data_model.mode:
+        raise LLMModeRequiredError("LLM mode is required.")
+
+    model = ModelManager().get_model_instance(
+        tenant_id=tenant_id,
+        model_type=ModelType.LLM,
+        provider=node_data_model.provider,
+        model=node_data_model.name,
+    )
+
+    model.model_type_instance = cast(LargeLanguageModel, model.model_type_instance)
+
+    # check model
+    provider_model = model.provider_model_bundle.configuration.get_provider_model(
+        model=node_data_model.name, model_type=ModelType.LLM
+    )
+
+    if provider_model is None:
+        raise ModelNotExistError(f"Model {node_data_model.name} not exist.")
+    provider_model.raise_for_status()
+
+    # model config
+    stop: list[str] = []
+    if "stop" in node_data_model.completion_params:
+        stop = node_data_model.completion_params.pop("stop")
+
+    model_schema = model.model_type_instance.get_model_schema(node_data_model.name, model.credentials)
+    if not model_schema:
+        raise ModelNotExistError(f"Model {node_data_model.name} not exist.")
+
+    return model, ModelConfigWithCredentialsEntity(
+        provider=node_data_model.provider,
+        model=node_data_model.name,
+        model_schema=model_schema,
+        mode=node_data_model.mode,
+        provider_model_bundle=model.provider_model_bundle,
+        credentials=model.credentials,
+        parameters=node_data_model.completion_params,
+        stop=stop,
+    )
+
+
+def fetch_files(variable_pool: VariablePool, selector: Sequence[str]) -> Sequence["File"]:
+    variable = variable_pool.get(selector)
+    if variable is None:
+        return []
+    elif isinstance(variable, FileSegment):
+        return [variable.value]
+    elif isinstance(variable, ArrayFileSegment):
+        return variable.value
+    elif isinstance(variable, NoneSegment | ArrayAnySegment):
+        return []
+    raise InvalidVariableTypeError(f"Invalid variable type: {type(variable)}")
+
+
+def fetch_memory(
+    variable_pool: VariablePool, app_id: str, node_data_memory: Optional[MemoryConfig], model_instance: ModelInstance
+) -> Optional[TokenBufferMemory]:
+    if not node_data_memory:
+        return None
+
+    # get conversation id
+    conversation_id_variable = variable_pool.get(["sys", SystemVariableKey.CONVERSATION_ID.value])
+    if not isinstance(conversation_id_variable, StringSegment):
+        return None
+    conversation_id = conversation_id_variable.value
+
+    with Session(db.engine, expire_on_commit=False) as session:
+        stmt = select(Conversation).where(Conversation.app_id == app_id, Conversation.id == conversation_id)
+        conversation = session.scalar(stmt)
+        if not conversation:
+            return None
+
+    memory = TokenBufferMemory(conversation=conversation, model_instance=model_instance)
+    return memory
+
+
+def deduct_llm_quota(tenant_id: str, model_instance: ModelInstance, usage: LLMUsage) -> None:
+    provider_model_bundle = model_instance.provider_model_bundle
+    provider_configuration = provider_model_bundle.configuration
+
+    if provider_configuration.using_provider_type != ProviderType.SYSTEM:
+        return
+
+    system_configuration = provider_configuration.system_configuration
+
+    quota_unit = None
+    for quota_configuration in system_configuration.quota_configurations:
+        if quota_configuration.quota_type == system_configuration.current_quota_type:
+            quota_unit = quota_configuration.quota_unit
+
+            if quota_configuration.quota_limit == -1:
+                return
+
+            break
+
+    used_quota = None
+    if quota_unit:
+        if quota_unit == QuotaUnit.TOKENS:
+            used_quota = usage.total_tokens
+        elif quota_unit == QuotaUnit.CREDITS:
+            used_quota = dify_config.get_model_credits(model_instance.model)
+        else:
+            used_quota = 1
+
+    if used_quota is not None and system_configuration.current_quota_type is not None:
+        with Session(db.engine) as session:
+            stmt = (
+                update(Provider)
+                .where(
+                    Provider.tenant_id == tenant_id,
+                    # TODO: Use provider name with prefix after the data migration.
+                    Provider.provider_name == ModelProviderID(model_instance.provider).provider_name,
+                    Provider.provider_type == ProviderType.SYSTEM.value,
+                    Provider.quota_type == system_configuration.current_quota_type.value,
+                    Provider.quota_limit > Provider.quota_used,
+                )
+                .values(
+                    quota_used=Provider.quota_used + used_quota,
+                    last_used=datetime.now(tz=UTC).replace(tzinfo=None),
+                )
+            )
+            session.execute(stmt)
+            session.commit()

api/core/workflow/nodes/llm/node.py

@@ -3,16 +3,11 @@ import io
 import json
 import logging
 from collections.abc import Generator, Mapping, Sequence
-from datetime import UTC, datetime
 from typing import TYPE_CHECKING, Any, Optional, cast
 
 import json_repair
 
-from configs import dify_config
 from core.app.entities.app_invoke_entities import ModelConfigWithCredentialsEntity
-from core.entities.model_entities import ModelStatus
-from core.entities.provider_entities import QuotaUnit
-from core.errors.error import ModelCurrentlyNotSupportError, ProviderTokenNotInitError, QuotaExceededError
 from core.file import FileType, file_manager
 from core.helper.code_executor import CodeExecutor, CodeLanguage
 from core.memory.token_buffer_memory import TokenBufferMemory
@@ -40,11 +35,10 @@ from core.model_runtime.entities.model_entities import (
 )
 from core.model_runtime.model_providers.__base.large_language_model import LargeLanguageModel
 from core.model_runtime.utils.encoders import jsonable_encoder
-from core.plugin.entities.plugin import ModelProviderID
 from core.prompt.entities.advanced_prompt_entities import CompletionModelPromptTemplate, MemoryConfig
 from core.prompt.utils.prompt_message_util import PromptMessageUtil
+from core.rag.entities.citation_metadata import RetrievalSourceMetadata
 from core.variables import (
-    ArrayAnySegment,
     ArrayFileSegment,
     ArraySegment,
     FileSegment,
@@ -71,14 +65,11 @@ from core.workflow.nodes.event import (
 from core.workflow.utils.structured_output.entities import (
     ResponseFormat,
     SpecialModelType,
-    SupportStructuredOutputStatus,
 )
 from core.workflow.utils.structured_output.prompt import STRUCTURED_OUTPUT_PROMPT
 from core.workflow.utils.variable_template_parser import VariableTemplateParser
-from extensions.ext_database import db
-from models.model import Conversation
-from models.provider import Provider, ProviderType
 
+from . import llm_utils
 from .entities import (
     LLMNodeChatModelMessage,
     LLMNodeCompletionModelPromptTemplate,
@@ -88,7 +79,6 @@ from .entities import (
 from .exc import (
     InvalidContextStructureError,
     InvalidVariableTypeError,
-    LLMModeRequiredError,
     LLMNodeError,
     MemoryRolePrefixRequiredError,
     ModelNotExistError,
@@ -160,6 +150,7 @@ class LLMNode(BaseNode[LLMNodeData]):
         result_text = ""
         usage = LLMUsage.empty_usage()
         finish_reason = None
+        variable_pool = self.graph_runtime_state.variable_pool
 
         try:
             # init messages template
@@ -178,7 +169,10 @@ class LLMNode(BaseNode[LLMNodeData]):
 
             # fetch files
             files = (
-                self._fetch_files(selector=self.node_data.vision.configs.variable_selector)
+                llm_utils.fetch_files(
+                    variable_pool=variable_pool,
+                    selector=self.node_data.vision.configs.variable_selector,
+                )
                 if self.node_data.vision.enabled
                 else []
             )
@@ -200,15 +194,18 @@ class LLMNode(BaseNode[LLMNodeData]):
             model_instance, model_config = self._fetch_model_config(self.node_data.model)
 
             # fetch memory
-            memory = self._fetch_memory(node_data_memory=self.node_data.memory, model_instance=model_instance)
+            memory = llm_utils.fetch_memory(
+                variable_pool=variable_pool,
+                app_id=self.app_id,
+                node_data_memory=self.node_data.memory,
+                model_instance=model_instance,
+            )
 
             query = None
             if self.node_data.memory:
                 query = self.node_data.memory.query_prompt_template
                 if not query and (
-                    query_variable := self.graph_runtime_state.variable_pool.get(
-                        (SYSTEM_VARIABLE_NODE_ID, SystemVariableKey.QUERY)
-                    )
+                    query_variable := variable_pool.get((SYSTEM_VARIABLE_NODE_ID, SystemVariableKey.QUERY))
                 ):
                     query = query_variable.text
 
@@ -222,7 +219,7 @@ class LLMNode(BaseNode[LLMNodeData]):
                 memory_config=self.node_data.memory,
                 vision_enabled=self.node_data.vision.enabled,
                 vision_detail=self.node_data.vision.configs.detail,
-                variable_pool=self.graph_runtime_state.variable_pool,
+                variable_pool=variable_pool,
                 jinja2_variables=self.node_data.prompt_config.jinja2_variables,
             )
 
@@ -251,7 +248,7 @@ class LLMNode(BaseNode[LLMNodeData]):
                     usage = event.usage
                     finish_reason = event.finish_reason
                     # deduct quota
-                    self.deduct_llm_quota(tenant_id=self.tenant_id, model_instance=model_instance, usage=usage)
+                    llm_utils.deduct_llm_quota(tenant_id=self.tenant_id, model_instance=model_instance, usage=usage)
                     break
             outputs = {"text": result_text, "usage": jsonable_encoder(usage), "finish_reason": finish_reason}
             structured_output = process_structured_output(result_text)
@@ -274,7 +271,7 @@ class LLMNode(BaseNode[LLMNodeData]):
                     llm_usage=usage,
                 )
             )
-        except LLMNodeError as e:
+        except ValueError as e:
             yield RunCompletedEvent(
                 run_result=NodeRunResult(
                     status=WorkflowNodeExecutionStatus.FAILED,
@@ -302,8 +299,6 @@ class LLMNode(BaseNode[LLMNodeData]):
         prompt_messages: Sequence[PromptMessage],
         stop: Optional[Sequence[str]] = None,
     ) -> Generator[NodeEvent, None, None]:
-        db.session.close()
-
         invoke_result = model_instance.invoke_llm(
             prompt_messages=list(prompt_messages),
             model_parameters=node_data_model.completion_params,
@@ -449,18 +444,6 @@ class LLMNode(BaseNode[LLMNodeData]):
 
         return inputs
 
-    def _fetch_files(self, *, selector: Sequence[str]) -> Sequence["File"]:
-        variable = self.graph_runtime_state.variable_pool.get(selector)
-        if variable is None:
-            return []
-        elif isinstance(variable, FileSegment):
-            return [variable.value]
-        elif isinstance(variable, ArrayFileSegment):
-            return variable.value
-        elif isinstance(variable, NoneSegment | ArrayAnySegment):
-            return []
-        raise InvalidVariableTypeError(f"Invalid variable type: {type(variable)}")
-
     def _fetch_context(self, node_data: LLMNodeData):
         if not node_data.context.enabled:
             return
@@ -474,7 +457,7 @@ class LLMNode(BaseNode[LLMNodeData]):
                 yield RunRetrieverResourceEvent(retriever_resources=[], context=context_value_variable.value)
             elif isinstance(context_value_variable, ArraySegment):
                 context_str = ""
-                original_retriever_resource = []
+                original_retriever_resource: list[RetrievalSourceMetadata] = []
                 for item in context_value_variable.value:
                     if isinstance(item, str):
                         context_str += item + "\n"
@@ -492,7 +475,7 @@ class LLMNode(BaseNode[LLMNodeData]):
                     retriever_resources=original_retriever_resource, context=context_str.strip()
                 )
 
-    def _convert_to_original_retriever_resource(self, context_dict: dict) -> Optional[dict]:
+    def _convert_to_original_retriever_resource(self, context_dict: dict):
         if (
             "metadata" in context_dict
             and "_source" in context_dict["metadata"]
@@ -500,24 +483,24 @@ class LLMNode(BaseNode[LLMNodeData]):
         ):
             metadata = context_dict.get("metadata", {})
 
-            source = {
-                "position": metadata.get("position"),
-                "dataset_id": metadata.get("dataset_id"),
-                "dataset_name": metadata.get("dataset_name"),
-                "document_id": metadata.get("document_id"),
-                "document_name": metadata.get("document_name"),
-                "data_source_type": metadata.get("data_source_type"),
-                "segment_id": metadata.get("segment_id"),
-                "retriever_from": metadata.get("retriever_from"),
-                "score": metadata.get("score"),
-                "hit_count": metadata.get("segment_hit_count"),
-                "word_count": metadata.get("segment_word_count"),
-                "segment_position": metadata.get("segment_position"),
-                "index_node_hash": metadata.get("segment_index_node_hash"),
-                "content": context_dict.get("content"),
-                "page": metadata.get("page"),
-                "doc_metadata": metadata.get("doc_metadata"),
-            }
+            source = RetrievalSourceMetadata(
+                position=metadata.get("position"),
+                dataset_id=metadata.get("dataset_id"),
+                dataset_name=metadata.get("dataset_name"),
+                document_id=metadata.get("document_id"),
+                document_name=metadata.get("document_name"),
+                data_source_type=metadata.get("data_source_type"),
+                segment_id=metadata.get("segment_id"),
+                retriever_from=metadata.get("retriever_from"),
+                score=metadata.get("score"),
+                hit_count=metadata.get("segment_hit_count"),
+                word_count=metadata.get("segment_word_count"),
+                segment_position=metadata.get("segment_position"),
+                index_node_hash=metadata.get("segment_index_node_hash"),
+                content=context_dict.get("content"),
+                page=metadata.get("page"),
+                doc_metadata=metadata.get("doc_metadata"),
+            )
 
             return source
 
@@ -526,95 +509,25 @@ class LLMNode(BaseNode[LLMNodeData]):
     def _fetch_model_config(
         self, node_data_model: ModelConfig
     ) -> tuple[ModelInstance, ModelConfigWithCredentialsEntity]:
-        model_name = node_data_model.name
-        provider_name = node_data_model.provider
-
-        model_manager = ModelManager()
-        model_instance = model_manager.get_model_instance(
-            tenant_id=self.tenant_id, model_type=ModelType.LLM, provider=provider_name, model=model_name
+        model, model_config_with_cred = llm_utils.fetch_model_config(
+            tenant_id=self.tenant_id, node_data_model=node_data_model
         )
+        completion_params = model_config_with_cred.parameters
 
-        provider_model_bundle = model_instance.provider_model_bundle
-        model_type_instance = model_instance.model_type_instance
-        model_type_instance = cast(LargeLanguageModel, model_type_instance)
-
-        model_credentials = model_instance.credentials
-
-        # check model
-        provider_model = provider_model_bundle.configuration.get_provider_model(
-            model=model_name, model_type=ModelType.LLM
-        )
-
-        if provider_model is None:
-            raise ModelNotExistError(f"Model {model_name} not exist.")
-
-        if provider_model.status == ModelStatus.NO_CONFIGURE:
-            raise ProviderTokenNotInitError(f"Model {model_name} credentials is not initialized.")
-        elif provider_model.status == ModelStatus.NO_PERMISSION:
-            raise ModelCurrentlyNotSupportError(f"Dify Hosted OpenAI {model_name} currently not support.")
-        elif provider_model.status == ModelStatus.QUOTA_EXCEEDED:
-            raise QuotaExceededError(f"Model provider {provider_name} quota exceeded.")
-
-        # model config
-        completion_params = node_data_model.completion_params
-        stop = []
-        if "stop" in completion_params:
-            stop = completion_params["stop"]
-            del completion_params["stop"]
-
-        # get model mode
-        model_mode = node_data_model.mode
-        if not model_mode:
-            raise LLMModeRequiredError("LLM mode is required.")
-
-        model_schema = model_type_instance.get_model_schema(model_name, model_credentials)
-
+        model_schema = model.model_type_instance.get_model_schema(node_data_model.name, model.credentials)
         if not model_schema:
-            raise ModelNotExistError(f"Model {model_name} not exist.")
-        support_structured_output = self._check_model_structured_output_support()
-        if support_structured_output == SupportStructuredOutputStatus.SUPPORTED:
-            completion_params = self._handle_native_json_schema(completion_params, model_schema.parameter_rules)
-        elif support_structured_output == SupportStructuredOutputStatus.UNSUPPORTED:
-            # Set appropriate response format based on model capabilities
-            self._set_response_format(completion_params, model_schema.parameter_rules)
-        return model_instance, ModelConfigWithCredentialsEntity(
-            provider=provider_name,
-            model=model_name,
-            model_schema=model_schema,
-            mode=model_mode,
-            provider_model_bundle=provider_model_bundle,
-            credentials=model_credentials,
-            parameters=completion_params,
-            stop=stop,
-        )
-
-    def _fetch_memory(
-        self, node_data_memory: Optional[MemoryConfig], model_instance: ModelInstance
-    ) -> Optional[TokenBufferMemory]:
-        if not node_data_memory:
-            return None
-
-        # get conversation id
-        conversation_id_variable = self.graph_runtime_state.variable_pool.get(
-            ["sys", SystemVariableKey.CONVERSATION_ID.value]
-        )
-        if not isinstance(conversation_id_variable, StringSegment):
-            return None
-        conversation_id = conversation_id_variable.value
-
-        # get conversation
-        conversation = (
-            db.session.query(Conversation)
-            .filter(Conversation.app_id == self.app_id, Conversation.id == conversation_id)
-            .first()
-        )
-
-        if not conversation:
-            return None
+            raise ModelNotExistError(f"Model {node_data_model.name} not exist.")
 
-        memory = TokenBufferMemory(conversation=conversation, model_instance=model_instance)
-
-        return memory
+        if self.node_data.structured_output_enabled:
+            if model_schema.support_structure_output:
+                completion_params = self._handle_native_json_schema(completion_params, model_schema.parameter_rules)
+            else:
+                # Set appropriate response format based on model capabilities
+                self._set_response_format(completion_params, model_schema.parameter_rules)
+        model_config_with_cred.parameters = completion_params
+        # NOTE(-LAN-): This line modify the `self.node_data.model`, which is used in `_invoke_llm()`.
+        node_data_model.completion_params = completion_params
+        return model, model_config_with_cred
 
     def _fetch_prompt_messages(
         self,
@@ -789,13 +702,25 @@ class LLMNode(BaseNode[LLMNodeData]):
                 "No prompt found in the LLM configuration. "
                 "Please ensure a prompt is properly configured before proceeding."
             )
-        support_structured_output = self._check_model_structured_output_support()
-        if support_structured_output == SupportStructuredOutputStatus.UNSUPPORTED:
-            filtered_prompt_messages = self._handle_prompt_based_schema(
-                prompt_messages=filtered_prompt_messages,
-            )
-        stop = model_config.stop
-        return filtered_prompt_messages, stop
+
+        model = ModelManager().get_model_instance(
+            tenant_id=self.tenant_id,
+            model_type=ModelType.LLM,
+            provider=model_config.provider,
+            model=model_config.model,
+        )
+        model_schema = model.model_type_instance.get_model_schema(
+            model=model_config.model,
+            credentials=model.credentials,
+        )
+        if not model_schema:
+            raise ModelNotExistError(f"Model {model_config.model} not exist.")
+        if self.node_data.structured_output_enabled:
+            if not model_schema.support_structure_output:
+                filtered_prompt_messages = self._handle_prompt_based_schema(
+                    prompt_messages=filtered_prompt_messages,
+                )
+        return filtered_prompt_messages, model_config.stop
 
     def _parse_structured_output(self, result_text: str) -> dict[str, Any]:
         structured_output: dict[str, Any] = {}
@@ -816,51 +741,6 @@ class LLMNode(BaseNode[LLMNodeData]):
             structured_output = parsed
         return structured_output
 
-    @classmethod
-    def deduct_llm_quota(cls, tenant_id: str, model_instance: ModelInstance, usage: LLMUsage) -> None:
-        provider_model_bundle = model_instance.provider_model_bundle
-        provider_configuration = provider_model_bundle.configuration
-
-        if provider_configuration.using_provider_type != ProviderType.SYSTEM:
-            return
-
-        system_configuration = provider_configuration.system_configuration
-
-        quota_unit = None
-        for quota_configuration in system_configuration.quota_configurations:
-            if quota_configuration.quota_type == system_configuration.current_quota_type:
-                quota_unit = quota_configuration.quota_unit
-
-                if quota_configuration.quota_limit == -1:
-                    return
-
-                break
-
-        used_quota = None
-        if quota_unit:
-            if quota_unit == QuotaUnit.TOKENS:
-                used_quota = usage.total_tokens
-            elif quota_unit == QuotaUnit.CREDITS:
-                used_quota = dify_config.get_model_credits(model_instance.model)
-            else:
-                used_quota = 1
-
-        if used_quota is not None and system_configuration.current_quota_type is not None:
-            db.session.query(Provider).filter(
-                Provider.tenant_id == tenant_id,
-                # TODO: Use provider name with prefix after the data migration.
-                Provider.provider_name == ModelProviderID(model_instance.provider).provider_name,
-                Provider.provider_type == ProviderType.SYSTEM.value,
-                Provider.quota_type == system_configuration.current_quota_type.value,
-                Provider.quota_limit > Provider.quota_used,
-            ).update(
-                {
-                    "quota_used": Provider.quota_used + used_quota,
-                    "last_used": datetime.now(tz=UTC).replace(tzinfo=None),
-                }
-            )
-            db.session.commit()
-
     @classmethod
     def _extract_variable_selector_to_variable_mapping(
         cls,
@@ -902,7 +782,7 @@ class LLMNode(BaseNode[LLMNodeData]):
             variable_mapping["#context#"] = node_data.context.variable_selector
 
         if node_data.vision.enabled:
-            variable_mapping["#files#"] = ["sys", SystemVariableKey.FILES.value]
+            variable_mapping["#files#"] = node_data.vision.configs.variable_selector
 
         if node_data.memory:
             variable_mapping["#sys.query#"] = ["sys", SystemVariableKey.QUERY.value]
@@ -1184,32 +1064,6 @@ class LLMNode(BaseNode[LLMNodeData]):
         except json.JSONDecodeError:
             raise LLMNodeError("structured_output_schema is not valid JSON format")
 
-    def _check_model_structured_output_support(self) -> SupportStructuredOutputStatus:
-        """
-        Check if the current model supports structured output.
-
-        Returns:
-            SupportStructuredOutput: The support status of structured output
-        """
-        # Early return if structured output is disabled
-        if (
-            not isinstance(self.node_data, LLMNodeData)
-            or not self.node_data.structured_output_enabled
-            or not self.node_data.structured_output
-        ):
-            return SupportStructuredOutputStatus.DISABLED
-        # Get model schema and check if it exists
-        model_schema = self._fetch_model_schema(self.node_data.model.provider)
-        if not model_schema:
-            return SupportStructuredOutputStatus.DISABLED
-
-        # Check if model supports structured output feature
-        return (
-            SupportStructuredOutputStatus.SUPPORTED
-            if bool(model_schema.features and ModelFeature.STRUCTURED_OUTPUT in model_schema.features)
-            else SupportStructuredOutputStatus.UNSUPPORTED
-        )
-
     def _save_multimodal_output_and_convert_result_to_markdown(
         self,
         contents: str | list[PromptMessageContentUnionTypes] | None,

api/core/workflow/nodes/parameter_extractor/parameter_extractor_node.py

@@ -28,10 +28,10 @@ from core.prompt.utils.prompt_message_util import PromptMessageUtil
 from core.workflow.entities.node_entities import NodeRunResult
 from core.workflow.entities.variable_pool import VariablePool
 from core.workflow.entities.workflow_node_execution import WorkflowNodeExecutionMetadataKey, WorkflowNodeExecutionStatus
+from core.workflow.nodes.base.node import BaseNode
 from core.workflow.nodes.enums import NodeType
-from core.workflow.nodes.llm import LLMNode, ModelConfig
+from core.workflow.nodes.llm import ModelConfig, llm_utils
 from core.workflow.utils import variable_template_parser
-from extensions.ext_database import db
 
 from .entities import ParameterExtractorNodeData
 from .exc import (
@@ -84,7 +84,7 @@ def extract_json(text):
     return None
 
 
-class ParameterExtractorNode(LLMNode):
+class ParameterExtractorNode(BaseNode):
     """
     Parameter Extractor Node.
     """
@@ -117,8 +117,11 @@ class ParameterExtractorNode(LLMNode):
         variable = self.graph_runtime_state.variable_pool.get(node_data.query)
         query = variable.text if variable else ""
 
+        variable_pool = self.graph_runtime_state.variable_pool
+
         files = (
-            self._fetch_files(
+            llm_utils.fetch_files(
+                variable_pool=variable_pool,
                 selector=node_data.vision.configs.variable_selector,
             )
             if node_data.vision.enabled
@@ -138,7 +141,9 @@ class ParameterExtractorNode(LLMNode):
             raise ModelSchemaNotFoundError("Model schema not found")
 
         # fetch memory
-        memory = self._fetch_memory(
+        memory = llm_utils.fetch_memory(
+            variable_pool=variable_pool,
+            app_id=self.app_id,
             node_data_memory=node_data.memory,
             model_instance=model_instance,
         )
@@ -259,8 +264,6 @@ class ParameterExtractorNode(LLMNode):
         tools: list[PromptMessageTool],
         stop: list[str],
     ) -> tuple[str, LLMUsage, Optional[AssistantPromptMessage.ToolCall]]:
-        db.session.close()
-
         invoke_result = model_instance.invoke_llm(
             prompt_messages=prompt_messages,
             model_parameters=node_data_model.completion_params,
@@ -282,7 +285,7 @@ class ParameterExtractorNode(LLMNode):
         tool_call = invoke_result.message.tool_calls[0] if invoke_result.message.tool_calls else None
 
         # deduct quota
-        self.deduct_llm_quota(tenant_id=self.tenant_id, model_instance=model_instance, usage=usage)
+        llm_utils.deduct_llm_quota(tenant_id=self.tenant_id, model_instance=model_instance, usage=usage)
 
         if text is None:
             text = ""
@@ -797,7 +800,9 @@ class ParameterExtractorNode(LLMNode):
         Fetch model config.
         """
         if not self._model_instance or not self._model_config:
-            self._model_instance, self._model_config = super()._fetch_model_config(node_data_model)
+            self._model_instance, self._model_config = llm_utils.fetch_model_config(
+                tenant_id=self.tenant_id, node_data_model=node_data_model
+            )
 
         return self._model_instance, self._model_config
 
@@ -816,7 +821,6 @@ class ParameterExtractorNode(LLMNode):
         :param node_data: node data
         :return:
         """
-        # FIXME: fix the type error later
         variable_mapping: dict[str, Sequence[str]] = {"query": node_data.query}
 
         if node_data.instruction:

api/core/workflow/nodes/question_classifier/entities.py

@@ -19,3 +19,12 @@ class QuestionClassifierNodeData(BaseNodeData):
     instruction: Optional[str] = None
     memory: Optional[MemoryConfig] = None
     vision: VisionConfig = Field(default_factory=VisionConfig)
+
+    @property
+    def structured_output_enabled(self) -> bool:
+        # NOTE(QuantumGhost): Temporary workaround for issue #20725
+        # (https://github.com/langgenius/dify/issues/20725).
+        #
+        # The proper fix would be to make `QuestionClassifierNode` inherit
+        # from `BaseNode` instead of `LLMNode`.
+        return False

api/core/workflow/nodes/question_classifier/question_classifier_node.py

@@ -18,6 +18,7 @@ from core.workflow.nodes.llm import (
     LLMNode,
     LLMNodeChatModelMessage,
     LLMNodeCompletionModelPromptTemplate,
+    llm_utils,
 )
 from core.workflow.utils.variable_template_parser import VariableTemplateParser
 from libs.json_in_md_parser import parse_and_check_json_markdown
@@ -50,7 +51,9 @@ class QuestionClassifierNode(LLMNode):
         # fetch model config
         model_instance, model_config = self._fetch_model_config(node_data.model)
         # fetch memory
-        memory = self._fetch_memory(
+        memory = llm_utils.fetch_memory(
+            variable_pool=variable_pool,
+            app_id=self.app_id,
             node_data_memory=node_data.memory,
             model_instance=model_instance,
         )
@@ -59,7 +62,8 @@ class QuestionClassifierNode(LLMNode):
         node_data.instruction = variable_pool.convert_template(node_data.instruction).text
 
         files = (
-            self._fetch_files(
+            llm_utils.fetch_files(
+                variable_pool=variable_pool,
                 selector=node_data.vision.configs.variable_selector,
             )
             if node_data.vision.enabled
@@ -79,9 +83,13 @@ class QuestionClassifierNode(LLMNode):
             memory=memory,
             max_token_limit=rest_token,
         )
+        # Some models (e.g. Gemma, Mistral) force roles alternation (user/assistant/user/assistant...).
+        # If both self._get_prompt_template and self._fetch_prompt_messages append a user prompt,
+        # two consecutive user prompts will be generated, causing model's error.
+        # To avoid this, set sys_query to an empty string so that only one user prompt is appended at the end.
         prompt_messages, stop = self._fetch_prompt_messages(
             prompt_template=prompt_template,
-            sys_query=query,
+            sys_query="",
             memory=memory,
             model_config=model_config,
             sys_files=files,

api/core/workflow/nodes/variable_aggregator/entities.py

@@ -1,7 +1,8 @@
-from typing import Literal, Optional
+from typing import Optional
 
 from pydantic import BaseModel
 
+from core.variables.types import SegmentType
 from core.workflow.nodes.base import BaseNodeData
 
 
@@ -17,7 +18,7 @@ class AdvancedSettings(BaseModel):
         Group.
         """
 
-        output_type: Literal["string", "number", "object", "array[string]", "array[number]", "array[object]"]
+        output_type: SegmentType
         variables: list[list[str]]
         group_name: str
 

api/core/workflow/utils/structured_output/entities.py

@@ -14,11 +14,3 @@ class SpecialModelType(StrEnum):
 
     GEMINI = "gemini"
     OLLAMA = "ollama"
-
-
-class SupportStructuredOutputStatus(StrEnum):
-    """Constants for structured output support status"""
-
-    SUPPORTED = "supported"
-    UNSUPPORTED = "unsupported"
-    DISABLED = "disabled"

api/extensions/ext_celery.py

@@ -70,6 +70,7 @@ def init_app(app: DifyApp) -> Celery:
         "schedule.update_tidb_serverless_status_task",
         "schedule.clean_messages",
         "schedule.mail_clean_document_notify_task",
+        "schedule.queue_monitor_task",
     ]
     day = dify_config.CELERY_BEAT_SCHEDULER_TIME
     beat_schedule = {
@@ -98,6 +99,12 @@ def init_app(app: DifyApp) -> Celery:
             "task": "schedule.mail_clean_document_notify_task.mail_clean_document_notify_task",
             "schedule": crontab(minute="0", hour="10", day_of_week="1"),
         },
+        "datasets-queue-monitor": {
+            "task": "schedule.queue_monitor_task.queue_monitor_task",
+            "schedule": timedelta(
+                minutes=dify_config.QUEUE_MONITOR_INTERVAL if dify_config.QUEUE_MONITOR_INTERVAL else 30
+            ),
+        },
     }
     celery_app.conf.update(beat_schedule=beat_schedule, imports=imports)
 

api/extensions/ext_login.py

@@ -57,6 +57,9 @@ def load_user_from_request(request_from_flask_login):
             raise Unauthorized("Invalid Authorization token.")
         decoded = PassportService().verify(auth_token)
         user_id = decoded.get("user_id")
+        source = decoded.get("token_source")
+        if source:
+            raise Unauthorized("Invalid Authorization token.")
         if not user_id:
             raise Unauthorized("Invalid Authorization token.")
 

api/factories/variable_factory.py

@@ -84,8 +84,8 @@ def _build_variable_from_mapping(*, mapping: Mapping[str, Any], selector: Sequen
         raise VariableError("missing value type")
     if (value := mapping.get("value")) is None:
         raise VariableError("missing value")
-    # FIXME: using Any here, fix it later
-    result: Any
+
+    result: Variable
     match value_type:
         case SegmentType.STRING:
             result = StringVariable.model_validate(mapping)

api/libs/helper.py

@@ -1,8 +1,9 @@
 import json
 import logging
-import random
 import re
+import secrets
 import string
+import struct
 import subprocess
 import time
 import uuid
@@ -14,10 +15,12 @@ from zoneinfo import available_timezones
 
 from flask import Response, stream_with_context
 from flask_restful import fields
+from pydantic import BaseModel
 
 from configs import dify_config
 from core.app.features.rate_limiting.rate_limit import RateLimitGenerator
 from core.file import helpers as file_helpers
+from core.model_runtime.utils.encoders import jsonable_encoder
 from extensions.ext_redis import redis_client
 
 if TYPE_CHECKING:
@@ -175,14 +178,14 @@ def generate_string(n):
     letters_digits = string.ascii_letters + string.digits
     result = ""
     for i in range(n):
-        result += random.choice(letters_digits)
+        result += secrets.choice(letters_digits)
 
     return result
 
 
 def extract_remote_ip(request) -> str:
     if request.headers.get("CF-Connecting-IP"):
-        return cast(str, request.headers.get("Cf-Connecting-Ip"))
+        return cast(str, request.headers.get("CF-Connecting-IP"))
     elif request.headers.getlist("X-Forwarded-For"):
         return cast(str, request.headers.getlist("X-Forwarded-For")[0])
     else:
@@ -196,7 +199,7 @@ def generate_text_hash(text: str) -> str:
 
 def compact_generate_response(response: Union[Mapping, Generator, RateLimitGenerator]) -> Response:
     if isinstance(response, dict):
-        return Response(response=json.dumps(response), status=200, mimetype="application/json")
+        return Response(response=json.dumps(jsonable_encoder(response)), status=200, mimetype="application/json")
     else:
 
         def generate() -> Generator:
@@ -205,6 +208,60 @@ def compact_generate_response(response: Union[Mapping, Generator, RateLimitGener
         return Response(stream_with_context(generate()), status=200, mimetype="text/event-stream")
 
 
+def length_prefixed_response(magic_number: int, response: Union[Mapping, Generator, RateLimitGenerator]) -> Response:
+    """
+    This function is used to return a response with a length prefix.
+    Magic number is a one byte number that indicates the type of the response.
+
+    For a compatibility with latest plugin daemon https://github.com/langgenius/dify-plugin-daemon/pull/341
+    Avoid using line-based response, it leads a memory issue.
+
+    We uses following format:
+    | Field         | Size     | Description                     |
+    |---------------|----------|---------------------------------|
+    | Magic Number  | 1 byte   | Magic number identifier         |
+    | Reserved      | 1 byte   | Reserved field                  |
+    | Header Length | 2 bytes  | Header length (usually 0xa)    |
+    | Data Length   | 4 bytes  | Length of the data              |
+    | Reserved      | 6 bytes  | Reserved fields                 |
+    | Data          | Variable | Actual data content             |
+
+    | Reserved Fields | Header   | Data     |
+    |-----------------|----------|----------|
+    | 4 bytes total   | Variable | Variable |
+
+    all data is in little endian
+    """
+
+    def pack_response_with_length_prefix(response: bytes) -> bytes:
+        header_length = 0xA
+        data_length = len(response)
+        # | Magic Number 1byte | Reserved 1byte | Header Length 2bytes | Data Length 4bytes | Reserved 6bytes | Data
+        return struct.pack("<BBHI", magic_number, 0, header_length, data_length) + b"\x00" * 6 + response
+
+    if isinstance(response, dict):
+        return Response(
+            response=pack_response_with_length_prefix(json.dumps(jsonable_encoder(response)).encode("utf-8")),
+            status=200,
+            mimetype="application/json",
+        )
+    elif isinstance(response, BaseModel):
+        return Response(
+            response=pack_response_with_length_prefix(response.model_dump_json().encode("utf-8")),
+            status=200,
+            mimetype="application/json",
+        )
+
+    def generate() -> Generator:
+        for chunk in response:
+            if isinstance(chunk, str):
+                yield pack_response_with_length_prefix(chunk.encode("utf-8"))
+            else:
+                yield pack_response_with_length_prefix(chunk)
+
+    return Response(stream_with_context(generate()), status=200, mimetype="text/event-stream")
+
+
 class TokenManager:
     @classmethod
     def generate_token(

api/libs/smtp.py

@@ -28,7 +28,8 @@ class SMTPClient:
             else:
                 smtp = smtplib.SMTP(self.server, self.port, timeout=10)
 
-            if self.username and self.password:
+            # Only authenticate if both username and password are non-empty
+            if self.username and self.password and self.username.strip() and self.password.strip():
                 smtp.login(self.username, self.password)
 
             msg = MIMEMultipart()

api/migrations/versions/2025_06_06_1424-4474872b0ee6_workflow_draft_varaibles_add_node_execution_id.py

@@ -0,0 +1,60 @@
+"""`workflow_draft_varaibles` add `node_execution_id` column, add an index for `workflow_node_executions`.
+
+Revision ID: 4474872b0ee6
+Revises: 2adcbe1f5dfb
+Create Date: 2025-06-06 14:24:44.213018
+
+"""
+from alembic import op
+import models as models
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision = '4474872b0ee6'
+down_revision = '2adcbe1f5dfb'
+branch_labels = None
+depends_on = None
+
+
+def upgrade():
+    # `CREATE INDEX CONCURRENTLY` cannot run within a transaction, so use the `autocommit_block`
+    # context manager to wrap the index creation statement.
+    # Reference:
+    #
+    # - https://www.postgresql.org/docs/current/sql-createindex.html#:~:text=Another%20difference%20is,CREATE%20INDEX%20CONCURRENTLY%20cannot.
+    # - https://alembic.sqlalchemy.org/en/latest/api/runtime.html#alembic.runtime.migration.MigrationContext.autocommit_block
+    with op.get_context().autocommit_block():
+        op.create_index(
+            op.f('workflow_node_executions_tenant_id_idx'),
+            "workflow_node_executions",
+            ['tenant_id', 'workflow_id', 'node_id', sa.literal_column('created_at DESC')],
+            unique=False,
+            postgresql_concurrently=True,
+        )
+
+    with op.batch_alter_table('workflow_draft_variables', schema=None) as batch_op:
+        batch_op.add_column(sa.Column('node_execution_id', models.types.StringUUID(), nullable=True))
+    # ### end Alembic commands ###
+
+
+def downgrade():
+    # ### commands auto generated by Alembic - please adjust! ###
+
+    # `DROP INDEX CONCURRENTLY` cannot run within a transaction, so use the `autocommit_block`
+    # context manager to wrap the index creation statement.
+    # Reference:
+    #
+    # - https://www.postgresql.org/docs/current/sql-createindex.html#:~:text=Another%20difference%20is,CREATE%20INDEX%20CONCURRENTLY%20cannot.
+    # - https://alembic.sqlalchemy.org/en/latest/api/runtime.html#alembic.runtime.migration.MigrationContext.autocommit_block
+    # `DROP INDEX CONCURRENTLY` cannot run within a transaction, so commit existing transactions first.
+    # Reference:
+    #
+    # https://www.postgresql.org/docs/current/sql-createindex.html#:~:text=Another%20difference%20is,CREATE%20INDEX%20CONCURRENTLY%20cannot.
+    with op.get_context().autocommit_block():
+        op.drop_index(op.f('workflow_node_executions_tenant_id_idx'), postgresql_concurrently=True)
+
+    with op.batch_alter_table('workflow_draft_variables', schema=None) as batch_op:
+        batch_op.drop_column('node_execution_id')
+
+    # ### end Alembic commands ###

api/models/provider.py

@@ -1,6 +1,9 @@
+from datetime import datetime
 from enum import Enum
+from typing import Optional
 
-from sqlalchemy import func
+from sqlalchemy import func, text
+from sqlalchemy.orm import Mapped, mapped_column
 
 from .base import Base
 from .engine import db
@@ -51,20 +54,24 @@ class Provider(Base):
         ),
     )
 
-    id = db.Column(StringUUID, server_default=db.text("uuid_generate_v4()"))
-    tenant_id = db.Column(StringUUID, nullable=False)
-    provider_name = db.Column(db.String(255), nullable=False)
-    provider_type = db.Column(db.String(40), nullable=False, server_default=db.text("'custom'::character varying"))
-    encrypted_config = db.Column(db.Text, nullable=True)
-    is_valid = db.Column(db.Boolean, nullable=False, server_default=db.text("false"))
-    last_used = db.Column(db.DateTime, nullable=True)
+    id: Mapped[str] = mapped_column(StringUUID, server_default=text("uuid_generate_v4()"))
+    tenant_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    provider_name: Mapped[str] = mapped_column(db.String(255), nullable=False)
+    provider_type: Mapped[str] = mapped_column(
+        db.String(40), nullable=False, server_default=text("'custom'::character varying")
+    )
+    encrypted_config: Mapped[Optional[str]] = mapped_column(db.Text, nullable=True)
+    is_valid: Mapped[bool] = mapped_column(db.Boolean, nullable=False, server_default=text("false"))
+    last_used: Mapped[Optional[datetime]] = mapped_column(db.DateTime, nullable=True)
 
-    quota_type = db.Column(db.String(40), nullable=True, server_default=db.text("''::character varying"))
-    quota_limit = db.Column(db.BigInteger, nullable=True)
-    quota_used = db.Column(db.BigInteger, default=0)
+    quota_type: Mapped[Optional[str]] = mapped_column(
+        db.String(40), nullable=True, server_default=text("''::character varying")
+    )
+    quota_limit: Mapped[Optional[int]] = mapped_column(db.BigInteger, nullable=True)
+    quota_used: Mapped[Optional[int]] = mapped_column(db.BigInteger, default=0)
 
-    created_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
-    updated_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    created_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    updated_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
 
     def __repr__(self):
         return (
@@ -104,15 +111,15 @@ class ProviderModel(Base):
         ),
     )
 
-    id = db.Column(StringUUID, server_default=db.text("uuid_generate_v4()"))
-    tenant_id = db.Column(StringUUID, nullable=False)
-    provider_name = db.Column(db.String(255), nullable=False)
-    model_name = db.Column(db.String(255), nullable=False)
-    model_type = db.Column(db.String(40), nullable=False)
-    encrypted_config = db.Column(db.Text, nullable=True)
-    is_valid = db.Column(db.Boolean, nullable=False, server_default=db.text("false"))
-    created_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
-    updated_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    id: Mapped[str] = mapped_column(StringUUID, server_default=text("uuid_generate_v4()"))
+    tenant_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    provider_name: Mapped[str] = mapped_column(db.String(255), nullable=False)
+    model_name: Mapped[str] = mapped_column(db.String(255), nullable=False)
+    model_type: Mapped[str] = mapped_column(db.String(40), nullable=False)
+    encrypted_config: Mapped[Optional[str]] = mapped_column(db.Text, nullable=True)
+    is_valid: Mapped[bool] = mapped_column(db.Boolean, nullable=False, server_default=text("false"))
+    created_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    updated_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
 
 
 class TenantDefaultModel(Base):
@@ -122,13 +129,13 @@ class TenantDefaultModel(Base):
         db.Index("tenant_default_model_tenant_id_provider_type_idx", "tenant_id", "provider_name", "model_type"),
     )
 
-    id = db.Column(StringUUID, server_default=db.text("uuid_generate_v4()"))
-    tenant_id = db.Column(StringUUID, nullable=False)
-    provider_name = db.Column(db.String(255), nullable=False)
-    model_name = db.Column(db.String(255), nullable=False)
-    model_type = db.Column(db.String(40), nullable=False)
-    created_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
-    updated_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    id: Mapped[str] = mapped_column(StringUUID, server_default=text("uuid_generate_v4()"))
+    tenant_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    provider_name: Mapped[str] = mapped_column(db.String(255), nullable=False)
+    model_name: Mapped[str] = mapped_column(db.String(255), nullable=False)
+    model_type: Mapped[str] = mapped_column(db.String(40), nullable=False)
+    created_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    updated_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
 
 
 class TenantPreferredModelProvider(Base):
@@ -138,12 +145,12 @@ class TenantPreferredModelProvider(Base):
         db.Index("tenant_preferred_model_provider_tenant_provider_idx", "tenant_id", "provider_name"),
     )
 
-    id = db.Column(StringUUID, server_default=db.text("uuid_generate_v4()"))
-    tenant_id = db.Column(StringUUID, nullable=False)
-    provider_name = db.Column(db.String(255), nullable=False)
-    preferred_provider_type = db.Column(db.String(40), nullable=False)
-    created_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
-    updated_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    id: Mapped[str] = mapped_column(StringUUID, server_default=text("uuid_generate_v4()"))
+    tenant_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    provider_name: Mapped[str] = mapped_column(db.String(255), nullable=False)
+    preferred_provider_type: Mapped[str] = mapped_column(db.String(40), nullable=False)
+    created_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    updated_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
 
 
 class ProviderOrder(Base):
@@ -153,22 +160,24 @@ class ProviderOrder(Base):
         db.Index("provider_order_tenant_provider_idx", "tenant_id", "provider_name"),
     )
 
-    id = db.Column(StringUUID, server_default=db.text("uuid_generate_v4()"))
-    tenant_id = db.Column(StringUUID, nullable=False)
-    provider_name = db.Column(db.String(255), nullable=False)
-    account_id = db.Column(StringUUID, nullable=False)
-    payment_product_id = db.Column(db.String(191), nullable=False)
-    payment_id = db.Column(db.String(191))
-    transaction_id = db.Column(db.String(191))
-    quantity = db.Column(db.Integer, nullable=False, server_default=db.text("1"))
-    currency = db.Column(db.String(40))
-    total_amount = db.Column(db.Integer)
-    payment_status = db.Column(db.String(40), nullable=False, server_default=db.text("'wait_pay'::character varying"))
-    paid_at = db.Column(db.DateTime)
-    pay_failed_at = db.Column(db.DateTime)
-    refunded_at = db.Column(db.DateTime)
-    created_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
-    updated_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    id: Mapped[str] = mapped_column(StringUUID, server_default=text("uuid_generate_v4()"))
+    tenant_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    provider_name: Mapped[str] = mapped_column(db.String(255), nullable=False)
+    account_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    payment_product_id: Mapped[str] = mapped_column(db.String(191), nullable=False)
+    payment_id: Mapped[Optional[str]] = mapped_column(db.String(191))
+    transaction_id: Mapped[Optional[str]] = mapped_column(db.String(191))
+    quantity: Mapped[int] = mapped_column(db.Integer, nullable=False, server_default=text("1"))
+    currency: Mapped[Optional[str]] = mapped_column(db.String(40))
+    total_amount: Mapped[Optional[int]] = mapped_column(db.Integer)
+    payment_status: Mapped[str] = mapped_column(
+        db.String(40), nullable=False, server_default=text("'wait_pay'::character varying")
+    )
+    paid_at: Mapped[Optional[datetime]] = mapped_column(db.DateTime)
+    pay_failed_at: Mapped[Optional[datetime]] = mapped_column(db.DateTime)
+    refunded_at: Mapped[Optional[datetime]] = mapped_column(db.DateTime)
+    created_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    updated_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
 
 
 class ProviderModelSetting(Base):
@@ -182,15 +191,15 @@ class ProviderModelSetting(Base):
         db.Index("provider_model_setting_tenant_provider_model_idx", "tenant_id", "provider_name", "model_type"),
     )
 
-    id = db.Column(StringUUID, server_default=db.text("uuid_generate_v4()"))
-    tenant_id = db.Column(StringUUID, nullable=False)
-    provider_name = db.Column(db.String(255), nullable=False)
-    model_name = db.Column(db.String(255), nullable=False)
-    model_type = db.Column(db.String(40), nullable=False)
-    enabled = db.Column(db.Boolean, nullable=False, server_default=db.text("true"))
-    load_balancing_enabled = db.Column(db.Boolean, nullable=False, server_default=db.text("false"))
-    created_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
-    updated_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    id: Mapped[str] = mapped_column(StringUUID, server_default=text("uuid_generate_v4()"))
+    tenant_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    provider_name: Mapped[str] = mapped_column(db.String(255), nullable=False)
+    model_name: Mapped[str] = mapped_column(db.String(255), nullable=False)
+    model_type: Mapped[str] = mapped_column(db.String(40), nullable=False)
+    enabled: Mapped[bool] = mapped_column(db.Boolean, nullable=False, server_default=text("true"))
+    load_balancing_enabled: Mapped[bool] = mapped_column(db.Boolean, nullable=False, server_default=text("false"))
+    created_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    updated_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
 
 
 class LoadBalancingModelConfig(Base):
@@ -204,13 +213,13 @@ class LoadBalancingModelConfig(Base):
         db.Index("load_balancing_model_config_tenant_provider_model_idx", "tenant_id", "provider_name", "model_type"),
     )
 
-    id = db.Column(StringUUID, server_default=db.text("uuid_generate_v4()"))
-    tenant_id = db.Column(StringUUID, nullable=False)
-    provider_name = db.Column(db.String(255), nullable=False)
-    model_name = db.Column(db.String(255), nullable=False)
-    model_type = db.Column(db.String(40), nullable=False)
-    name = db.Column(db.String(255), nullable=False)
-    encrypted_config = db.Column(db.Text, nullable=True)
-    enabled = db.Column(db.Boolean, nullable=False, server_default=db.text("true"))
-    created_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
-    updated_at = db.Column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    id: Mapped[str] = mapped_column(StringUUID, server_default=text("uuid_generate_v4()"))
+    tenant_id: Mapped[str] = mapped_column(StringUUID, nullable=False)
+    provider_name: Mapped[str] = mapped_column(db.String(255), nullable=False)
+    model_name: Mapped[str] = mapped_column(db.String(255), nullable=False)
+    model_type: Mapped[str] = mapped_column(db.String(40), nullable=False)
+    name: Mapped[str] = mapped_column(db.String(255), nullable=False)
+    encrypted_config: Mapped[Optional[str]] = mapped_column(db.Text, nullable=True)
+    enabled: Mapped[bool] = mapped_column(db.Boolean, nullable=False, server_default=text("true"))
+    created_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())
+    updated_at: Mapped[datetime] = mapped_column(db.DateTime, nullable=False, server_default=func.current_timestamp())

api/models/workflow.py

@@ -16,8 +16,8 @@ if TYPE_CHECKING:
     from models.model import AppMode
 
 import sqlalchemy as sa
-from sqlalchemy import UniqueConstraint, func
-from sqlalchemy.orm import Mapped, mapped_column
+from sqlalchemy import Index, PrimaryKeyConstraint, UniqueConstraint, func
+from sqlalchemy.orm import Mapped, declared_attr, mapped_column
 
 from constants import DEFAULT_FILE_NUMBER_LIMITS, HIDDEN_VALUE
 from core.helper import encrypter
@@ -590,28 +590,48 @@ class WorkflowNodeExecutionModel(Base):
     """
 
     __tablename__ = "workflow_node_executions"
-    __table_args__ = (
-        db.PrimaryKeyConstraint("id", name="workflow_node_execution_pkey"),
-        db.Index(
-            "workflow_node_execution_workflow_run_idx",
-            "tenant_id",
-            "app_id",
-            "workflow_id",
-            "triggered_from",
-            "workflow_run_id",
-        ),
-        db.Index(
-            "workflow_node_execution_node_run_idx", "tenant_id", "app_id", "workflow_id", "triggered_from", "node_id"
-        ),
-        db.Index(
-            "workflow_node_execution_id_idx",
-            "tenant_id",
-            "app_id",
-            "workflow_id",
-            "triggered_from",
-            "node_execution_id",
-        ),
-    )
+
+    @declared_attr
+    def __table_args__(cls):  # noqa
+        return (
+            PrimaryKeyConstraint("id", name="workflow_node_execution_pkey"),
+            Index(
+                "workflow_node_execution_workflow_run_idx",
+                "tenant_id",
+                "app_id",
+                "workflow_id",
+                "triggered_from",
+                "workflow_run_id",
+            ),
+            Index(
+                "workflow_node_execution_node_run_idx",
+                "tenant_id",
+                "app_id",
+                "workflow_id",
+                "triggered_from",
+                "node_id",
+            ),
+            Index(
+                "workflow_node_execution_id_idx",
+                "tenant_id",
+                "app_id",
+                "workflow_id",
+                "triggered_from",
+                "node_execution_id",
+            ),
+            Index(
+                # The first argument is the index name,
+                # which we leave as `None`` to allow auto-generation by the ORM.
+                None,
+                cls.tenant_id,
+                cls.workflow_id,
+                cls.node_id,
+                # MyPy may flag the following line because it doesn't recognize that
+                # the `declared_attr` decorator passes the receiving class as the first
+                # argument to this method, allowing us to reference class attributes.
+                cls.created_at.desc(),  # type: ignore
+            ),
+        )
 
     id: Mapped[str] = mapped_column(StringUUID, server_default=db.text("uuid_generate_v4()"))
     tenant_id: Mapped[str] = mapped_column(StringUUID)
@@ -885,14 +905,29 @@ class WorkflowDraftVariable(Base):
 
     selector: Mapped[str] = mapped_column(sa.String(255), nullable=False, name="selector")
 
+    # The data type of this variable's value
     value_type: Mapped[SegmentType] = mapped_column(EnumText(SegmentType, length=20))
-    # JSON string
+
+    # The variable's value serialized as a JSON string
     value: Mapped[str] = mapped_column(sa.Text, nullable=False, name="value")
 
-    # visible
+    # Controls whether the variable should be displayed in the variable inspection panel
     visible: Mapped[bool] = mapped_column(sa.Boolean, nullable=False, default=True)
+
+    # Determines whether this variable can be modified by users
     editable: Mapped[bool] = mapped_column(sa.Boolean, nullable=False, default=False)
 
+    # The `node_execution_id` field identifies the workflow node execution that created this variable.
+    # It corresponds to the `id` field in the `WorkflowNodeExecutionModel` model.
+    #
+    # This field is not `None` for system variables and node variables, and is  `None`
+    # for conversation variables.
+    node_execution_id: Mapped[str | None] = mapped_column(
+        StringUUID,
+        nullable=True,
+        default=None,
+    )
+
     def get_selector(self) -> list[str]:
         selector = json.loads(self.selector)
         if not isinstance(selector, list):