Merge branch 'main' into feat/pull-a-variable

.github/workflows/deploy-agent-dev.yml

@@ -1,4 +1,4 @@
-name: Deploy Trigger Dev
+name: Deploy Agent Dev
 
 permissions:
   contents: read
@@ -7,7 +7,7 @@ on:
   workflow_run:
     workflows: ["Build and Push API & Web"]
     branches:
-      - "deploy/trigger-dev"
+      - "deploy/agent-dev"
     types:
       - completed
 
@@ -16,12 +16,12 @@ jobs:
     runs-on: ubuntu-latest
     if: |
       github.event.workflow_run.conclusion == 'success' &&
-      github.event.workflow_run.head_branch == 'deploy/trigger-dev'
+      github.event.workflow_run.head_branch == 'deploy/agent-dev'
     steps:
       - name: Deploy to server
         uses: appleboy/ssh-action@v0.1.8
         with:
-          host: ${{ secrets.TRIGGER_SSH_HOST }}
+          host: ${{ secrets.AGENT_DEV_SSH_HOST }}
           username: ${{ secrets.SSH_USER }}
           key: ${{ secrets.SSH_PRIVATE_KEY }}
           script: |

api/core/app/app_config/entities.py

@@ -1,4 +1,3 @@
-import json
 from collections.abc import Sequence
 from enum import StrEnum, auto
 from typing import Any, Literal
@@ -121,7 +120,7 @@ class VariableEntity(BaseModel):
     allowed_file_types: Sequence[FileType] | None = Field(default_factory=list)
     allowed_file_extensions: Sequence[str] | None = Field(default_factory=list)
     allowed_file_upload_methods: Sequence[FileTransferMethod] | None = Field(default_factory=list)
-    json_schema: str | None = Field(default=None)
+    json_schema: dict | None = Field(default=None)
 
     @field_validator("description", mode="before")
     @classmethod
@@ -135,17 +134,11 @@ class VariableEntity(BaseModel):
 
     @field_validator("json_schema")
     @classmethod
-    def validate_json_schema(cls, schema: str | None) -> str | None:
+    def validate_json_schema(cls, schema: dict | None) -> dict | None:
         if schema is None:
             return None
-
         try:
-            json_schema = json.loads(schema)
-        except json.JSONDecodeError:
-            raise ValueError(f"invalid json_schema value {schema}")
-
-        try:
-            Draft7Validator.check_schema(json_schema)
+            Draft7Validator.check_schema(schema)
         except SchemaError as e:
             raise ValueError(f"Invalid JSON schema: {e.message}")
         return schema

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

@@ -26,7 +26,6 @@ class AdvancedChatAppConfigManager(BaseAppConfigManager):
     @classmethod
     def get_app_config(cls, app_model: App, workflow: Workflow) -> AdvancedChatAppConfig:
         features_dict = workflow.features_dict
-
         app_mode = AppMode.value_of(app_model.mode)
         app_config = AdvancedChatAppConfig(
             tenant_id=app_model.tenant_id,

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

@@ -358,25 +358,6 @@ class AdvancedChatAppGenerateTaskPipeline(GraphRuntimeStateSupport):
         if node_finish_resp:
             yield node_finish_resp
 
-        # For ANSWER nodes, check if we need to send a message_replace event
-        # Only send if the final output differs from the accumulated task_state.answer
-        # This happens when variables were updated by variable_assigner during workflow execution
-        if event.node_type == NodeType.ANSWER and event.outputs:
-            final_answer = event.outputs.get("answer")
-            if final_answer is not None and final_answer != self._task_state.answer:
-                logger.info(
-                    "ANSWER node final output '%s' differs from accumulated answer '%s', sending message_replace event",
-                    final_answer,
-                    self._task_state.answer,
-                )
-                # Update the task state answer
-                self._task_state.answer = str(final_answer)
-                # Send message_replace event to update the UI
-                yield self._message_cycle_manager.message_replace_to_stream_response(
-                    answer=str(final_answer),
-                    reason="variable_update",
-                )
-
     def _handle_node_failed_events(
         self,
         event: Union[QueueNodeFailedEvent, QueueNodeExceptionEvent],

api/core/app/apps/base_app_generator.py

@@ -1,4 +1,3 @@
-import json
 from collections.abc import Generator, Mapping, Sequence
 from typing import TYPE_CHECKING, Any, Union, final
 
@@ -76,12 +75,24 @@ class BaseAppGenerator:
         user_inputs = {**user_inputs, **files_inputs, **file_list_inputs}
 
         # Check if all files are converted to File
-        if any(filter(lambda v: isinstance(v, dict), user_inputs.values())):
-            raise ValueError("Invalid input type")
-        if any(
-            filter(lambda v: isinstance(v, dict), filter(lambda item: isinstance(item, list), user_inputs.values()))
-        ):
-            raise ValueError("Invalid input type")
+        invalid_dict_keys = [
+            k
+            for k, v in user_inputs.items()
+            if isinstance(v, dict)
+            and entity_dictionary[k].type not in {VariableEntityType.FILE, VariableEntityType.JSON_OBJECT}
+        ]
+        if invalid_dict_keys:
+            raise ValueError(f"Invalid input type for {invalid_dict_keys}")
+
+        invalid_list_dict_keys = [
+            k
+            for k, v in user_inputs.items()
+            if isinstance(v, list)
+            and any(isinstance(item, dict) for item in v)
+            and entity_dictionary[k].type != VariableEntityType.FILE_LIST
+        ]
+        if invalid_list_dict_keys:
+            raise ValueError(f"Invalid input type for {invalid_list_dict_keys}")
 
         return user_inputs
 
@@ -178,12 +189,8 @@ class BaseAppGenerator:
                     elif value == 0:
                         value = False
             case VariableEntityType.JSON_OBJECT:
-                if not isinstance(value, str):
-                    raise ValueError(f"{variable_entity.variable} in input form must be a string")
-                try:
-                    json.loads(value)
-                except json.JSONDecodeError:
-                    raise ValueError(f"{variable_entity.variable} in input form must be a valid JSON object")
+                if not isinstance(value, dict):
+                    raise ValueError(f"{variable_entity.variable} in input form must be a dict")
             case _:
                 raise AssertionError("this statement should be unreachable.")
 

api/core/workflow/nodes/start/start_node.py

@@ -1,4 +1,3 @@
-import json
 from typing import Any
 
 from jsonschema import Draft7Validator, ValidationError
@@ -43,25 +42,22 @@ class StartNode(Node[StartNodeData]):
             if value is None and variable.required:
                 raise ValueError(f"{key} is required in input form")
 
+            # If no value provided, skip further processing for this key
+            if not value:
+                continue
+
+            if not isinstance(value, dict):
+                raise ValueError(f"JSON object for '{key}' must be an object")
+
+            # Overwrite with normalized dict to ensure downstream consistency
+            node_inputs[key] = value
+
+            # If schema exists, then validate against it
             schema = variable.json_schema
             if not schema:
                 continue
 
-            if not value:
-                continue
-
             try:
-                json_schema = json.loads(schema)
-            except json.JSONDecodeError as e:
-                raise ValueError(f"{schema} must be a valid JSON object")
-
-            try:
-                json_value = json.loads(value)
-            except json.JSONDecodeError as e:
-                raise ValueError(f"{value} must be a valid JSON object")
-
-            try:
-                Draft7Validator(json_schema).validate(json_value)
+                Draft7Validator(schema).validate(value)
             except ValidationError as e:
                 raise ValueError(f"JSON object for '{key}' does not match schema: {e.message}")
-            node_inputs[key] = json_value

api/tests/unit_tests/core/app/apps/advanced_chat/test_generate_task_pipeline_answer_node.py

@@ -1,390 +0,0 @@
-"""
-Tests for AdvancedChatAppGenerateTaskPipeline._handle_node_succeeded_event method,
-specifically testing the ANSWER node message_replace logic.
-"""
-
-from datetime import datetime
-from types import SimpleNamespace
-from unittest.mock import MagicMock, Mock, patch
-
-import pytest
-
-from core.app.entities.app_invoke_entities import AdvancedChatAppGenerateEntity
-from core.app.entities.queue_entities import QueueNodeSucceededEvent
-from core.workflow.enums import NodeType
-from models import EndUser
-from models.model import AppMode
-
-
-class TestAnswerNodeMessageReplace:
-    """Test cases for ANSWER node message_replace event logic."""
-
-    @pytest.fixture
-    def mock_application_generate_entity(self):
-        """Create a mock application generate entity."""
-        entity = Mock(spec=AdvancedChatAppGenerateEntity)
-        entity.task_id = "test-task-id"
-        entity.app_id = "test-app-id"
-        entity.workflow_run_id = "test-workflow-run-id"
-        # minimal app_config used by pipeline internals
-        entity.app_config = SimpleNamespace(
-            tenant_id="test-tenant-id",
-            app_id="test-app-id",
-            app_mode=AppMode.ADVANCED_CHAT,
-            app_model_config_dict={},
-            additional_features=None,
-            sensitive_word_avoidance=None,
-        )
-        entity.query = "test query"
-        entity.files = []
-        entity.extras = {}
-        entity.trace_manager = None
-        entity.inputs = {}
-        entity.invoke_from = "debugger"
-        return entity
-
-    @pytest.fixture
-    def mock_workflow(self):
-        """Create a mock workflow."""
-        workflow = Mock()
-        workflow.id = "test-workflow-id"
-        workflow.features_dict = {}
-        return workflow
-
-    @pytest.fixture
-    def mock_queue_manager(self):
-        """Create a mock queue manager."""
-        manager = Mock()
-        manager.listen.return_value = []
-        manager.graph_runtime_state = None
-        return manager
-
-    @pytest.fixture
-    def mock_conversation(self):
-        """Create a mock conversation."""
-        conversation = Mock()
-        conversation.id = "test-conversation-id"
-        conversation.mode = "advanced_chat"
-        return conversation
-
-    @pytest.fixture
-    def mock_message(self):
-        """Create a mock message."""
-        message = Mock()
-        message.id = "test-message-id"
-        message.query = "test query"
-        message.created_at = Mock()
-        message.created_at.timestamp.return_value = 1234567890
-        return message
-
-    @pytest.fixture
-    def mock_user(self):
-        """Create a mock end user."""
-        user = MagicMock(spec=EndUser)
-        user.id = "test-user-id"
-        user.session_id = "test-session-id"
-        return user
-
-    @pytest.fixture
-    def mock_draft_var_saver_factory(self):
-        """Create a mock draft variable saver factory."""
-        return Mock()
-
-    @pytest.fixture
-    def pipeline(
-        self,
-        mock_application_generate_entity,
-        mock_workflow,
-        mock_queue_manager,
-        mock_conversation,
-        mock_message,
-        mock_user,
-        mock_draft_var_saver_factory,
-    ):
-        """Create an AdvancedChatAppGenerateTaskPipeline instance with mocked dependencies."""
-        from core.app.apps.advanced_chat.generate_task_pipeline import AdvancedChatAppGenerateTaskPipeline
-
-        with patch("core.app.apps.advanced_chat.generate_task_pipeline.db"):
-            pipeline = AdvancedChatAppGenerateTaskPipeline(
-                application_generate_entity=mock_application_generate_entity,
-                workflow=mock_workflow,
-                queue_manager=mock_queue_manager,
-                conversation=mock_conversation,
-                message=mock_message,
-                user=mock_user,
-                stream=True,
-                dialogue_count=1,
-                draft_var_saver_factory=mock_draft_var_saver_factory,
-            )
-            # Initialize workflow run id to avoid validation errors
-            pipeline._workflow_run_id = "test-workflow-run-id"
-            # Mock the message cycle manager methods we need to track
-            pipeline._message_cycle_manager.message_replace_to_stream_response = Mock()
-            return pipeline
-
-    def test_answer_node_with_different_output_sends_message_replace(self, pipeline, mock_application_generate_entity):
-        """
-        Test that when an ANSWER node's final output differs from accumulated answer,
-        a message_replace event is sent.
-        """
-        # Arrange: Set initial accumulated answer
-        pipeline._task_state.answer = "initial answer"
-
-        # Create ANSWER node succeeded event with different final output
-        event = QueueNodeSucceededEvent(
-            node_execution_id="test-node-execution-id",
-            node_id="test-answer-node",
-            node_type=NodeType.ANSWER,
-            start_at=datetime.now(),
-            outputs={"answer": "updated final answer"},
-        )
-
-        # Mock the workflow response converter to avoid extra processing
-        pipeline._workflow_response_converter.workflow_node_finish_to_stream_response = Mock(return_value=None)
-        pipeline._save_output_for_event = Mock()
-
-        # Act
-        responses = list(pipeline._handle_node_succeeded_event(event))
-
-        # Assert
-        assert pipeline._task_state.answer == "updated final answer"
-        # Verify message_replace was called
-        pipeline._message_cycle_manager.message_replace_to_stream_response.assert_called_once_with(
-            answer="updated final answer", reason="variable_update"
-        )
-
-    def test_answer_node_with_same_output_does_not_send_message_replace(self, pipeline):
-        """
-        Test that when an ANSWER node's final output is the same as accumulated answer,
-        no message_replace event is sent.
-        """
-        # Arrange: Set initial accumulated answer
-        pipeline._task_state.answer = "same answer"
-
-        # Create ANSWER node succeeded event with same output
-        event = QueueNodeSucceededEvent(
-            node_execution_id="test-node-execution-id",
-            node_id="test-answer-node",
-            node_type=NodeType.ANSWER,
-            start_at=datetime.now(),
-            outputs={"answer": "same answer"},
-        )
-
-        # Mock the workflow response converter
-        pipeline._workflow_response_converter.workflow_node_finish_to_stream_response = Mock(return_value=None)
-        pipeline._save_output_for_event = Mock()
-
-        # Act
-        list(pipeline._handle_node_succeeded_event(event))
-
-        # Assert: answer should remain unchanged
-        assert pipeline._task_state.answer == "same answer"
-        # Verify message_replace was NOT called
-        pipeline._message_cycle_manager.message_replace_to_stream_response.assert_not_called()
-
-    def test_answer_node_with_none_output_does_not_send_message_replace(self, pipeline):
-        """
-        Test that when an ANSWER node's output is None or missing 'answer' key,
-        no message_replace event is sent.
-        """
-        # Arrange: Set initial accumulated answer
-        pipeline._task_state.answer = "existing answer"
-
-        # Create ANSWER node succeeded event with None output
-        event = QueueNodeSucceededEvent(
-            node_execution_id="test-node-execution-id",
-            node_id="test-answer-node",
-            node_type=NodeType.ANSWER,
-            start_at=datetime.now(),
-            outputs={"answer": None},
-        )
-
-        # Mock the workflow response converter
-        pipeline._workflow_response_converter.workflow_node_finish_to_stream_response = Mock(return_value=None)
-        pipeline._save_output_for_event = Mock()
-
-        # Act
-        list(pipeline._handle_node_succeeded_event(event))
-
-        # Assert: answer should remain unchanged
-        assert pipeline._task_state.answer == "existing answer"
-        # Verify message_replace was NOT called
-        pipeline._message_cycle_manager.message_replace_to_stream_response.assert_not_called()
-
-    def test_answer_node_with_empty_outputs_does_not_send_message_replace(self, pipeline):
-        """
-        Test that when an ANSWER node has empty outputs dict,
-        no message_replace event is sent.
-        """
-        # Arrange: Set initial accumulated answer
-        pipeline._task_state.answer = "existing answer"
-
-        # Create ANSWER node succeeded event with empty outputs
-        event = QueueNodeSucceededEvent(
-            node_execution_id="test-node-execution-id",
-            node_id="test-answer-node",
-            node_type=NodeType.ANSWER,
-            start_at=datetime.now(),
-            outputs={},
-        )
-
-        # Mock the workflow response converter
-        pipeline._workflow_response_converter.workflow_node_finish_to_stream_response = Mock(return_value=None)
-        pipeline._save_output_for_event = Mock()
-
-        # Act
-        list(pipeline._handle_node_succeeded_event(event))
-
-        # Assert: answer should remain unchanged
-        assert pipeline._task_state.answer == "existing answer"
-        # Verify message_replace was NOT called
-        pipeline._message_cycle_manager.message_replace_to_stream_response.assert_not_called()
-
-    def test_answer_node_with_no_answer_key_in_outputs(self, pipeline):
-        """
-        Test that when an ANSWER node's outputs don't contain 'answer' key,
-        no message_replace event is sent.
-        """
-        # Arrange: Set initial accumulated answer
-        pipeline._task_state.answer = "existing answer"
-
-        # Create ANSWER node succeeded event without 'answer' key in outputs
-        event = QueueNodeSucceededEvent(
-            node_execution_id="test-node-execution-id",
-            node_id="test-answer-node",
-            node_type=NodeType.ANSWER,
-            start_at=datetime.now(),
-            outputs={"other_key": "some value"},
-        )
-
-        # Mock the workflow response converter
-        pipeline._workflow_response_converter.workflow_node_finish_to_stream_response = Mock(return_value=None)
-        pipeline._save_output_for_event = Mock()
-
-        # Act
-        list(pipeline._handle_node_succeeded_event(event))
-
-        # Assert: answer should remain unchanged
-        assert pipeline._task_state.answer == "existing answer"
-        # Verify message_replace was NOT called
-        pipeline._message_cycle_manager.message_replace_to_stream_response.assert_not_called()
-
-    def test_non_answer_node_does_not_send_message_replace(self, pipeline):
-        """
-        Test that non-ANSWER nodes (e.g., LLM, END) don't trigger message_replace events.
-        """
-        # Arrange: Set initial accumulated answer
-        pipeline._task_state.answer = "existing answer"
-
-        # Test with LLM node
-        llm_event = QueueNodeSucceededEvent(
-            node_execution_id="test-llm-execution-id",
-            node_id="test-llm-node",
-            node_type=NodeType.LLM,
-            start_at=datetime.now(),
-            outputs={"answer": "different answer"},
-        )
-
-        # Mock the workflow response converter
-        pipeline._workflow_response_converter.workflow_node_finish_to_stream_response = Mock(return_value=None)
-        pipeline._save_output_for_event = Mock()
-
-        # Act
-        list(pipeline._handle_node_succeeded_event(llm_event))
-
-        # Assert: answer should remain unchanged
-        assert pipeline._task_state.answer == "existing answer"
-        # Verify message_replace was NOT called
-        pipeline._message_cycle_manager.message_replace_to_stream_response.assert_not_called()
-
-    def test_end_node_does_not_send_message_replace(self, pipeline):
-        """
-        Test that END nodes don't trigger message_replace events even with 'answer' output.
-        """
-        # Arrange: Set initial accumulated answer
-        pipeline._task_state.answer = "existing answer"
-
-        # Create END node succeeded event with answer output
-        event = QueueNodeSucceededEvent(
-            node_execution_id="test-end-execution-id",
-            node_id="test-end-node",
-            node_type=NodeType.END,
-            start_at=datetime.now(),
-            outputs={"answer": "different answer"},
-        )
-
-        # Mock the workflow response converter
-        pipeline._workflow_response_converter.workflow_node_finish_to_stream_response = Mock(return_value=None)
-        pipeline._save_output_for_event = Mock()
-
-        # Act
-        list(pipeline._handle_node_succeeded_event(event))
-
-        # Assert: answer should remain unchanged
-        assert pipeline._task_state.answer == "existing answer"
-        # Verify message_replace was NOT called
-        pipeline._message_cycle_manager.message_replace_to_stream_response.assert_not_called()
-
-    def test_answer_node_with_numeric_output_converts_to_string(self, pipeline):
-        """
-        Test that when an ANSWER node's final output is numeric,
-        it gets converted to string properly.
-        """
-        # Arrange: Set initial accumulated answer
-        pipeline._task_state.answer = "text answer"
-
-        # Create ANSWER node succeeded event with numeric output
-        event = QueueNodeSucceededEvent(
-            node_execution_id="test-node-execution-id",
-            node_id="test-answer-node",
-            node_type=NodeType.ANSWER,
-            start_at=datetime.now(),
-            outputs={"answer": 12345},
-        )
-
-        # Mock the workflow response converter
-        pipeline._workflow_response_converter.workflow_node_finish_to_stream_response = Mock(return_value=None)
-        pipeline._save_output_for_event = Mock()
-
-        # Act
-        list(pipeline._handle_node_succeeded_event(event))
-
-        # Assert: answer should be converted to string
-        assert pipeline._task_state.answer == "12345"
-        # Verify message_replace was called with string
-        pipeline._message_cycle_manager.message_replace_to_stream_response.assert_called_once_with(
-            answer="12345", reason="variable_update"
-        )
-
-    def test_answer_node_files_are_recorded(self, pipeline):
-        """
-        Test that ANSWER nodes properly record files from outputs.
-        """
-        # Arrange
-        pipeline._task_state.answer = "existing answer"
-
-        # Create ANSWER node succeeded event with files
-        event = QueueNodeSucceededEvent(
-            node_execution_id="test-node-execution-id",
-            node_id="test-answer-node",
-            node_type=NodeType.ANSWER,
-            start_at=datetime.now(),
-            outputs={
-                "answer": "same answer",
-                "files": [
-                    {"type": "image", "transfer_method": "remote_url", "remote_url": "http://example.com/img.png"}
-                ],
-            },
-        )
-
-        # Mock the workflow response converter
-        pipeline._workflow_response_converter.fetch_files_from_node_outputs = Mock(return_value=event.outputs["files"])
-        pipeline._workflow_response_converter.workflow_node_finish_to_stream_response = Mock(return_value=None)
-        pipeline._save_output_for_event = Mock()
-
-        # Act
-        list(pipeline._handle_node_succeeded_event(event))
-
-        # Assert: files should be recorded
-        assert len(pipeline._recorded_files) == 1
-        assert pipeline._recorded_files[0] == event.outputs["files"][0]

api/tests/unit_tests/core/workflow/nodes/test_start_node_json_object.py

@@ -58,6 +58,8 @@ def test_json_object_valid_schema():
         }
     )
 
+    schema = json.loads(schema)
+
     variables = [
         VariableEntity(
             variable="profile",
@@ -68,7 +70,7 @@ def test_json_object_valid_schema():
         )
     ]
 
-    user_inputs = {"profile": json.dumps({"age": 20, "name": "Tom"})}
+    user_inputs = {"profile": {"age": 20, "name": "Tom"}}
 
     node = make_start_node(user_inputs, variables)
     result = node._run()
@@ -87,6 +89,8 @@ def test_json_object_invalid_json_string():
             "required": ["age", "name"],
         }
     )
+
+    schema = json.loads(schema)
     variables = [
         VariableEntity(
             variable="profile",
@@ -97,12 +101,12 @@ def test_json_object_invalid_json_string():
         )
     ]
 
-    # Missing closing brace makes this invalid JSON
+    # Providing a string instead of an object should raise a type error
     user_inputs = {"profile": '{"age": 20, "name": "Tom"'}
 
     node = make_start_node(user_inputs, variables)
 
-    with pytest.raises(ValueError, match='{"age": 20, "name": "Tom" must be a valid JSON object'):
+    with pytest.raises(ValueError, match="JSON object for 'profile' must be an object"):
         node._run()
 
 
@@ -118,6 +122,8 @@ def test_json_object_does_not_match_schema():
         }
     )
 
+    schema = json.loads(schema)
+
     variables = [
         VariableEntity(
             variable="profile",
@@ -129,7 +135,7 @@ def test_json_object_does_not_match_schema():
     ]
 
     # age is a string, which violates the schema (expects number)
-    user_inputs = {"profile": json.dumps({"age": "twenty", "name": "Tom"})}
+    user_inputs = {"profile": {"age": "twenty", "name": "Tom"}}
 
     node = make_start_node(user_inputs, variables)
 
@@ -149,6 +155,8 @@ def test_json_object_missing_required_schema_field():
         }
     )
 
+    schema = json.loads(schema)
+
     variables = [
         VariableEntity(
             variable="profile",
@@ -160,7 +168,7 @@ def test_json_object_missing_required_schema_field():
     ]
 
     # Missing required field "name"
-    user_inputs = {"profile": json.dumps({"age": 20})}
+    user_inputs = {"profile": {"age": 20}}
 
     node = make_start_node(user_inputs, variables)
 

web/app/(commonLayout)/plugins/page.tsx

@@ -2,11 +2,11 @@ import Marketplace from '@/app/components/plugins/marketplace'
 import PluginPage from '@/app/components/plugins/plugin-page'
 import PluginsPanel from '@/app/components/plugins/plugin-page/plugins-panel'
 
-const PluginList = async () => {
+const PluginList = () => {
   return (
     <PluginPage
       plugins={<PluginsPanel />}
-      marketplace={<Marketplace pluginTypeSwitchClassName="top-[60px]" showSearchParams={false} />}
+      marketplace={<Marketplace pluginTypeSwitchClassName="top-[60px]" />}
     />
   )
 }

web/app/components/app-sidebar/app-info.tsx

@@ -26,6 +26,7 @@ import { NEED_REFRESH_APP_LIST_KEY } from '@/config'
 import { useAppContext } from '@/context/app-context'
 import { useProviderContext } from '@/context/provider-context'
 import { copyApp, deleteApp, exportAppConfig, updateAppInfo } from '@/service/apps'
+import { useInvalidateAppList } from '@/service/use-apps'
 import { fetchWorkflowDraft } from '@/service/workflow'
 import { AppModeEnum } from '@/types/app'
 import { getRedirection } from '@/utils/app-redirection'
@@ -66,6 +67,7 @@ const AppInfo = ({ expand, onlyShowDetail = false, openState = false, onDetailEx
   const { onPlanInfoChanged } = useProviderContext()
   const appDetail = useAppStore(state => state.appDetail)
   const setAppDetail = useAppStore(state => state.setAppDetail)
+  const invalidateAppList = useInvalidateAppList()
   const [open, setOpen] = useState(openState)
   const [showEditModal, setShowEditModal] = useState(false)
   const [showDuplicateModal, setShowDuplicateModal] = useState(false)
@@ -191,6 +193,7 @@ const AppInfo = ({ expand, onlyShowDetail = false, openState = false, onDetailEx
     try {
       await deleteApp(appDetail.id)
       notify({ type: 'success', message: t('appDeleted', { ns: 'app' }) })
+      invalidateAppList()
       onPlanInfoChanged()
       setAppDetail()
       replace('/apps')
@@ -202,7 +205,7 @@ const AppInfo = ({ expand, onlyShowDetail = false, openState = false, onDetailEx
       })
     }
     setShowConfirmDelete(false)
-  }, [appDetail, notify, onPlanInfoChanged, replace, setAppDetail, t])
+  }, [appDetail, invalidateAppList, notify, onPlanInfoChanged, replace, setAppDetail, t])
 
   const { isCurrentWorkspaceEditor } = useAppContext()
 

web/app/components/app/configuration/config-var/config-modal/index.tsx

@@ -83,7 +83,7 @@ const ConfigModal: FC<IConfigModalProps> = ({
     if (!isJsonObject || !tempPayload.json_schema)
       return ''
     try {
-      return JSON.stringify(JSON.parse(tempPayload.json_schema), null, 2)
+      return tempPayload.json_schema
     }
     catch {
       return ''

web/app/components/apps/list.spec.tsx

@@ -10,6 +10,7 @@ const mockReplace = vi.fn()
 const mockRouter = { replace: mockReplace }
 vi.mock('next/navigation', () => ({
   useRouter: () => mockRouter,
+  useSearchParams: () => new URLSearchParams(''),
 }))
 
 // Mock app context

web/app/components/apps/list.tsx

@@ -12,6 +12,7 @@ import { useDebounceFn } from 'ahooks'
 import dynamic from 'next/dynamic'
 import {
   useRouter,
+  useSearchParams,
 } from 'next/navigation'
 import { parseAsString, useQueryState } from 'nuqs'
 import { useCallback, useEffect, useRef, useState } from 'react'
@@ -36,6 +37,16 @@ import useAppsQueryState from './hooks/use-apps-query-state'
 import { useDSLDragDrop } from './hooks/use-dsl-drag-drop'
 import NewAppCard from './new-app-card'
 
+// Define valid tabs at module scope to avoid re-creation on each render and stale closures
+const validTabs = new Set<string | AppModeEnum>([
+  'all',
+  AppModeEnum.WORKFLOW,
+  AppModeEnum.ADVANCED_CHAT,
+  AppModeEnum.CHAT,
+  AppModeEnum.AGENT_CHAT,
+  AppModeEnum.COMPLETION,
+])
+
 const TagManagementModal = dynamic(() => import('@/app/components/base/tag-management'), {
   ssr: false,
 })
@@ -47,12 +58,41 @@ const List = () => {
   const { t } = useTranslation()
   const { systemFeatures } = useGlobalPublicStore()
   const router = useRouter()
+  const searchParams = useSearchParams()
   const { isCurrentWorkspaceEditor, isCurrentWorkspaceDatasetOperator, isLoadingCurrentWorkspace } = useAppContext()
   const showTagManagementModal = useTagStore(s => s.showTagManagementModal)
   const [activeTab, setActiveTab] = useQueryState(
     'category',
     parseAsString.withDefault('all').withOptions({ history: 'push' }),
   )
+
+  // valid tabs for apps list; anything else should fallback to 'all'
+
+  // 1) Normalize legacy/incorrect query params like ?mode=discover -> ?category=all
+  useEffect(() => {
+    // avoid running on server
+    if (typeof window === 'undefined')
+      return
+    const mode = searchParams.get('mode')
+    if (!mode)
+      return
+    const url = new URL(window.location.href)
+    url.searchParams.delete('mode')
+    if (validTabs.has(mode)) {
+      // migrate to category key
+      url.searchParams.set('category', mode)
+    }
+    else {
+      url.searchParams.set('category', 'all')
+    }
+    router.replace(url.pathname + url.search)
+  }, [router, searchParams])
+
+  // 2) If category has an invalid value (e.g., 'discover'), reset to 'all'
+  useEffect(() => {
+    if (!validTabs.has(activeTab))
+      setActiveTab('all')
+  }, [activeTab, setActiveTab])
   const { query: { tagIDs = [], keywords = '', isCreatedByMe: queryIsCreatedByMe = false }, setQuery } = useAppsQueryState()
   const [isCreatedByMe, setIsCreatedByMe] = useState(queryIsCreatedByMe)
   const [tagFilterValue, setTagFilterValue] = useState<string[]>(tagIDs)

web/app/components/base/chat/chat/utils.ts

@@ -37,7 +37,7 @@ export const getProcessedInputs = (inputs: Record<string, any>, inputsForm: Inpu
       return
     }
 
-    if (!inputValue)
+    if (inputValue == null)
       return
 
     if (item.type === InputVarType.singleFile) {
@@ -52,6 +52,20 @@ export const getProcessedInputs = (inputs: Record<string, any>, inputsForm: Inpu
       else
         processedInputs[item.variable] = getProcessedFiles(inputValue)
     }
+    else if (item.type === InputVarType.jsonObject) {
+      // Prefer sending an object if the user entered valid JSON; otherwise keep the raw string.
+      try {
+        const v = typeof inputValue === 'string' ? JSON.parse(inputValue) : inputValue
+        if (v && typeof v === 'object' && !Array.isArray(v))
+          processedInputs[item.variable] = v
+        else
+          processedInputs[item.variable] = inputValue
+      }
+      catch {
+        // keep original string; backend will parse/validate
+        processedInputs[item.variable] = inputValue
+      }
+    }
   })
 
   return processedInputs

web/app/components/datasets/create/embedding-process/index.tsx

@@ -1,47 +1,29 @@
 import type { FC } from 'react'
-import type {
-  DataSourceInfo,
-  FullDocumentDetail,
-  IndexingStatusResponse,
-  LegacyDataSourceInfo,
-  ProcessRuleResponse,
-} from '@/models/datasets'
+import type { FullDocumentDetail } from '@/models/datasets'
+import type { RETRIEVE_METHOD } from '@/types/app'
 import {
   RiArrowRightLine,
-  RiCheckboxCircleFill,
-  RiErrorWarningFill,
   RiLoader2Fill,
   RiTerminalBoxLine,
 } from '@remixicon/react'
-import Image from 'next/image'
 import Link from 'next/link'
 import { useRouter } from 'next/navigation'
-import * as React from 'react'
-import { useCallback, useEffect, useMemo, useRef, useState } from 'react'
+import { useMemo } from 'react'
 import { useTranslation } from 'react-i18next'
 import Button from '@/app/components/base/button'
 import Divider from '@/app/components/base/divider'
-import { ZapFast } from '@/app/components/base/icons/src/vender/solid/general'
-import NotionIcon from '@/app/components/base/notion-icon'
-import Tooltip from '@/app/components/base/tooltip'
-import PriorityLabel from '@/app/components/billing/priority-label'
 import { Plan } from '@/app/components/billing/type'
-import UpgradeBtn from '@/app/components/billing/upgrade-btn'
-import { FieldInfo } from '@/app/components/datasets/documents/detail/metadata'
 import { useProviderContext } from '@/context/provider-context'
 import { useDatasetApiAccessUrl } from '@/hooks/use-api-access-url'
-import { DataSourceType, ProcessMode } from '@/models/datasets'
-import { fetchIndexingStatusBatch as doFetchIndexingStatus } from '@/service/datasets'
 import { useProcessRule } from '@/service/knowledge/use-dataset'
 import { useInvalidDocumentList } from '@/service/knowledge/use-document'
-import { RETRIEVE_METHOD } from '@/types/app'
-import { sleep } from '@/utils'
-import { cn } from '@/utils/classnames'
-import DocumentFileIcon from '../../common/document-file-icon'
-import { indexMethodIcon, retrievalIcon } from '../icons'
-import { IndexingType } from '../step-two'
+import IndexingProgressItem from './indexing-progress-item'
+import RuleDetail from './rule-detail'
+import UpgradeBanner from './upgrade-banner'
+import { useIndexingStatusPolling } from './use-indexing-status-polling'
+import { createDocumentLookup } from './utils'
 
-type Props = {
+type EmbeddingProcessProps = {
   datasetId: string
   batchId: string
   documents?: FullDocumentDetail[]
@@ -49,333 +31,121 @@ type Props = {
   retrievalMethod?: RETRIEVE_METHOD
 }
 
-const RuleDetail: FC<{
-  sourceData?: ProcessRuleResponse
-  indexingType?: string
-  retrievalMethod?: RETRIEVE_METHOD
-}> = ({ sourceData, indexingType, retrievalMethod }) => {
+// Status header component
+const StatusHeader: FC<{ isEmbedding: boolean, isCompleted: boolean }> = ({
+  isEmbedding,
+  isCompleted,
+}) => {
   const { t } = useTranslation()
 
-  const segmentationRuleMap = {
-    mode: t('embedding.mode', { ns: 'datasetDocuments' }),
-    segmentLength: t('embedding.segmentLength', { ns: 'datasetDocuments' }),
-    textCleaning: t('embedding.textCleaning', { ns: 'datasetDocuments' }),
-  }
-
-  const getRuleName = (key: string) => {
-    if (key === 'remove_extra_spaces')
-      return t('stepTwo.removeExtraSpaces', { ns: 'datasetCreation' })
-
-    if (key === 'remove_urls_emails')
-      return t('stepTwo.removeUrlEmails', { ns: 'datasetCreation' })
-
-    if (key === 'remove_stopwords')
-      return t('stepTwo.removeStopwords', { ns: 'datasetCreation' })
-  }
-
-  const isNumber = (value: unknown) => {
-    return typeof value === 'number'
-  }
-
-  const getValue = useCallback((field: string) => {
-    let value: string | number | undefined = '-'
-    const maxTokens = isNumber(sourceData?.rules?.segmentation?.max_tokens)
-      ? sourceData.rules.segmentation.max_tokens
-      : value
-    const childMaxTokens = isNumber(sourceData?.rules?.subchunk_segmentation?.max_tokens)
-      ? sourceData.rules.subchunk_segmentation.max_tokens
-      : value
-    switch (field) {
-      case 'mode':
-        value = !sourceData?.mode
-          ? value
-          : sourceData.mode === ProcessMode.general
-            ? (t('embedding.custom', { ns: 'datasetDocuments' }) as string)
-            : `${t('embedding.hierarchical', { ns: 'datasetDocuments' })} · ${sourceData?.rules?.parent_mode === 'paragraph'
-              ? t('parentMode.paragraph', { ns: 'dataset' })
-              : t('parentMode.fullDoc', { ns: 'dataset' })}`
-        break
-      case 'segmentLength':
-        value = !sourceData?.mode
-          ? value
-          : sourceData.mode === ProcessMode.general
-            ? maxTokens
-            : `${t('embedding.parentMaxTokens', { ns: 'datasetDocuments' })} ${maxTokens}; ${t('embedding.childMaxTokens', { ns: 'datasetDocuments' })} ${childMaxTokens}`
-        break
-      default:
-        value = !sourceData?.mode
-          ? value
-          : sourceData?.rules?.pre_processing_rules?.filter(rule =>
-              rule.enabled).map(rule => getRuleName(rule.id)).join(',')
-        break
-    }
-    return value
-  }, [sourceData])
-
   return (
-    <div className="flex flex-col gap-1">
-      {Object.keys(segmentationRuleMap).map((field) => {
-        return (
-          <FieldInfo
-            key={field}
-            label={segmentationRuleMap[field as keyof typeof segmentationRuleMap]}
-            displayedValue={String(getValue(field))}
-          />
-        )
-      })}
-      <FieldInfo
-        label={t('stepTwo.indexMode', { ns: 'datasetCreation' })}
-        displayedValue={t(`stepTwo.${indexingType === IndexingType.ECONOMICAL ? 'economical' : 'qualified'}`, { ns: 'datasetCreation' }) as string}
-        valueIcon={(
-          <Image
-            className="size-4"
-            src={
-              indexingType === IndexingType.ECONOMICAL
-                ? indexMethodIcon.economical
-                : indexMethodIcon.high_quality
-            }
-            alt=""
-          />
-        )}
-      />
-      <FieldInfo
-        label={t('form.retrievalSetting.title', { ns: 'datasetSettings' })}
-        // displayedValue={t(`datasetSettings.form.retrievalSetting.${retrievalMethod}`) as string}
-        displayedValue={t(`retrieval.${indexingType === IndexingType.ECONOMICAL ? 'keyword_search' : retrievalMethod ?? 'semantic_search'}.title`, { ns: 'dataset' })}
-        valueIcon={(
-          <Image
-            className="size-4"
-            src={
-              retrievalMethod === RETRIEVE_METHOD.fullText
-                ? retrievalIcon.fullText
-                : retrievalMethod === RETRIEVE_METHOD.hybrid
-                  ? retrievalIcon.hybrid
-                  : retrievalIcon.vector
-            }
-            alt=""
-          />
-        )}
-      />
+    <div className="system-md-semibold-uppercase flex items-center gap-x-1 text-text-secondary">
+      {isEmbedding && (
+        <>
+          <RiLoader2Fill className="size-4 animate-spin" />
+          <span>{t('embedding.processing', { ns: 'datasetDocuments' })}</span>
+        </>
+      )}
+      {isCompleted && t('embedding.completed', { ns: 'datasetDocuments' })}
     </div>
   )
 }
 
-const EmbeddingProcess: FC<Props> = ({ datasetId, batchId, documents = [], indexingType, retrievalMethod }) => {
+// Action buttons component
+const ActionButtons: FC<{
+  apiReferenceUrl: string
+  onNavToDocuments: () => void
+}> = ({ apiReferenceUrl, onNavToDocuments }) => {
   const { t } = useTranslation()
+
+  return (
+    <div className="mt-6 flex items-center gap-x-2 py-2">
+      <Link href={apiReferenceUrl} target="_blank" rel="noopener noreferrer">
+        <Button className="w-fit gap-x-0.5 px-3">
+          <RiTerminalBoxLine className="size-4" />
+          <span className="px-0.5">Access the API</span>
+        </Button>
+      </Link>
+      <Button
+        className="w-fit gap-x-0.5 px-3"
+        variant="primary"
+        onClick={onNavToDocuments}
+      >
+        <span className="px-0.5">{t('stepThree.navTo', { ns: 'datasetCreation' })}</span>
+        <RiArrowRightLine className="size-4 stroke-current stroke-1" />
+      </Button>
+    </div>
+  )
+}
+
+const EmbeddingProcess: FC<EmbeddingProcessProps> = ({
+  datasetId,
+  batchId,
+  documents = [],
+  indexingType,
+  retrievalMethod,
+}) => {
   const { enableBilling, plan } = useProviderContext()
-
-  const getFirstDocument = documents[0]
-
-  const [indexingStatusBatchDetail, setIndexingStatusDetail] = useState<IndexingStatusResponse[]>([])
-  const fetchIndexingStatus = async () => {
-    const status = await doFetchIndexingStatus({ datasetId, batchId })
-    setIndexingStatusDetail(status.data)
-    return status.data
-  }
-
-  const [isStopQuery, setIsStopQuery] = useState(false)
-  const isStopQueryRef = useRef(isStopQuery)
-  useEffect(() => {
-    isStopQueryRef.current = isStopQuery
-  }, [isStopQuery])
-  const stopQueryStatus = () => {
-    setIsStopQuery(true)
-  }
-
-  const startQueryStatus = async () => {
-    if (isStopQueryRef.current)
-      return
-
-    try {
-      const indexingStatusBatchDetail = await fetchIndexingStatus()
-      const isCompleted = indexingStatusBatchDetail.every(indexingStatusDetail => ['completed', 'error', 'paused'].includes(indexingStatusDetail.indexing_status))
-      if (isCompleted) {
-        stopQueryStatus()
-        return
-      }
-      await sleep(2500)
-      await startQueryStatus()
-    }
-    catch {
-      await sleep(2500)
-      await startQueryStatus()
-    }
-  }
-
-  useEffect(() => {
-    setIsStopQuery(false)
-    startQueryStatus()
-    return () => {
-      stopQueryStatus()
-    }
-  }, [])
-
-  // get rule
-  const { data: ruleDetail } = useProcessRule(getFirstDocument?.id)
-
   const router = useRouter()
   const invalidDocumentList = useInvalidDocumentList()
-  const navToDocumentList = () => {
+  const apiReferenceUrl = useDatasetApiAccessUrl()
+
+  // Polling hook for indexing status
+  const { statusList, isEmbedding, isEmbeddingCompleted } = useIndexingStatusPolling({
+    datasetId,
+    batchId,
+  })
+
+  // Get process rule for the first document
+  const firstDocumentId = documents[0]?.id
+  const { data: ruleDetail } = useProcessRule(firstDocumentId)
+
+  // Document lookup utilities - memoized for performance
+  const documentLookup = useMemo(
+    () => createDocumentLookup(documents),
+    [documents],
+  )
+
+  const handleNavToDocuments = () => {
     invalidDocumentList()
     router.push(`/datasets/${datasetId}/documents`)
   }
-  const apiReferenceUrl = useDatasetApiAccessUrl()
 
-  const isEmbedding = useMemo(() => {
-    return indexingStatusBatchDetail.some(indexingStatusDetail => ['indexing', 'splitting', 'parsing', 'cleaning'].includes(indexingStatusDetail?.indexing_status || ''))
-  }, [indexingStatusBatchDetail])
-  const isEmbeddingCompleted = useMemo(() => {
-    return indexingStatusBatchDetail.every(indexingStatusDetail => ['completed', 'error', 'paused'].includes(indexingStatusDetail?.indexing_status || ''))
-  }, [indexingStatusBatchDetail])
-
-  const getSourceName = (id: string) => {
-    const doc = documents.find(document => document.id === id)
-    return doc?.name
-  }
-  const getFileType = (name?: string) => name?.split('.').pop() || 'txt'
-  const getSourcePercent = (detail: IndexingStatusResponse) => {
-    const completedCount = detail.completed_segments || 0
-    const totalCount = detail.total_segments || 0
-    if (totalCount === 0)
-      return 0
-    const percent = Math.round(completedCount * 100 / totalCount)
-    return percent > 100 ? 100 : percent
-  }
-  const getSourceType = (id: string) => {
-    const doc = documents.find(document => document.id === id)
-    return doc?.data_source_type as DataSourceType
-  }
-
-  const isLegacyDataSourceInfo = (info: DataSourceInfo): info is LegacyDataSourceInfo => {
-    return info != null && typeof (info as LegacyDataSourceInfo).upload_file === 'object'
-  }
-
-  const getIcon = (id: string) => {
-    const doc = documents.find(document => document.id === id)
-    const info = doc?.data_source_info
-    if (info && isLegacyDataSourceInfo(info))
-      return info.notion_page_icon
-    return undefined
-  }
-  const isSourceEmbedding = (detail: IndexingStatusResponse) =>
-    ['indexing', 'splitting', 'parsing', 'cleaning', 'waiting'].includes(detail.indexing_status || '')
+  const showUpgradeBanner = enableBilling && plan.type !== Plan.team
 
   return (
     <>
       <div className="flex flex-col gap-y-3">
-        <div className="system-md-semibold-uppercase flex items-center gap-x-1 text-text-secondary">
-          {isEmbedding && (
-            <>
-              <RiLoader2Fill className="size-4 animate-spin" />
-              <span>{t('embedding.processing', { ns: 'datasetDocuments' })}</span>
-            </>
-          )}
-          {isEmbeddingCompleted && t('embedding.completed', { ns: 'datasetDocuments' })}
-        </div>
-        {
-          enableBilling && plan.type !== Plan.team && (
-            <div className="flex h-14 items-center rounded-xl border-[0.5px] border-black/5 bg-white p-3 shadow-md">
-              <div className="flex h-8 w-8 shrink-0 items-center justify-center rounded-lg bg-[#FFF6ED]">
-                <ZapFast className="h-4 w-4 text-[#FB6514]" />
-              </div>
-              <div className="mx-3 grow text-[13px] font-medium text-gray-700">
-                {t('plansCommon.documentProcessingPriorityUpgrade', { ns: 'billing' })}
-              </div>
-              <UpgradeBtn loc="knowledge-speed-up" />
-            </div>
-          )
-        }
+        <StatusHeader isEmbedding={isEmbedding} isCompleted={isEmbeddingCompleted} />
+
+        {showUpgradeBanner && <UpgradeBanner />}
+
         <div className="flex flex-col gap-0.5 pb-2">
-          {indexingStatusBatchDetail.map(indexingStatusDetail => (
-            <div
-              key={indexingStatusDetail.id}
-              className={cn(
-                'relative h-[26px] overflow-hidden rounded-md bg-components-progress-bar-bg',
-                indexingStatusDetail.indexing_status === 'error' && 'bg-state-destructive-hover-alt',
-              )}
-            >
-              {isSourceEmbedding(indexingStatusDetail) && (
-                <div
-                  className="absolute left-0 top-0 h-full min-w-0.5 border-r-[2px] border-r-components-progress-bar-progress-highlight bg-components-progress-bar-progress"
-                  style={{ width: `${getSourcePercent(indexingStatusDetail)}%` }}
-                />
-              )}
-              <div className="z-[1] flex h-full items-center gap-1 pl-[6px] pr-2">
-                {getSourceType(indexingStatusDetail.id) === DataSourceType.FILE && (
-                  <DocumentFileIcon
-                    size="sm"
-                    className="shrink-0"
-                    name={getSourceName(indexingStatusDetail.id)}
-                    extension={getFileType(getSourceName(indexingStatusDetail.id))}
-                  />
-                )}
-                {getSourceType(indexingStatusDetail.id) === DataSourceType.NOTION && (
-                  <NotionIcon
-                    className="shrink-0"
-                    type="page"
-                    src={getIcon(indexingStatusDetail.id)}
-                  />
-                )}
-                <div className="flex w-0 grow items-center gap-1" title={getSourceName(indexingStatusDetail.id)}>
-                  <div className="system-xs-medium truncate text-text-secondary">
-                    {getSourceName(indexingStatusDetail.id)}
-                  </div>
-                  {
-                    enableBilling && (
-                      <PriorityLabel className="ml-0" />
-                    )
-                  }
-                </div>
-                {isSourceEmbedding(indexingStatusDetail) && (
-                  <div className="shrink-0 text-xs text-text-secondary">{`${getSourcePercent(indexingStatusDetail)}%`}</div>
-                )}
-                {indexingStatusDetail.indexing_status === 'error' && (
-                  <Tooltip
-                    popupClassName="px-4 py-[14px] max-w-60 body-xs-regular text-text-secondary border-[0.5px] border-components-panel-border rounded-xl"
-                    offset={4}
-                    popupContent={indexingStatusDetail.error}
-                  >
-                    <span>
-                      <RiErrorWarningFill className="size-4 shrink-0 text-text-destructive" />
-                    </span>
-                  </Tooltip>
-                )}
-                {indexingStatusDetail.indexing_status === 'completed' && (
-                  <RiCheckboxCircleFill className="size-4 shrink-0 text-text-success" />
-                )}
-              </div>
-            </div>
+          {statusList.map(detail => (
+            <IndexingProgressItem
+              key={detail.id}
+              detail={detail}
+              name={documentLookup.getName(detail.id)}
+              sourceType={documentLookup.getSourceType(detail.id)}
+              notionIcon={documentLookup.getNotionIcon(detail.id)}
+              enableBilling={enableBilling}
+            />
           ))}
         </div>
+
         <Divider type="horizontal" className="my-0 bg-divider-subtle" />
+
         <RuleDetail
           sourceData={ruleDetail}
           indexingType={indexingType}
           retrievalMethod={retrievalMethod}
         />
       </div>
-      <div className="mt-6 flex items-center gap-x-2 py-2">
-        <Link
-          href={apiReferenceUrl}
-          target="_blank"
-          rel="noopener noreferrer"
-        >
-          <Button
-            className="w-fit gap-x-0.5 px-3"
-          >
-            <RiTerminalBoxLine className="size-4" />
-            <span className="px-0.5">Access the API</span>
-          </Button>
-        </Link>
-        <Button
-          className="w-fit gap-x-0.5 px-3"
-          variant="primary"
-          onClick={navToDocumentList}
-        >
-          <span className="px-0.5">{t('stepThree.navTo', { ns: 'datasetCreation' })}</span>
-          <RiArrowRightLine className="size-4 stroke-current stroke-1" />
-        </Button>
-      </div>
+
+      <ActionButtons
+        apiReferenceUrl={apiReferenceUrl}
+        onNavToDocuments={handleNavToDocuments}
+      />
     </>
   )
 }

web/app/components/datasets/create/embedding-process/indexing-progress-item.tsx

@@ -0,0 +1,120 @@
+import type { FC } from 'react'
+import type { IndexingStatusResponse } from '@/models/datasets'
+import {
+  RiCheckboxCircleFill,
+  RiErrorWarningFill,
+} from '@remixicon/react'
+import NotionIcon from '@/app/components/base/notion-icon'
+import Tooltip from '@/app/components/base/tooltip'
+import PriorityLabel from '@/app/components/billing/priority-label'
+import { DataSourceType } from '@/models/datasets'
+import { cn } from '@/utils/classnames'
+import DocumentFileIcon from '../../common/document-file-icon'
+import { getFileType, getSourcePercent, isSourceEmbedding } from './utils'
+
+type IndexingProgressItemProps = {
+  detail: IndexingStatusResponse
+  name?: string
+  sourceType?: DataSourceType
+  notionIcon?: string
+  enableBilling?: boolean
+}
+
+// Status icon component for completed/error states
+const StatusIcon: FC<{ status: string, error?: string }> = ({ status, error }) => {
+  if (status === 'completed')
+    return <RiCheckboxCircleFill className="size-4 shrink-0 text-text-success" />
+
+  if (status === 'error') {
+    return (
+      <Tooltip
+        popupClassName="px-4 py-[14px] max-w-60 body-xs-regular text-text-secondary border-[0.5px] border-components-panel-border rounded-xl"
+        offset={4}
+        popupContent={error}
+      >
+        <span>
+          <RiErrorWarningFill className="size-4 shrink-0 text-text-destructive" />
+        </span>
+      </Tooltip>
+    )
+  }
+
+  return null
+}
+
+// Source type icon component
+const SourceTypeIcon: FC<{
+  sourceType?: DataSourceType
+  name?: string
+  notionIcon?: string
+}> = ({ sourceType, name, notionIcon }) => {
+  if (sourceType === DataSourceType.FILE) {
+    return (
+      <DocumentFileIcon
+        size="sm"
+        className="shrink-0"
+        name={name}
+        extension={getFileType(name)}
+      />
+    )
+  }
+
+  if (sourceType === DataSourceType.NOTION) {
+    return (
+      <NotionIcon
+        className="shrink-0"
+        type="page"
+        src={notionIcon}
+      />
+    )
+  }
+
+  return null
+}
+
+const IndexingProgressItem: FC<IndexingProgressItemProps> = ({
+  detail,
+  name,
+  sourceType,
+  notionIcon,
+  enableBilling,
+}) => {
+  const isEmbedding = isSourceEmbedding(detail)
+  const percent = getSourcePercent(detail)
+  const isError = detail.indexing_status === 'error'
+
+  return (
+    <div
+      className={cn(
+        'relative h-[26px] overflow-hidden rounded-md bg-components-progress-bar-bg',
+        isError && 'bg-state-destructive-hover-alt',
+      )}
+    >
+      {isEmbedding && (
+        <div
+          className="absolute left-0 top-0 h-full min-w-0.5 border-r-[2px] border-r-components-progress-bar-progress-highlight bg-components-progress-bar-progress"
+          style={{ width: `${percent}%` }}
+        />
+      )}
+      <div className="z-[1] flex h-full items-center gap-1 pl-[6px] pr-2">
+        <SourceTypeIcon
+          sourceType={sourceType}
+          name={name}
+          notionIcon={notionIcon}
+        />
+        <div className="flex w-0 grow items-center gap-1" title={name}>
+          <div className="system-xs-medium truncate text-text-secondary">
+            {name}
+          </div>
+          {enableBilling && <PriorityLabel className="ml-0" />}
+        </div>
+        {isEmbedding && (
+          <div className="shrink-0 text-xs text-text-secondary">{`${percent}%`}</div>
+        )}
+        <StatusIcon status={detail.indexing_status} error={detail.error} />
+      </div>
+    </div>
+  )
+}
+
+export default IndexingProgressItem

web/app/components/datasets/create/embedding-process/rule-detail.tsx

@@ -0,0 +1,133 @@
+import type { FC } from 'react'
+import type { ProcessRuleResponse } from '@/models/datasets'
+import Image from 'next/image'
+import { useCallback } from 'react'
+import { useTranslation } from 'react-i18next'
+import { FieldInfo } from '@/app/components/datasets/documents/detail/metadata'
+import { ProcessMode } from '@/models/datasets'
+import { RETRIEVE_METHOD } from '@/types/app'
+import { indexMethodIcon, retrievalIcon } from '../icons'
+import { IndexingType } from '../step-two'
+
+type RuleDetailProps = {
+  sourceData?: ProcessRuleResponse
+  indexingType?: string
+  retrievalMethod?: RETRIEVE_METHOD
+}
+
+// Lookup table for pre-processing rule names
+const PRE_PROCESSING_RULE_KEYS = {
+  remove_extra_spaces: 'stepTwo.removeExtraSpaces',
+  remove_urls_emails: 'stepTwo.removeUrlEmails',
+  remove_stopwords: 'stepTwo.removeStopwords',
+} as const
+
+// Lookup table for retrieval method icons
+const RETRIEVAL_ICON_MAP: Partial<Record<RETRIEVE_METHOD, string>> = {
+  [RETRIEVE_METHOD.fullText]: retrievalIcon.fullText,
+  [RETRIEVE_METHOD.hybrid]: retrievalIcon.hybrid,
+  [RETRIEVE_METHOD.semantic]: retrievalIcon.vector,
+  [RETRIEVE_METHOD.invertedIndex]: retrievalIcon.fullText,
+  [RETRIEVE_METHOD.keywordSearch]: retrievalIcon.fullText,
+}
+
+const isNumber = (value: unknown): value is number => typeof value === 'number'
+
+const RuleDetail: FC<RuleDetailProps> = ({ sourceData, indexingType, retrievalMethod }) => {
+  const { t } = useTranslation()
+
+  const segmentationRuleLabels = {
+    mode: t('embedding.mode', { ns: 'datasetDocuments' }),
+    segmentLength: t('embedding.segmentLength', { ns: 'datasetDocuments' }),
+    textCleaning: t('embedding.textCleaning', { ns: 'datasetDocuments' }),
+  }
+
+  const getRuleName = useCallback((key: string): string | undefined => {
+    const translationKey = PRE_PROCESSING_RULE_KEYS[key as keyof typeof PRE_PROCESSING_RULE_KEYS]
+    return translationKey ? t(translationKey, { ns: 'datasetCreation' }) : undefined
+  }, [t])
+
+  const getModeValue = useCallback((): string => {
+    if (!sourceData?.mode)
+      return '-'
+
+    if (sourceData.mode === ProcessMode.general)
+      return t('embedding.custom', { ns: 'datasetDocuments' })
+
+    const parentModeLabel = sourceData.rules?.parent_mode === 'paragraph'
+      ? t('parentMode.paragraph', { ns: 'dataset' })
+      : t('parentMode.fullDoc', { ns: 'dataset' })
+
+    return `${t('embedding.hierarchical', { ns: 'datasetDocuments' })} · ${parentModeLabel}`
+  }, [sourceData, t])
+
+  const getSegmentLengthValue = useCallback((): string | number => {
+    if (!sourceData?.mode)
+      return '-'
+
+    const maxTokens = isNumber(sourceData.rules?.segmentation?.max_tokens)
+      ? sourceData.rules.segmentation.max_tokens
+      : '-'
+
+    if (sourceData.mode === ProcessMode.general)
+      return maxTokens
+
+    const childMaxTokens = isNumber(sourceData.rules?.subchunk_segmentation?.max_tokens)
+      ? sourceData.rules.subchunk_segmentation.max_tokens
+      : '-'
+
+    return `${t('embedding.parentMaxTokens', { ns: 'datasetDocuments' })} ${maxTokens}; ${t('embedding.childMaxTokens', { ns: 'datasetDocuments' })} ${childMaxTokens}`
+  }, [sourceData, t])
+
+  const getTextCleaningValue = useCallback((): string => {
+    if (!sourceData?.mode)
+      return '-'
+
+    const enabledRules = sourceData.rules?.pre_processing_rules?.filter(rule => rule.enabled) || []
+    const ruleNames = enabledRules
+      .map((rule) => {
+        const name = getRuleName(rule.id)
+        return typeof name === 'string' ? name : ''
+      })
+      .filter(name => name)
+    return ruleNames.length > 0 ? ruleNames.join(',') : '-'
+  }, [sourceData, getRuleName])
+
+  const fieldValueGetters: Record<string, () => string | number> = {
+    mode: getModeValue,
+    segmentLength: getSegmentLengthValue,
+    textCleaning: getTextCleaningValue,
+  }
+
+  const isEconomical = indexingType === IndexingType.ECONOMICAL
+  const indexMethodIconSrc = isEconomical ? indexMethodIcon.economical : indexMethodIcon.high_quality
+  const indexModeLabel = t(`stepTwo.${isEconomical ? 'economical' : 'qualified'}`, { ns: 'datasetCreation' })
+
+  const effectiveRetrievalMethod = isEconomical ? 'keyword_search' : (retrievalMethod ?? 'semantic_search')
+  const retrievalLabel = t(`retrieval.${effectiveRetrievalMethod}.title`, { ns: 'dataset' })
+  const retrievalIconSrc = RETRIEVAL_ICON_MAP[retrievalMethod as keyof typeof RETRIEVAL_ICON_MAP] ?? retrievalIcon.vector
+
+  return (
+    <div className="flex flex-col gap-1">
+      {Object.keys(segmentationRuleLabels).map(field => (
+        <FieldInfo
+          key={field}
+          label={segmentationRuleLabels[field as keyof typeof segmentationRuleLabels]}
+          displayedValue={String(fieldValueGetters[field]())}
+        />
+      ))}
+      <FieldInfo
+        label={t('stepTwo.indexMode', { ns: 'datasetCreation' })}
+        displayedValue={indexModeLabel}
+        valueIcon={<Image className="size-4" src={indexMethodIconSrc} alt="" />}
+      />
+      <FieldInfo
+        label={t('form.retrievalSetting.title', { ns: 'datasetSettings' })}
+        displayedValue={retrievalLabel}
+        valueIcon={<Image className="size-4" src={retrievalIconSrc} alt="" />}
+      />
+    </div>
+  )
+}
+
+export default RuleDetail

web/app/components/datasets/create/embedding-process/upgrade-banner.tsx

@@ -0,0 +1,22 @@
+import type { FC } from 'react'
+import { useTranslation } from 'react-i18next'
+import { ZapFast } from '@/app/components/base/icons/src/vender/solid/general'
+import UpgradeBtn from '@/app/components/billing/upgrade-btn'
+
+const UpgradeBanner: FC = () => {
+  const { t } = useTranslation()
+
+  return (
+    <div className="flex h-14 items-center rounded-xl border-[0.5px] border-black/5 bg-white p-3 shadow-md">
+      <div className="flex h-8 w-8 shrink-0 items-center justify-center rounded-lg bg-[#FFF6ED]">
+        <ZapFast className="h-4 w-4 text-[#FB6514]" />
+      </div>
+      <div className="mx-3 grow text-[13px] font-medium text-gray-700">
+        {t('plansCommon.documentProcessingPriorityUpgrade', { ns: 'billing' })}
+      </div>
+      <UpgradeBtn loc="knowledge-speed-up" />
+    </div>
+  )
+}
+
+export default UpgradeBanner

web/app/components/datasets/create/embedding-process/use-indexing-status-polling.ts

@@ -0,0 +1,90 @@
+import type { IndexingStatusResponse } from '@/models/datasets'
+import { useEffect, useRef, useState } from 'react'
+import { fetchIndexingStatusBatch } from '@/service/datasets'
+
+const POLLING_INTERVAL = 2500
+const COMPLETED_STATUSES = ['completed', 'error', 'paused'] as const
+const EMBEDDING_STATUSES = ['indexing', 'splitting', 'parsing', 'cleaning', 'waiting'] as const
+
+type IndexingStatusPollingParams = {
+  datasetId: string
+  batchId: string
+}
+
+type IndexingStatusPollingResult = {
+  statusList: IndexingStatusResponse[]
+  isEmbedding: boolean
+  isEmbeddingCompleted: boolean
+}
+
+const isStatusCompleted = (status: string): boolean =>
+  COMPLETED_STATUSES.includes(status as typeof COMPLETED_STATUSES[number])
+
+const isAllCompleted = (statusList: IndexingStatusResponse[]): boolean =>
+  statusList.every(item => isStatusCompleted(item.indexing_status))
+
+/**
+ * Custom hook for polling indexing status with automatic stop on completion.
+ * Handles the polling lifecycle and provides derived states for UI rendering.
+ */
+export const useIndexingStatusPolling = ({
+  datasetId,
+  batchId,
+}: IndexingStatusPollingParams): IndexingStatusPollingResult => {
+  const [statusList, setStatusList] = useState<IndexingStatusResponse[]>([])
+  const isStopPollingRef = useRef(false)
+
+  useEffect(() => {
+    // Reset polling state on mount
+    isStopPollingRef.current = false
+    let timeoutId: ReturnType<typeof setTimeout> | null = null
+
+    const fetchStatus = async (): Promise<IndexingStatusResponse[]> => {
+      const response = await fetchIndexingStatusBatch({ datasetId, batchId })
+      setStatusList(response.data)
+      return response.data
+    }
+
+    const poll = async (): Promise<void> => {
+      if (isStopPollingRef.current)
+        return
+
+      try {
+        const data = await fetchStatus()
+        if (isAllCompleted(data)) {
+          isStopPollingRef.current = true
+          return
+        }
+      }
+      catch {
+        // Continue polling on error
+      }
+
+      if (!isStopPollingRef.current) {
+        timeoutId = setTimeout(() => {
+          poll()
+        }, POLLING_INTERVAL)
+      }
+    }
+
+    poll()
+
+    return () => {
+      isStopPollingRef.current = true
+      if (timeoutId)
+        clearTimeout(timeoutId)
+    }
+  }, [datasetId, batchId])
+
+  const isEmbedding = statusList.some(item =>
+    EMBEDDING_STATUSES.includes(item?.indexing_status as typeof EMBEDDING_STATUSES[number]),
+  )
+
+  const isEmbeddingCompleted = statusList.length > 0 && isAllCompleted(statusList)
+
+  return {
+    statusList,
+    isEmbedding,
+    isEmbeddingCompleted,
+  }
+}

web/app/components/datasets/create/embedding-process/utils.ts

@@ -0,0 +1,64 @@
+import type {
+  DataSourceInfo,
+  DataSourceType,
+  FullDocumentDetail,
+  IndexingStatusResponse,
+  LegacyDataSourceInfo,
+} from '@/models/datasets'
+
+const EMBEDDING_STATUSES = ['indexing', 'splitting', 'parsing', 'cleaning', 'waiting'] as const
+
+/**
+ * Type guard for legacy data source info with upload_file property
+ */
+export const isLegacyDataSourceInfo = (info: DataSourceInfo): info is LegacyDataSourceInfo => {
+  return info != null && typeof (info as LegacyDataSourceInfo).upload_file === 'object'
+}
+
+/**
+ * Check if a status indicates the source is being embedded
+ */
+export const isSourceEmbedding = (detail: IndexingStatusResponse): boolean =>
+  EMBEDDING_STATUSES.includes(detail.indexing_status as typeof EMBEDDING_STATUSES[number])
+
+/**
+ * Calculate the progress percentage for a document
+ */
+export const getSourcePercent = (detail: IndexingStatusResponse): number => {
+  const completedCount = detail.completed_segments || 0
+  const totalCount = detail.total_segments || 0
+
+  if (totalCount === 0)
+    return 0
+
+  const percent = Math.round(completedCount * 100 / totalCount)
+  return Math.min(percent, 100)
+}
+
+/**
+ * Get file extension from filename, defaults to 'txt'
+ */
+export const getFileType = (name?: string): string =>
+  name?.split('.').pop() || 'txt'
+
+/**
+ * Document lookup utilities - provides document info by ID from a list
+ */
+export const createDocumentLookup = (documents: FullDocumentDetail[]) => {
+  const documentMap = new Map(documents.map(doc => [doc.id, doc]))
+
+  return {
+    getDocument: (id: string) => documentMap.get(id),
+
+    getName: (id: string) => documentMap.get(id)?.name,
+
+    getSourceType: (id: string) => documentMap.get(id)?.data_source_type as DataSourceType | undefined,
+
+    getNotionIcon: (id: string) => {
+      const info = documentMap.get(id)?.data_source_info
+      if (info && isLegacyDataSourceInfo(info))
+        return info.notion_page_icon
+      return undefined
+    },
+  }
+}

web/app/components/datasets/create/step-two/components/general-chunking-options.tsx

@@ -0,0 +1,199 @@
+'use client'
+
+import type { FC } from 'react'
+import type { PreProcessingRule } from '@/models/datasets'
+import {
+  RiAlertFill,
+  RiSearchEyeLine,
+} from '@remixicon/react'
+import Image from 'next/image'
+import { useTranslation } from 'react-i18next'
+import Button from '@/app/components/base/button'
+import Checkbox from '@/app/components/base/checkbox'
+import Divider from '@/app/components/base/divider'
+import Tooltip from '@/app/components/base/tooltip'
+import { IS_CE_EDITION } from '@/config'
+import { ChunkingMode } from '@/models/datasets'
+import SettingCog from '../../assets/setting-gear-mod.svg'
+import s from '../index.module.css'
+import LanguageSelect from '../language-select'
+import { DelimiterInput, MaxLengthInput, OverlapInput } from './inputs'
+import { OptionCard } from './option-card'
+
+type TextLabelProps = {
+  children: React.ReactNode
+}
+
+const TextLabel: FC<TextLabelProps> = ({ children }) => {
+  return <label className="system-sm-semibold text-text-secondary">{children}</label>
+}
+
+type GeneralChunkingOptionsProps = {
+  // State
+  segmentIdentifier: string
+  maxChunkLength: number
+  overlap: number
+  rules: PreProcessingRule[]
+  currentDocForm: ChunkingMode
+  docLanguage: string
+  // Flags
+  isActive: boolean
+  isInUpload: boolean
+  isNotUploadInEmptyDataset: boolean
+  hasCurrentDatasetDocForm: boolean
+  // Actions
+  onSegmentIdentifierChange: (value: string) => void
+  onMaxChunkLengthChange: (value: number) => void
+  onOverlapChange: (value: number) => void
+  onRuleToggle: (id: string) => void
+  onDocFormChange: (form: ChunkingMode) => void
+  onDocLanguageChange: (lang: string) => void
+  onPreview: () => void
+  onReset: () => void
+  // Locale
+  locale: string
+}
+
+export const GeneralChunkingOptions: FC<GeneralChunkingOptionsProps> = ({
+  segmentIdentifier,
+  maxChunkLength,
+  overlap,
+  rules,
+  currentDocForm,
+  docLanguage,
+  isActive,
+  isInUpload,
+  isNotUploadInEmptyDataset,
+  hasCurrentDatasetDocForm,
+  onSegmentIdentifierChange,
+  onMaxChunkLengthChange,
+  onOverlapChange,
+  onRuleToggle,
+  onDocFormChange,
+  onDocLanguageChange,
+  onPreview,
+  onReset,
+  locale,
+}) => {
+  const { t } = useTranslation()
+
+  const getRuleName = (key: string): string => {
+    const ruleNameMap: Record<string, string> = {
+      remove_extra_spaces: t('stepTwo.removeExtraSpaces', { ns: 'datasetCreation' }),
+      remove_urls_emails: t('stepTwo.removeUrlEmails', { ns: 'datasetCreation' }),
+      remove_stopwords: t('stepTwo.removeStopwords', { ns: 'datasetCreation' }),
+    }
+    return ruleNameMap[key] ?? key
+  }
+
+  return (
+    <OptionCard
+      className="mb-2 bg-background-section"
+      title={t('stepTwo.general', { ns: 'datasetCreation' })}
+      icon={<Image width={20} height={20} src={SettingCog} alt={t('stepTwo.general', { ns: 'datasetCreation' })} />}
+      activeHeaderClassName="bg-dataset-option-card-blue-gradient"
+      description={t('stepTwo.generalTip', { ns: 'datasetCreation' })}
+      isActive={isActive}
+      onSwitched={() => onDocFormChange(ChunkingMode.text)}
+      actions={(
+        <>
+          <Button variant="secondary-accent" onClick={onPreview}>
+            <RiSearchEyeLine className="mr-0.5 h-4 w-4" />
+            {t('stepTwo.previewChunk', { ns: 'datasetCreation' })}
+          </Button>
+          <Button variant="ghost" onClick={onReset}>
+            {t('stepTwo.reset', { ns: 'datasetCreation' })}
+          </Button>
+        </>
+      )}
+      noHighlight={isInUpload && isNotUploadInEmptyDataset}
+    >
+      <div className="flex flex-col gap-y-4">
+        <div className="flex gap-3">
+          <DelimiterInput
+            value={segmentIdentifier}
+            onChange={e => onSegmentIdentifierChange(e.target.value)}
+          />
+          <MaxLengthInput
+            unit="characters"
+            value={maxChunkLength}
+            onChange={onMaxChunkLengthChange}
+          />
+          <OverlapInput
+            unit="characters"
+            value={overlap}
+            min={1}
+            onChange={onOverlapChange}
+          />
+        </div>
+        <div className="flex w-full flex-col">
+          <div className="flex items-center gap-x-2">
+            <div className="inline-flex shrink-0">
+              <TextLabel>{t('stepTwo.rules', { ns: 'datasetCreation' })}</TextLabel>
+            </div>
+            <Divider className="grow" bgStyle="gradient" />
+          </div>
+          <div className="mt-1">
+            {rules.map(rule => (
+              <div
+                key={rule.id}
+                className={s.ruleItem}
+                onClick={() => onRuleToggle(rule.id)}
+              >
+                <Checkbox checked={rule.enabled} />
+                <label className="system-sm-regular ml-2 cursor-pointer text-text-secondary">
+                  {getRuleName(rule.id)}
+                </label>
+              </div>
+            ))}
+            {IS_CE_EDITION && (
+              <>
+                <Divider type="horizontal" className="my-4 bg-divider-subtle" />
+                <div className="flex items-center py-0.5">
+                  <div
+                    className="flex items-center"
+                    onClick={() => {
+                      if (hasCurrentDatasetDocForm)
+                        return
+                      if (currentDocForm === ChunkingMode.qa)
+                        onDocFormChange(ChunkingMode.text)
+                      else
+                        onDocFormChange(ChunkingMode.qa)
+                    }}
+                  >
+                    <Checkbox
+                      checked={currentDocForm === ChunkingMode.qa}
+                      disabled={hasCurrentDatasetDocForm}
+                    />
+                    <label className="system-sm-regular ml-2 cursor-pointer text-text-secondary">
+                      {t('stepTwo.useQALanguage', { ns: 'datasetCreation' })}
+                    </label>
+                  </div>
+                  <LanguageSelect
+                    currentLanguage={docLanguage || locale}
+                    onSelect={onDocLanguageChange}
+                    disabled={currentDocForm !== ChunkingMode.qa}
+                  />
+                  <Tooltip popupContent={t('stepTwo.QATip', { ns: 'datasetCreation' })} />
+                </div>
+                {currentDocForm === ChunkingMode.qa && (
+                  <div
+                    style={{
+                      background: 'linear-gradient(92deg, rgba(247, 144, 9, 0.1) 0%, rgba(255, 255, 255, 0.00) 100%)',
+                    }}
+                    className="mt-2 flex h-10 items-center gap-2 rounded-xl border border-components-panel-border px-3 text-xs shadow-xs backdrop-blur-[5px]"
+                  >
+                    <RiAlertFill className="size-4 text-text-warning-secondary" />
+                    <span className="system-xs-medium text-text-primary">
+                      {t('stepTwo.QATip', { ns: 'datasetCreation' })}
+                    </span>
+                  </div>
+                )}
+              </>
+            )}
+          </div>
+        </div>
+      </div>
+    </OptionCard>
+  )
+}

web/app/components/datasets/create/step-two/components/index.ts

@@ -0,0 +1,5 @@
+export { GeneralChunkingOptions } from './general-chunking-options'
+export { IndexingModeSection } from './indexing-mode-section'
+export { ParentChildOptions } from './parent-child-options'
+export { PreviewPanel } from './preview-panel'
+export { StepTwoFooter } from './step-two-footer'

web/app/components/datasets/create/step-two/components/indexing-mode-section.tsx

@@ -0,0 +1,253 @@
+'use client'
+
+import type { FC } from 'react'
+import type { DefaultModel, Model } from '@/app/components/header/account-setting/model-provider-page/declarations'
+import type { RetrievalConfig } from '@/types/app'
+import Image from 'next/image'
+import Link from 'next/link'
+import { useTranslation } from 'react-i18next'
+import Badge from '@/app/components/base/badge'
+import Button from '@/app/components/base/button'
+import CustomDialog from '@/app/components/base/dialog'
+import Divider from '@/app/components/base/divider'
+import { AlertTriangle } from '@/app/components/base/icons/src/vender/solid/alertsAndFeedback'
+import Tooltip from '@/app/components/base/tooltip'
+import EconomicalRetrievalMethodConfig from '@/app/components/datasets/common/economical-retrieval-method-config'
+import RetrievalMethodConfig from '@/app/components/datasets/common/retrieval-method-config'
+import ModelSelector from '@/app/components/header/account-setting/model-provider-page/model-selector'
+import { useDocLink } from '@/context/i18n'
+import { ChunkingMode } from '@/models/datasets'
+import { cn } from '@/utils/classnames'
+import { indexMethodIcon } from '../../icons'
+import { IndexingType } from '../hooks'
+import s from '../index.module.css'
+import { OptionCard } from './option-card'
+
+type IndexingModeSectionProps = {
+  // State
+  indexType: IndexingType
+  hasSetIndexType: boolean
+  docForm: ChunkingMode
+  embeddingModel: DefaultModel
+  embeddingModelList?: Model[]
+  retrievalConfig: RetrievalConfig
+  showMultiModalTip: boolean
+  // Flags
+  isModelAndRetrievalConfigDisabled: boolean
+  datasetId?: string
+  // Modal state
+  isQAConfirmDialogOpen: boolean
+  // Actions
+  onIndexTypeChange: (type: IndexingType) => void
+  onEmbeddingModelChange: (model: DefaultModel) => void
+  onRetrievalConfigChange: (config: RetrievalConfig) => void
+  onQAConfirmDialogClose: () => void
+  onQAConfirmDialogConfirm: () => void
+}
+
+export const IndexingModeSection: FC<IndexingModeSectionProps> = ({
+  indexType,
+  hasSetIndexType,
+  docForm,
+  embeddingModel,
+  embeddingModelList,
+  retrievalConfig,
+  showMultiModalTip,
+  isModelAndRetrievalConfigDisabled,
+  datasetId,
+  isQAConfirmDialogOpen,
+  onIndexTypeChange,
+  onEmbeddingModelChange,
+  onRetrievalConfigChange,
+  onQAConfirmDialogClose,
+  onQAConfirmDialogConfirm,
+}) => {
+  const { t } = useTranslation()
+  const docLink = useDocLink()
+
+  const getIndexingTechnique = () => indexType
+
+  return (
+    <>
+      {/* Index Mode */}
+      <div className="system-md-semibold mb-1 text-text-secondary">
+        {t('stepTwo.indexMode', { ns: 'datasetCreation' })}
+      </div>
+      <div className="flex items-center gap-2">
+        {/* Qualified option */}
+        {(!hasSetIndexType || (hasSetIndexType && indexType === IndexingType.QUALIFIED)) && (
+          <OptionCard
+            className="flex-1 self-stretch"
+            title={(
+              <div className="flex items-center">
+                {t('stepTwo.qualified', { ns: 'datasetCreation' })}
+                <Badge
+                  className={cn(
+                    'ml-1 h-[18px]',
+                    (!hasSetIndexType && indexType === IndexingType.QUALIFIED)
+                      ? 'border-text-accent-secondary text-text-accent-secondary'
+                      : '',
+                  )}
+                  uppercase
+                >
+                  {t('stepTwo.recommend', { ns: 'datasetCreation' })}
+                </Badge>
+                <span className="ml-auto">
+                  {!hasSetIndexType && <span className={cn(s.radio)} />}
+                </span>
+              </div>
+            )}
+            description={t('stepTwo.qualifiedTip', { ns: 'datasetCreation' })}
+            icon={<Image src={indexMethodIcon.high_quality} alt="" />}
+            isActive={!hasSetIndexType && indexType === IndexingType.QUALIFIED}
+            disabled={hasSetIndexType}
+            onSwitched={() => onIndexTypeChange(IndexingType.QUALIFIED)}
+          />
+        )}
+
+        {/* Economical option */}
+        {(!hasSetIndexType || (hasSetIndexType && indexType === IndexingType.ECONOMICAL)) && (
+          <>
+            <CustomDialog show={isQAConfirmDialogOpen} onClose={onQAConfirmDialogClose} className="w-[432px]">
+              <header className="mb-4 pt-6">
+                <h2 className="text-lg font-semibold text-text-primary">
+                  {t('stepTwo.qaSwitchHighQualityTipTitle', { ns: 'datasetCreation' })}
+                </h2>
+                <p className="mt-2 text-sm font-normal text-text-secondary">
+                  {t('stepTwo.qaSwitchHighQualityTipContent', { ns: 'datasetCreation' })}
+                </p>
+              </header>
+              <div className="flex gap-2 pb-6">
+                <Button className="ml-auto" onClick={onQAConfirmDialogClose}>
+                  {t('stepTwo.cancel', { ns: 'datasetCreation' })}
+                </Button>
+                <Button variant="primary" onClick={onQAConfirmDialogConfirm}>
+                  {t('stepTwo.switch', { ns: 'datasetCreation' })}
+                </Button>
+              </div>
+            </CustomDialog>
+            <Tooltip
+              popupContent={(
+                <div className="rounded-lg border-components-panel-border bg-components-tooltip-bg p-3 text-xs font-medium text-text-secondary shadow-lg">
+                  {docForm === ChunkingMode.qa
+                    ? t('stepTwo.notAvailableForQA', { ns: 'datasetCreation' })
+                    : t('stepTwo.notAvailableForParentChild', { ns: 'datasetCreation' })}
+                </div>
+              )}
+              noDecoration
+              position="top"
+              asChild={false}
+              triggerClassName="flex-1 self-stretch"
+            >
+              <OptionCard
+                className="h-full"
+                title={t('stepTwo.economical', { ns: 'datasetCreation' })}
+                description={t('stepTwo.economicalTip', { ns: 'datasetCreation' })}
+                icon={<Image src={indexMethodIcon.economical} alt="" />}
+                isActive={!hasSetIndexType && indexType === IndexingType.ECONOMICAL}
+                disabled={hasSetIndexType || docForm !== ChunkingMode.text}
+                onSwitched={() => onIndexTypeChange(IndexingType.ECONOMICAL)}
+              />
+            </Tooltip>
+          </>
+        )}
+      </div>
+
+      {/* High quality tip */}
+      {!hasSetIndexType && indexType === IndexingType.QUALIFIED && (
+        <div className="mt-2 flex h-10 items-center gap-x-0.5 overflow-hidden rounded-xl border-[0.5px] border-components-panel-border bg-components-panel-bg-blur p-2 shadow-xs backdrop-blur-[5px]">
+          <div className="absolute bottom-0 left-0 right-0 top-0 bg-dataset-warning-message-bg opacity-40"></div>
+          <div className="p-1">
+            <AlertTriangle className="size-4 text-text-warning-secondary" />
+          </div>
+          <span className="system-xs-medium text-text-primary">
+            {t('stepTwo.highQualityTip', { ns: 'datasetCreation' })}
+          </span>
+        </div>
+      )}
+
+      {/* Economical index setting tip */}
+      {hasSetIndexType && indexType === IndexingType.ECONOMICAL && (
+        <div className="system-xs-medium mt-2 text-text-tertiary">
+          {t('stepTwo.indexSettingTip', { ns: 'datasetCreation' })}
+          <Link className="text-text-accent" href={`/datasets/${datasetId}/settings`}>
+            {t('stepTwo.datasetSettingLink', { ns: 'datasetCreation' })}
+          </Link>
+        </div>
+      )}
+
+      {/* Embedding model */}
+      {indexType === IndexingType.QUALIFIED && (
+        <div className="mt-5">
+          <div className={cn('system-md-semibold mb-1 text-text-secondary', datasetId && 'flex items-center justify-between')}>
+            {t('form.embeddingModel', { ns: 'datasetSettings' })}
+          </div>
+          <ModelSelector
+            readonly={isModelAndRetrievalConfigDisabled}
+            triggerClassName={isModelAndRetrievalConfigDisabled ? 'opacity-50' : ''}
+            defaultModel={embeddingModel}
+            modelList={embeddingModelList ?? []}
+            onSelect={onEmbeddingModelChange}
+          />
+          {isModelAndRetrievalConfigDisabled && (
+            <div className="system-xs-medium mt-2 text-text-tertiary">
+              {t('stepTwo.indexSettingTip', { ns: 'datasetCreation' })}
+              <Link className="text-text-accent" href={`/datasets/${datasetId}/settings`}>
+                {t('stepTwo.datasetSettingLink', { ns: 'datasetCreation' })}
+              </Link>
+            </div>
+          )}
+        </div>
+      )}
+
+      <Divider className="my-5" />
+
+      {/* Retrieval Method Config */}
+      <div>
+        {!isModelAndRetrievalConfigDisabled
+          ? (
+              <div className="mb-1">
+                <div className="system-md-semibold mb-0.5 text-text-secondary">
+                  {t('form.retrievalSetting.title', { ns: 'datasetSettings' })}
+                </div>
+                <div className="body-xs-regular text-text-tertiary">
+                  <a
+                    target="_blank"
+                    rel="noopener noreferrer"
+                    href={docLink('/guides/knowledge-base/create-knowledge-and-upload-documents')}
+                    className="text-text-accent"
+                  >
+                    {t('form.retrievalSetting.learnMore', { ns: 'datasetSettings' })}
+                  </a>
+                  {t('form.retrievalSetting.longDescription', { ns: 'datasetSettings' })}
+                </div>
+              </div>
+            )
+          : (
+              <div className={cn('system-md-semibold mb-0.5 text-text-secondary', 'flex items-center justify-between')}>
+                <div>{t('form.retrievalSetting.title', { ns: 'datasetSettings' })}</div>
+              </div>
+            )}
+
+        <div>
+          {getIndexingTechnique() === IndexingType.QUALIFIED
+            ? (
+                <RetrievalMethodConfig
+                  disabled={isModelAndRetrievalConfigDisabled}
+                  value={retrievalConfig}
+                  onChange={onRetrievalConfigChange}
+                  showMultiModalTip={showMultiModalTip}
+                />
+              )
+            : (
+                <EconomicalRetrievalMethodConfig
+                  disabled={isModelAndRetrievalConfigDisabled}
+                  value={retrievalConfig}
+                  onChange={onRetrievalConfigChange}
+                />
+              )}
+        </div>
+      </div>
+    </>
+  )
+}

web/app/components/datasets/create/step-two/components/parent-child-options.tsx

@@ -0,0 +1,191 @@
+'use client'
+
+import type { FC } from 'react'
+import type { ParentChildConfig } from '../hooks'
+import type { ParentMode, PreProcessingRule } from '@/models/datasets'
+import { RiSearchEyeLine } from '@remixicon/react'
+import Image from 'next/image'
+import { useTranslation } from 'react-i18next'
+import Button from '@/app/components/base/button'
+import Checkbox from '@/app/components/base/checkbox'
+import Divider from '@/app/components/base/divider'
+import { ParentChildChunk } from '@/app/components/base/icons/src/vender/knowledge'
+import RadioCard from '@/app/components/base/radio-card'
+import { ChunkingMode } from '@/models/datasets'
+import FileList from '../../assets/file-list-3-fill.svg'
+import Note from '../../assets/note-mod.svg'
+import BlueEffect from '../../assets/option-card-effect-blue.svg'
+import s from '../index.module.css'
+import { DelimiterInput, MaxLengthInput } from './inputs'
+import { OptionCard } from './option-card'
+
+type TextLabelProps = {
+  children: React.ReactNode
+}
+
+const TextLabel: FC<TextLabelProps> = ({ children }) => {
+  return <label className="system-sm-semibold text-text-secondary">{children}</label>
+}
+
+type ParentChildOptionsProps = {
+  // State
+  parentChildConfig: ParentChildConfig
+  rules: PreProcessingRule[]
+  currentDocForm: ChunkingMode
+  // Flags
+  isActive: boolean
+  isInUpload: boolean
+  isNotUploadInEmptyDataset: boolean
+  // Actions
+  onDocFormChange: (form: ChunkingMode) => void
+  onChunkForContextChange: (mode: ParentMode) => void
+  onParentDelimiterChange: (value: string) => void
+  onParentMaxLengthChange: (value: number) => void
+  onChildDelimiterChange: (value: string) => void
+  onChildMaxLengthChange: (value: number) => void
+  onRuleToggle: (id: string) => void
+  onPreview: () => void
+  onReset: () => void
+}
+
+export const ParentChildOptions: FC<ParentChildOptionsProps> = ({
+  parentChildConfig,
+  rules,
+  currentDocForm: _currentDocForm,
+  isActive,
+  isInUpload,
+  isNotUploadInEmptyDataset,
+  onDocFormChange,
+  onChunkForContextChange,
+  onParentDelimiterChange,
+  onParentMaxLengthChange,
+  onChildDelimiterChange,
+  onChildMaxLengthChange,
+  onRuleToggle,
+  onPreview,
+  onReset,
+}) => {
+  const { t } = useTranslation()
+
+  const getRuleName = (key: string): string => {
+    const ruleNameMap: Record<string, string> = {
+      remove_extra_spaces: t('stepTwo.removeExtraSpaces', { ns: 'datasetCreation' }),
+      remove_urls_emails: t('stepTwo.removeUrlEmails', { ns: 'datasetCreation' }),
+      remove_stopwords: t('stepTwo.removeStopwords', { ns: 'datasetCreation' }),
+    }
+    return ruleNameMap[key] ?? key
+  }
+
+  return (
+    <OptionCard
+      title={t('stepTwo.parentChild', { ns: 'datasetCreation' })}
+      icon={<ParentChildChunk className="h-[20px] w-[20px]" />}
+      effectImg={BlueEffect.src}
+      className="text-util-colors-blue-light-blue-light-500"
+      activeHeaderClassName="bg-dataset-option-card-blue-gradient"
+      description={t('stepTwo.parentChildTip', { ns: 'datasetCreation' })}
+      isActive={isActive}
+      onSwitched={() => onDocFormChange(ChunkingMode.parentChild)}
+      actions={(
+        <>
+          <Button variant="secondary-accent" onClick={onPreview}>
+            <RiSearchEyeLine className="mr-0.5 h-4 w-4" />
+            {t('stepTwo.previewChunk', { ns: 'datasetCreation' })}
+          </Button>
+          <Button variant="ghost" onClick={onReset}>
+            {t('stepTwo.reset', { ns: 'datasetCreation' })}
+          </Button>
+        </>
+      )}
+      noHighlight={isInUpload && isNotUploadInEmptyDataset}
+    >
+      <div className="flex flex-col gap-4">
+        {/* Parent chunk for context */}
+        <div>
+          <div className="flex items-center gap-x-2">
+            <div className="inline-flex shrink-0">
+              <TextLabel>{t('stepTwo.parentChunkForContext', { ns: 'datasetCreation' })}</TextLabel>
+            </div>
+            <Divider className="grow" bgStyle="gradient" />
+          </div>
+          <RadioCard
+            className="mt-1"
+            icon={<Image src={Note} alt="" />}
+            title={t('stepTwo.paragraph', { ns: 'datasetCreation' })}
+            description={t('stepTwo.paragraphTip', { ns: 'datasetCreation' })}
+            isChosen={parentChildConfig.chunkForContext === 'paragraph'}
+            onChosen={() => onChunkForContextChange('paragraph')}
+            chosenConfig={(
+              <div className="flex gap-3">
+                <DelimiterInput
+                  value={parentChildConfig.parent.delimiter}
+                  tooltip={t('stepTwo.parentChildDelimiterTip', { ns: 'datasetCreation' })!}
+                  onChange={e => onParentDelimiterChange(e.target.value)}
+                />
+                <MaxLengthInput
+                  unit="characters"
+                  value={parentChildConfig.parent.maxLength}
+                  onChange={onParentMaxLengthChange}
+                />
+              </div>
+            )}
+          />
+          <RadioCard
+            className="mt-2"
+            icon={<Image src={FileList} alt="" />}
+            title={t('stepTwo.fullDoc', { ns: 'datasetCreation' })}
+            description={t('stepTwo.fullDocTip', { ns: 'datasetCreation' })}
+            onChosen={() => onChunkForContextChange('full-doc')}
+            isChosen={parentChildConfig.chunkForContext === 'full-doc'}
+          />
+        </div>
+
+        {/* Child chunk for retrieval */}
+        <div>
+          <div className="flex items-center gap-x-2">
+            <div className="inline-flex shrink-0">
+              <TextLabel>{t('stepTwo.childChunkForRetrieval', { ns: 'datasetCreation' })}</TextLabel>
+            </div>
+            <Divider className="grow" bgStyle="gradient" />
+          </div>
+          <div className="mt-1 flex gap-3">
+            <DelimiterInput
+              value={parentChildConfig.child.delimiter}
+              tooltip={t('stepTwo.parentChildChunkDelimiterTip', { ns: 'datasetCreation' })!}
+              onChange={e => onChildDelimiterChange(e.target.value)}
+            />
+            <MaxLengthInput
+              unit="characters"
+              value={parentChildConfig.child.maxLength}
+              onChange={onChildMaxLengthChange}
+            />
+          </div>
+        </div>
+
+        {/* Rules */}
+        <div>
+          <div className="flex items-center gap-x-2">
+            <div className="inline-flex shrink-0">
+              <TextLabel>{t('stepTwo.rules', { ns: 'datasetCreation' })}</TextLabel>
+            </div>
+            <Divider className="grow" bgStyle="gradient" />
+          </div>
+          <div className="mt-1">
+            {rules.map(rule => (
+              <div
+                key={rule.id}
+                className={s.ruleItem}
+                onClick={() => onRuleToggle(rule.id)}
+              >
+                <Checkbox checked={rule.enabled} />
+                <label className="system-sm-regular ml-2 cursor-pointer text-text-secondary">
+                  {getRuleName(rule.id)}
+                </label>
+              </div>
+            ))}
+          </div>
+        </div>
+      </div>
+    </OptionCard>
+  )
+}

web/app/components/datasets/create/step-two/components/preview-panel.tsx

@@ -0,0 +1,171 @@
+'use client'
+
+import type { FC } from 'react'
+import type { ParentChildConfig } from '../hooks'
+import type { DataSourceType, FileIndexingEstimateResponse } from '@/models/datasets'
+import { RiSearchEyeLine } from '@remixicon/react'
+import { noop } from 'es-toolkit/function'
+import { useTranslation } from 'react-i18next'
+import Badge from '@/app/components/base/badge'
+import FloatRightContainer from '@/app/components/base/float-right-container'
+import { SkeletonContainer, SkeletonPoint, SkeletonRectangle, SkeletonRow } from '@/app/components/base/skeleton'
+import { FULL_DOC_PREVIEW_LENGTH } from '@/config'
+import { ChunkingMode } from '@/models/datasets'
+import { cn } from '@/utils/classnames'
+import { ChunkContainer, QAPreview } from '../../../chunk'
+import PreviewDocumentPicker from '../../../common/document-picker/preview-document-picker'
+import { PreviewSlice } from '../../../formatted-text/flavours/preview-slice'
+import { FormattedText } from '../../../formatted-text/formatted'
+import PreviewContainer from '../../../preview/container'
+import { PreviewHeader } from '../../../preview/header'
+
+type PreviewPanelProps = {
+  // State
+  isMobile: boolean
+  dataSourceType: DataSourceType
+  currentDocForm: ChunkingMode
+  estimate?: FileIndexingEstimateResponse
+  parentChildConfig: ParentChildConfig
+  isSetting?: boolean
+  // Picker
+  pickerFiles: Array<{ id: string, name: string, extension: string }>
+  pickerValue: { id: string, name: string, extension: string }
+  // Mutation state
+  isIdle: boolean
+  isPending: boolean
+  // Actions
+  onPickerChange: (selected: { id: string, name: string }) => void
+}
+
+export const PreviewPanel: FC<PreviewPanelProps> = ({
+  isMobile,
+  dataSourceType: _dataSourceType,
+  currentDocForm,
+  estimate,
+  parentChildConfig,
+  isSetting,
+  pickerFiles,
+  pickerValue,
+  isIdle,
+  isPending,
+  onPickerChange,
+}) => {
+  const { t } = useTranslation()
+
+  return (
+    <FloatRightContainer isMobile={isMobile} isOpen={true} onClose={noop} footer={null}>
+      <PreviewContainer
+        header={(
+          <PreviewHeader title={t('stepTwo.preview', { ns: 'datasetCreation' })}>
+            <div className="flex items-center gap-1">
+              <PreviewDocumentPicker
+                files={pickerFiles as Array<Required<{ id: string, name: string, extension: string }>>}
+                onChange={onPickerChange}
+                value={isSetting ? pickerFiles[0] : pickerValue}
+              />
+              {currentDocForm !== ChunkingMode.qa && (
+                <Badge
+                  text={t('stepTwo.previewChunkCount', {
+                    ns: 'datasetCreation',
+                    count: estimate?.total_segments || 0,
+                  }) as string}
+                />
+              )}
+            </div>
+          </PreviewHeader>
+        )}
+        className={cn('relative flex h-full w-1/2 shrink-0 p-4 pr-0', isMobile && 'w-full max-w-[524px]')}
+        mainClassName="space-y-6"
+      >
+        {/* QA Preview */}
+        {currentDocForm === ChunkingMode.qa && estimate?.qa_preview && (
+          estimate.qa_preview.map((item, index) => (
+            <ChunkContainer
+              key={item.question}
+              label={`Chunk-${index + 1}`}
+              characterCount={item.question.length + item.answer.length}
+            >
+              <QAPreview qa={item} />
+            </ChunkContainer>
+          ))
+        )}
+
+        {/* Text Preview */}
+        {currentDocForm === ChunkingMode.text && estimate?.preview && (
+          estimate.preview.map((item, index) => (
+            <ChunkContainer
+              key={item.content}
+              label={`Chunk-${index + 1}`}
+              characterCount={item.content.length}
+            >
+              {item.content}
+            </ChunkContainer>
+          ))
+        )}
+
+        {/* Parent-Child Preview */}
+        {currentDocForm === ChunkingMode.parentChild && estimate?.preview && (
+          estimate.preview.map((item, index) => {
+            const indexForLabel = index + 1
+            const childChunks = parentChildConfig.chunkForContext === 'full-doc'
+              ? item.child_chunks.slice(0, FULL_DOC_PREVIEW_LENGTH)
+              : item.child_chunks
+            return (
+              <ChunkContainer
+                key={item.content}
+                label={`Chunk-${indexForLabel}`}
+                characterCount={item.content.length}
+              >
+                <FormattedText>
+                  {childChunks.map((child, childIndex) => {
+                    const childIndexForLabel = childIndex + 1
+                    return (
+                      <PreviewSlice
+                        key={`C-${childIndexForLabel}-${child}`}
+                        label={`C-${childIndexForLabel}`}
+                        text={child}
+                        tooltip={`Child-chunk-${childIndexForLabel} · ${child.length} Characters`}
+                        labelInnerClassName="text-[10px] font-semibold align-bottom leading-7"
+                        dividerClassName="leading-7"
+                      />
+                    )
+                  })}
+                </FormattedText>
+              </ChunkContainer>
+            )
+          })
+        )}
+
+        {/* Idle State */}
+        {isIdle && (
+          <div className="flex h-full w-full items-center justify-center">
+            <div className="flex flex-col items-center justify-center gap-3">
+              <RiSearchEyeLine className="size-10 text-text-empty-state-icon" />
+              <p className="text-sm text-text-tertiary">
+                {t('stepTwo.previewChunkTip', { ns: 'datasetCreation' })}
+              </p>
+            </div>
+          </div>
+        )}
+
+        {/* Loading State */}
+        {isPending && (
+          <div className="space-y-6">
+            {Array.from({ length: 10 }, (_, i) => (
+              <SkeletonContainer key={i}>
+                <SkeletonRow>
+                  <SkeletonRectangle className="w-20" />
+                  <SkeletonPoint />
+                  <SkeletonRectangle className="w-24" />
+                </SkeletonRow>
+                <SkeletonRectangle className="w-full" />
+                <SkeletonRectangle className="w-full" />
+                <SkeletonRectangle className="w-[422px]" />
+              </SkeletonContainer>
+            ))}
+          </div>
+        )}
+      </PreviewContainer>
+    </FloatRightContainer>
+  )
+}

web/app/components/datasets/create/step-two/components/step-two-footer.tsx

@@ -0,0 +1,58 @@
+'use client'
+
+import type { FC } from 'react'
+import { RiArrowLeftLine } from '@remixicon/react'
+import { useTranslation } from 'react-i18next'
+import Button from '@/app/components/base/button'
+
+type StepTwoFooterProps = {
+  isSetting?: boolean
+  isCreating: boolean
+  onPrevious: () => void
+  onCreate: () => void
+  onCancel?: () => void
+}
+
+export const StepTwoFooter: FC<StepTwoFooterProps> = ({
+  isSetting,
+  isCreating,
+  onPrevious,
+  onCreate,
+  onCancel,
+}) => {
+  const { t } = useTranslation()
+
+  if (!isSetting) {
+    return (
+      <div className="mt-8 flex items-center py-2">
+        <Button onClick={onPrevious}>
+          <RiArrowLeftLine className="mr-1 h-4 w-4" />
+          {t('stepTwo.previousStep', { ns: 'datasetCreation' })}
+        </Button>
+        <Button
+          className="ml-auto"
+          loading={isCreating}
+          variant="primary"
+          onClick={onCreate}
+        >
+          {t('stepTwo.nextStep', { ns: 'datasetCreation' })}
+        </Button>
+      </div>
+    )
+  }
+
+  return (
+    <div className="mt-8 flex items-center py-2">
+      <Button
+        loading={isCreating}
+        variant="primary"
+        onClick={onCreate}
+      >
+        {t('stepTwo.save', { ns: 'datasetCreation' })}
+      </Button>
+      <Button className="ml-2" onClick={onCancel}>
+        {t('stepTwo.cancel', { ns: 'datasetCreation' })}
+      </Button>
+    </div>
+  )
+}

web/app/components/datasets/create/step-two/hooks/index.ts

@@ -0,0 +1,14 @@
+export { useDocumentCreation } from './use-document-creation'
+export type { DocumentCreation, ValidationParams } from './use-document-creation'
+
+export { IndexingType, useIndexingConfig } from './use-indexing-config'
+export type { IndexingConfig } from './use-indexing-config'
+
+export { useIndexingEstimate } from './use-indexing-estimate'
+export type { IndexingEstimate } from './use-indexing-estimate'
+
+export { usePreviewState } from './use-preview-state'
+export type { PreviewState } from './use-preview-state'
+
+export { DEFAULT_MAXIMUM_CHUNK_LENGTH, DEFAULT_OVERLAP, DEFAULT_SEGMENT_IDENTIFIER, defaultParentChildConfig, MAXIMUM_CHUNK_TOKEN_LENGTH, useSegmentationState } from './use-segmentation-state'
+export type { ParentChildConfig, SegmentationState } from './use-segmentation-state'

web/app/components/datasets/create/step-two/hooks/use-document-creation.ts

@@ -0,0 +1,279 @@
+import type { DefaultModel, Model } from '@/app/components/header/account-setting/model-provider-page/declarations'
+import type { NotionPage } from '@/models/common'
+import type {
+  ChunkingMode,
+  CrawlOptions,
+  CrawlResultItem,
+  CreateDocumentReq,
+  createDocumentResponse,
+  CustomFile,
+  FullDocumentDetail,
+  ProcessRule,
+} from '@/models/datasets'
+import type { RetrievalConfig, RETRIEVE_METHOD } from '@/types/app'
+import { useCallback } from 'react'
+import { useTranslation } from 'react-i18next'
+import { trackEvent } from '@/app/components/base/amplitude'
+import Toast from '@/app/components/base/toast'
+import { isReRankModelSelected } from '@/app/components/datasets/common/check-rerank-model'
+import { DataSourceProvider } from '@/models/common'
+import {
+  DataSourceType,
+} from '@/models/datasets'
+import { getNotionInfo, getWebsiteInfo, useCreateDocument, useCreateFirstDocument } from '@/service/knowledge/use-create-dataset'
+import { useInvalidDatasetList } from '@/service/knowledge/use-dataset'
+import { IndexingType } from './use-indexing-config'
+import { MAXIMUM_CHUNK_TOKEN_LENGTH } from './use-segmentation-state'
+
+export type UseDocumentCreationOptions = {
+  datasetId?: string
+  isSetting?: boolean
+  documentDetail?: FullDocumentDetail
+  dataSourceType: DataSourceType
+  files: CustomFile[]
+  notionPages: NotionPage[]
+  notionCredentialId: string
+  websitePages: CrawlResultItem[]
+  crawlOptions?: CrawlOptions
+  websiteCrawlProvider?: DataSourceProvider
+  websiteCrawlJobId?: string
+  // Callbacks
+  onStepChange?: (delta: number) => void
+  updateIndexingTypeCache?: (type: string) => void
+  updateResultCache?: (res: createDocumentResponse) => void
+  updateRetrievalMethodCache?: (method: RETRIEVE_METHOD | '') => void
+  onSave?: () => void
+  mutateDatasetRes?: () => void
+}
+
+export type ValidationParams = {
+  segmentationType: string
+  maxChunkLength: number
+  limitMaxChunkLength: number
+  overlap: number
+  indexType: IndexingType
+  embeddingModel: DefaultModel
+  rerankModelList: Model[]
+  retrievalConfig: RetrievalConfig
+}
+
+export const useDocumentCreation = (options: UseDocumentCreationOptions) => {
+  const { t } = useTranslation()
+  const {
+    datasetId,
+    isSetting,
+    documentDetail,
+    dataSourceType,
+    files,
+    notionPages,
+    notionCredentialId,
+    websitePages,
+    crawlOptions,
+    websiteCrawlProvider = DataSourceProvider.jinaReader,
+    websiteCrawlJobId = '',
+    onStepChange,
+    updateIndexingTypeCache,
+    updateResultCache,
+    updateRetrievalMethodCache,
+    onSave,
+    mutateDatasetRes,
+  } = options
+
+  const createFirstDocumentMutation = useCreateFirstDocument()
+  const createDocumentMutation = useCreateDocument(datasetId!)
+  const invalidDatasetList = useInvalidDatasetList()
+
+  const isCreating = createFirstDocumentMutation.isPending || createDocumentMutation.isPending
+
+  // Validate creation params
+  const validateParams = useCallback((params: ValidationParams): boolean => {
+    const {
+      segmentationType,
+      maxChunkLength,
+      limitMaxChunkLength,
+      overlap,
+      indexType,
+      embeddingModel,
+      rerankModelList,
+      retrievalConfig,
+    } = params
+
+    if (segmentationType === 'general' && overlap > maxChunkLength) {
+      Toast.notify({ type: 'error', message: t('stepTwo.overlapCheck', { ns: 'datasetCreation' }) })
+      return false
+    }
+
+    if (segmentationType === 'general' && maxChunkLength > limitMaxChunkLength) {
+      Toast.notify({
+        type: 'error',
+        message: t('stepTwo.maxLengthCheck', { ns: 'datasetCreation', limit: limitMaxChunkLength }),
+      })
+      return false
+    }
+
+    if (!isSetting) {
+      if (indexType === IndexingType.QUALIFIED && (!embeddingModel.model || !embeddingModel.provider)) {
+        Toast.notify({
+          type: 'error',
+          message: t('datasetConfig.embeddingModelRequired', { ns: 'appDebug' }),
+        })
+        return false
+      }
+
+      if (!isReRankModelSelected({
+        rerankModelList,
+        retrievalConfig,
+        indexMethod: indexType,
+      })) {
+        Toast.notify({ type: 'error', message: t('datasetConfig.rerankModelRequired', { ns: 'appDebug' }) })
+        return false
+      }
+    }
+
+    return true
+  }, [t, isSetting])
+
+  // Build creation params
+  const buildCreationParams = useCallback((
+    currentDocForm: ChunkingMode,
+    docLanguage: string,
+    processRule: ProcessRule,
+    retrievalConfig: RetrievalConfig,
+    embeddingModel: DefaultModel,
+    indexingTechnique: string,
+  ): CreateDocumentReq | null => {
+    if (isSetting) {
+      return {
+        original_document_id: documentDetail?.id,
+        doc_form: currentDocForm,
+        doc_language: docLanguage,
+        process_rule: processRule,
+        retrieval_model: retrievalConfig,
+        embedding_model: embeddingModel.model,
+        embedding_model_provider: embeddingModel.provider,
+        indexing_technique: indexingTechnique,
+      } as CreateDocumentReq
+    }
+
+    const params: CreateDocumentReq = {
+      data_source: {
+        type: dataSourceType,
+        info_list: {
+          data_source_type: dataSourceType,
+        },
+      },
+      indexing_technique: indexingTechnique,
+      process_rule: processRule,
+      doc_form: currentDocForm,
+      doc_language: docLanguage,
+      retrieval_model: retrievalConfig,
+      embedding_model: embeddingModel.model,
+      embedding_model_provider: embeddingModel.provider,
+    } as CreateDocumentReq
+
+    // Add data source specific info
+    if (dataSourceType === DataSourceType.FILE) {
+      params.data_source!.info_list.file_info_list = {
+        file_ids: files.map(file => file.id || '').filter(Boolean),
+      }
+    }
+    if (dataSourceType === DataSourceType.NOTION)
+      params.data_source!.info_list.notion_info_list = getNotionInfo(notionPages, notionCredentialId)
+
+    if (dataSourceType === DataSourceType.WEB) {
+      params.data_source!.info_list.website_info_list = getWebsiteInfo({
+        websiteCrawlProvider,
+        websiteCrawlJobId,
+        websitePages,
+        crawlOptions,
+      })
+    }
+
+    return params
+  }, [
+    isSetting,
+    documentDetail,
+    dataSourceType,
+    files,
+    notionPages,
+    notionCredentialId,
+    websitePages,
+    websiteCrawlProvider,
+    websiteCrawlJobId,
+    crawlOptions,
+  ])
+
+  // Execute creation
+  const executeCreation = useCallback(async (
+    params: CreateDocumentReq,
+    indexType: IndexingType,
+    retrievalConfig: RetrievalConfig,
+  ) => {
+    if (!datasetId) {
+      await createFirstDocumentMutation.mutateAsync(params, {
+        onSuccess(data) {
+          updateIndexingTypeCache?.(indexType)
+          updateResultCache?.(data)
+          updateRetrievalMethodCache?.(retrievalConfig.search_method as RETRIEVE_METHOD)
+        },
+      })
+    }
+    else {
+      await createDocumentMutation.mutateAsync(params, {
+        onSuccess(data) {
+          updateIndexingTypeCache?.(indexType)
+          updateResultCache?.(data)
+          updateRetrievalMethodCache?.(retrievalConfig.search_method as RETRIEVE_METHOD)
+        },
+      })
+    }
+
+    mutateDatasetRes?.()
+    invalidDatasetList()
+
+    trackEvent('create_datasets', {
+      data_source_type: dataSourceType,
+      indexing_technique: indexType,
+    })
+
+    onStepChange?.(+1)
+
+    if (isSetting)
+      onSave?.()
+  }, [
+    datasetId,
+    createFirstDocumentMutation,
+    createDocumentMutation,
+    updateIndexingTypeCache,
+    updateResultCache,
+    updateRetrievalMethodCache,
+    mutateDatasetRes,
+    invalidDatasetList,
+    dataSourceType,
+    onStepChange,
+    isSetting,
+    onSave,
+  ])
+
+  // Validate preview params
+  const validatePreviewParams = useCallback((maxChunkLength: number): boolean => {
+    if (maxChunkLength > MAXIMUM_CHUNK_TOKEN_LENGTH) {
+      Toast.notify({
+        type: 'error',
+        message: t('stepTwo.maxLengthCheck', { ns: 'datasetCreation', limit: MAXIMUM_CHUNK_TOKEN_LENGTH }),
+      })
+      return false
+    }
+    return true
+  }, [t])
+
+  return {
+    isCreating,
+    validateParams,
+    buildCreationParams,
+    executeCreation,
+    validatePreviewParams,
+  }
+}
+
+export type DocumentCreation = ReturnType<typeof useDocumentCreation>

web/app/components/datasets/create/step-two/hooks/use-indexing-config.ts

@@ -0,0 +1,143 @@
+import type { DefaultModel } from '@/app/components/header/account-setting/model-provider-page/declarations'
+import type { RetrievalConfig } from '@/types/app'
+import { useEffect, useMemo, useState } from 'react'
+import { checkShowMultiModalTip } from '@/app/components/datasets/settings/utils'
+import { ModelTypeEnum } from '@/app/components/header/account-setting/model-provider-page/declarations'
+import { useDefaultModel, useModelList, useModelListAndDefaultModelAndCurrentProviderAndModel } from '@/app/components/header/account-setting/model-provider-page/hooks'
+import { RETRIEVE_METHOD } from '@/types/app'
+
+export enum IndexingType {
+  QUALIFIED = 'high_quality',
+  ECONOMICAL = 'economy',
+}
+
+const DEFAULT_RETRIEVAL_CONFIG: RetrievalConfig = {
+  search_method: RETRIEVE_METHOD.semantic,
+  reranking_enable: false,
+  reranking_model: {
+    reranking_provider_name: '',
+    reranking_model_name: '',
+  },
+  top_k: 3,
+  score_threshold_enabled: false,
+  score_threshold: 0.5,
+}
+
+export type UseIndexingConfigOptions = {
+  initialIndexType?: IndexingType
+  initialEmbeddingModel?: DefaultModel
+  initialRetrievalConfig?: RetrievalConfig
+  isAPIKeySet: boolean
+  hasSetIndexType: boolean
+}
+
+export const useIndexingConfig = (options: UseIndexingConfigOptions) => {
+  const {
+    initialIndexType,
+    initialEmbeddingModel,
+    initialRetrievalConfig,
+    isAPIKeySet,
+    hasSetIndexType,
+  } = options
+
+  // Rerank model
+  const {
+    modelList: rerankModelList,
+    defaultModel: rerankDefaultModel,
+    currentModel: isRerankDefaultModelValid,
+  } = useModelListAndDefaultModelAndCurrentProviderAndModel(ModelTypeEnum.rerank)
+
+  // Embedding model list
+  const { data: embeddingModelList } = useModelList(ModelTypeEnum.textEmbedding)
+  const { data: defaultEmbeddingModel } = useDefaultModel(ModelTypeEnum.textEmbedding)
+
+  // Index type state
+  const [indexType, setIndexType] = useState<IndexingType>(() => {
+    if (initialIndexType)
+      return initialIndexType
+    return isAPIKeySet ? IndexingType.QUALIFIED : IndexingType.ECONOMICAL
+  })
+
+  // Embedding model state
+  const [embeddingModel, setEmbeddingModel] = useState<DefaultModel>(
+    initialEmbeddingModel ?? {
+      provider: defaultEmbeddingModel?.provider.provider || '',
+      model: defaultEmbeddingModel?.model || '',
+    },
+  )
+
+  // Retrieval config state
+  const [retrievalConfig, setRetrievalConfig] = useState<RetrievalConfig>(
+    initialRetrievalConfig ?? DEFAULT_RETRIEVAL_CONFIG,
+  )
+
+  // Sync retrieval config with rerank model when available
+  useEffect(() => {
+    if (initialRetrievalConfig)
+      return
+
+    setRetrievalConfig({
+      search_method: RETRIEVE_METHOD.semantic,
+      reranking_enable: !!isRerankDefaultModelValid,
+      reranking_model: {
+        reranking_provider_name: isRerankDefaultModelValid ? rerankDefaultModel?.provider.provider ?? '' : '',
+        reranking_model_name: isRerankDefaultModelValid ? rerankDefaultModel?.model ?? '' : '',
+      },
+      top_k: 3,
+      score_threshold_enabled: false,
+      score_threshold: 0.5,
+    })
+  }, [rerankDefaultModel, isRerankDefaultModelValid, initialRetrievalConfig])
+
+  // Sync index type with props
+  useEffect(() => {
+    if (initialIndexType)
+      setIndexType(initialIndexType)
+    else
+      setIndexType(isAPIKeySet ? IndexingType.QUALIFIED : IndexingType.ECONOMICAL)
+  }, [isAPIKeySet, initialIndexType])
+
+  // Show multimodal tip
+  const showMultiModalTip = useMemo(() => {
+    return checkShowMultiModalTip({
+      embeddingModel,
+      rerankingEnable: retrievalConfig.reranking_enable,
+      rerankModel: {
+        rerankingProviderName: retrievalConfig.reranking_model.reranking_provider_name,
+        rerankingModelName: retrievalConfig.reranking_model.reranking_model_name,
+      },
+      indexMethod: indexType,
+      embeddingModelList,
+      rerankModelList,
+    })
+  }, [embeddingModel, retrievalConfig, indexType, embeddingModelList, rerankModelList])
+
+  // Get effective indexing technique
+  const getIndexingTechnique = () => initialIndexType || indexType
+
+  return {
+    // Index type
+    indexType,
+    setIndexType,
+    hasSetIndexType,
+    getIndexingTechnique,
+
+    // Embedding model
+    embeddingModel,
+    setEmbeddingModel,
+    embeddingModelList,
+    defaultEmbeddingModel,
+
+    // Retrieval config
+    retrievalConfig,
+    setRetrievalConfig,
+    rerankModelList,
+    rerankDefaultModel,
+    isRerankDefaultModelValid,
+
+    // Computed
+    showMultiModalTip,
+  }
+}
+
+export type IndexingConfig = ReturnType<typeof useIndexingConfig>

web/app/components/datasets/create/step-two/hooks/use-indexing-estimate.ts

@@ -0,0 +1,123 @@
+import type { IndexingType } from './use-indexing-config'
+import type { NotionPage } from '@/models/common'
+import type { ChunkingMode, CrawlOptions, CrawlResultItem, CustomFile, ProcessRule } from '@/models/datasets'
+import { useCallback } from 'react'
+import { DataSourceProvider } from '@/models/common'
+import { DataSourceType } from '@/models/datasets'
+import {
+  useFetchFileIndexingEstimateForFile,
+  useFetchFileIndexingEstimateForNotion,
+  useFetchFileIndexingEstimateForWeb,
+} from '@/service/knowledge/use-create-dataset'
+
+export type UseIndexingEstimateOptions = {
+  dataSourceType: DataSourceType
+  datasetId?: string
+  // Document settings
+  currentDocForm: ChunkingMode
+  docLanguage: string
+  // File data source
+  files: CustomFile[]
+  previewFileName?: string
+  // Notion data source
+  previewNotionPage: NotionPage
+  notionCredentialId: string
+  // Website data source
+  previewWebsitePage: CrawlResultItem
+  crawlOptions?: CrawlOptions
+  websiteCrawlProvider?: DataSourceProvider
+  websiteCrawlJobId?: string
+  // Processing
+  indexingTechnique: IndexingType
+  processRule: ProcessRule
+}
+
+export const useIndexingEstimate = (options: UseIndexingEstimateOptions) => {
+  const {
+    dataSourceType,
+    datasetId,
+    currentDocForm,
+    docLanguage,
+    files,
+    previewFileName,
+    previewNotionPage,
+    notionCredentialId,
+    previewWebsitePage,
+    crawlOptions,
+    websiteCrawlProvider,
+    websiteCrawlJobId,
+    indexingTechnique,
+    processRule,
+  } = options
+
+  // File indexing estimate
+  const fileQuery = useFetchFileIndexingEstimateForFile({
+    docForm: currentDocForm,
+    docLanguage,
+    dataSourceType: DataSourceType.FILE,
+    files: previewFileName
+      ? [files.find(file => file.name === previewFileName)!]
+      : files,
+    indexingTechnique,
+    processRule,
+    dataset_id: datasetId!,
+  })
+
+  // Notion indexing estimate
+  const notionQuery = useFetchFileIndexingEstimateForNotion({
+    docForm: currentDocForm,
+    docLanguage,
+    dataSourceType: DataSourceType.NOTION,
+    notionPages: [previewNotionPage],
+    indexingTechnique,
+    processRule,
+    dataset_id: datasetId || '',
+    credential_id: notionCredentialId,
+  })
+
+  // Website indexing estimate
+  const websiteQuery = useFetchFileIndexingEstimateForWeb({
+    docForm: currentDocForm,
+    docLanguage,
+    dataSourceType: DataSourceType.WEB,
+    websitePages: [previewWebsitePage],
+    crawlOptions,
+    websiteCrawlProvider: websiteCrawlProvider ?? DataSourceProvider.jinaReader,
+    websiteCrawlJobId: websiteCrawlJobId ?? '',
+    indexingTechnique,
+    processRule,
+    dataset_id: datasetId || '',
+  })
+
+  // Get current mutation based on data source type
+  const getCurrentMutation = useCallback(() => {
+    if (dataSourceType === DataSourceType.FILE)
+      return fileQuery
+    if (dataSourceType === DataSourceType.NOTION)
+      return notionQuery
+    return websiteQuery
+  }, [dataSourceType, fileQuery, notionQuery, websiteQuery])
+
+  const currentMutation = getCurrentMutation()
+
+  // Trigger estimate fetch
+  const fetchEstimate = useCallback(() => {
+    if (dataSourceType === DataSourceType.FILE)
+      fileQuery.mutate()
+    else if (dataSourceType === DataSourceType.NOTION)
+      notionQuery.mutate()
+    else
+      websiteQuery.mutate()
+  }, [dataSourceType, fileQuery, notionQuery, websiteQuery])
+
+  return {
+    currentMutation,
+    estimate: currentMutation.data,
+    isIdle: currentMutation.isIdle,
+    isPending: currentMutation.isPending,
+    fetchEstimate,
+    reset: currentMutation.reset,
+  }
+}
+
+export type IndexingEstimate = ReturnType<typeof useIndexingEstimate>

web/app/components/datasets/create/step-two/hooks/use-preview-state.ts

@@ -0,0 +1,127 @@
+import type { NotionPage } from '@/models/common'
+import type { CrawlResultItem, CustomFile, DocumentItem, FullDocumentDetail } from '@/models/datasets'
+import { useCallback, useState } from 'react'
+import { DataSourceType } from '@/models/datasets'
+
+export type UsePreviewStateOptions = {
+  dataSourceType: DataSourceType
+  files: CustomFile[]
+  notionPages: NotionPage[]
+  websitePages: CrawlResultItem[]
+  documentDetail?: FullDocumentDetail
+  datasetId?: string
+}
+
+export const usePreviewState = (options: UsePreviewStateOptions) => {
+  const {
+    dataSourceType,
+    files,
+    notionPages,
+    websitePages,
+    documentDetail,
+    datasetId,
+  } = options
+
+  // File preview state
+  const [previewFile, setPreviewFile] = useState<DocumentItem>(
+    (datasetId && documentDetail)
+      ? documentDetail.file
+      : files[0],
+  )
+
+  // Notion page preview state
+  const [previewNotionPage, setPreviewNotionPage] = useState<NotionPage>(
+    (datasetId && documentDetail)
+      ? documentDetail.notion_page
+      : notionPages[0],
+  )
+
+  // Website page preview state
+  const [previewWebsitePage, setPreviewWebsitePage] = useState<CrawlResultItem>(
+    (datasetId && documentDetail)
+      ? documentDetail.website_page
+      : websitePages[0],
+  )
+
+  // Get preview items for document picker based on data source type
+  const getPreviewPickerItems = useCallback(() => {
+    if (dataSourceType === DataSourceType.FILE) {
+      return files as Array<Required<CustomFile>>
+    }
+    if (dataSourceType === DataSourceType.NOTION) {
+      return notionPages.map(page => ({
+        id: page.page_id,
+        name: page.page_name,
+        extension: 'md',
+      }))
+    }
+    if (dataSourceType === DataSourceType.WEB) {
+      return websitePages.map(page => ({
+        id: page.source_url,
+        name: page.title,
+        extension: 'md',
+      }))
+    }
+    return []
+  }, [dataSourceType, files, notionPages, websitePages])
+
+  // Get current preview value for picker
+  const getPreviewPickerValue = useCallback(() => {
+    if (dataSourceType === DataSourceType.FILE) {
+      return previewFile as Required<CustomFile>
+    }
+    if (dataSourceType === DataSourceType.NOTION) {
+      return {
+        id: previewNotionPage?.page_id || '',
+        name: previewNotionPage?.page_name || '',
+        extension: 'md',
+      }
+    }
+    if (dataSourceType === DataSourceType.WEB) {
+      return {
+        id: previewWebsitePage?.source_url || '',
+        name: previewWebsitePage?.title || '',
+        extension: 'md',
+      }
+    }
+    return { id: '', name: '', extension: '' }
+  }, [dataSourceType, previewFile, previewNotionPage, previewWebsitePage])
+
+  // Handle preview change
+  const handlePreviewChange = useCallback((selected: { id: string, name: string }) => {
+    if (dataSourceType === DataSourceType.FILE) {
+      setPreviewFile(selected as DocumentItem)
+    }
+    else if (dataSourceType === DataSourceType.NOTION) {
+      const selectedPage = notionPages.find(page => page.page_id === selected.id)
+      if (selectedPage)
+        setPreviewNotionPage(selectedPage)
+    }
+    else if (dataSourceType === DataSourceType.WEB) {
+      const selectedPage = websitePages.find(page => page.source_url === selected.id)
+      if (selectedPage)
+        setPreviewWebsitePage(selectedPage)
+    }
+  }, [dataSourceType, notionPages, websitePages])
+
+  return {
+    // File preview
+    previewFile,
+    setPreviewFile,
+
+    // Notion preview
+    previewNotionPage,
+    setPreviewNotionPage,
+
+    // Website preview
+    previewWebsitePage,
+    setPreviewWebsitePage,
+
+    // Picker helpers
+    getPreviewPickerItems,
+    getPreviewPickerValue,
+    handlePreviewChange,
+  }
+}
+
+export type PreviewState = ReturnType<typeof usePreviewState>

web/app/components/datasets/create/step-two/hooks/use-segmentation-state.ts

@@ -0,0 +1,222 @@
+import type { ParentMode, PreProcessingRule, ProcessRule, Rules } from '@/models/datasets'
+import { useCallback, useState } from 'react'
+import { ChunkingMode, ProcessMode } from '@/models/datasets'
+import escape from './escape'
+import unescape from './unescape'
+
+// Constants
+export const DEFAULT_SEGMENT_IDENTIFIER = '\\n\\n'
+export const DEFAULT_MAXIMUM_CHUNK_LENGTH = 1024
+export const DEFAULT_OVERLAP = 50
+export const MAXIMUM_CHUNK_TOKEN_LENGTH = Number.parseInt(
+  globalThis.document?.body?.getAttribute('data-public-indexing-max-segmentation-tokens-length') || '4000',
+  10,
+)
+
+export type ParentChildConfig = {
+  chunkForContext: ParentMode
+  parent: {
+    delimiter: string
+    maxLength: number
+  }
+  child: {
+    delimiter: string
+    maxLength: number
+  }
+}
+
+export const defaultParentChildConfig: ParentChildConfig = {
+  chunkForContext: 'paragraph',
+  parent: {
+    delimiter: '\\n\\n',
+    maxLength: 1024,
+  },
+  child: {
+    delimiter: '\\n',
+    maxLength: 512,
+  },
+}
+
+export type UseSegmentationStateOptions = {
+  initialSegmentationType?: ProcessMode
+}
+
+export const useSegmentationState = (options: UseSegmentationStateOptions = {}) => {
+  const { initialSegmentationType } = options
+
+  // Segmentation type (general or parent-child)
+  const [segmentationType, setSegmentationType] = useState<ProcessMode>(
+    initialSegmentationType ?? ProcessMode.general,
+  )
+
+  // General chunking settings
+  const [segmentIdentifier, doSetSegmentIdentifier] = useState(DEFAULT_SEGMENT_IDENTIFIER)
+  const [maxChunkLength, setMaxChunkLength] = useState(DEFAULT_MAXIMUM_CHUNK_LENGTH)
+  const [limitMaxChunkLength, setLimitMaxChunkLength] = useState(MAXIMUM_CHUNK_TOKEN_LENGTH)
+  const [overlap, setOverlap] = useState(DEFAULT_OVERLAP)
+
+  // Pre-processing rules
+  const [rules, setRules] = useState<PreProcessingRule[]>([])
+  const [defaultConfig, setDefaultConfig] = useState<Rules>()
+
+  // Parent-child config
+  const [parentChildConfig, setParentChildConfig] = useState<ParentChildConfig>(defaultParentChildConfig)
+
+  // Escaped segment identifier setter
+  const setSegmentIdentifier = useCallback((value: string, canEmpty?: boolean) => {
+    if (value) {
+      doSetSegmentIdentifier(escape(value))
+    }
+    else {
+      doSetSegmentIdentifier(canEmpty ? '' : DEFAULT_SEGMENT_IDENTIFIER)
+    }
+  }, [])
+
+  // Rule toggle handler
+  const toggleRule = useCallback((id: string) => {
+    setRules(prev => prev.map(rule =>
+      rule.id === id ? { ...rule, enabled: !rule.enabled } : rule,
+    ))
+  }, [])
+
+  // Reset to defaults
+  const resetToDefaults = useCallback(() => {
+    if (defaultConfig) {
+      setSegmentIdentifier(defaultConfig.segmentation.separator)
+      setMaxChunkLength(defaultConfig.segmentation.max_tokens)
+      setOverlap(defaultConfig.segmentation.chunk_overlap!)
+      setRules(defaultConfig.pre_processing_rules)
+    }
+    setParentChildConfig(defaultParentChildConfig)
+  }, [defaultConfig, setSegmentIdentifier])
+
+  // Apply config from document detail
+  const applyConfigFromRules = useCallback((rulesConfig: Rules, isHierarchical: boolean) => {
+    const separator = rulesConfig.segmentation.separator
+    const max = rulesConfig.segmentation.max_tokens
+    const chunkOverlap = rulesConfig.segmentation.chunk_overlap
+
+    setSegmentIdentifier(separator)
+    setMaxChunkLength(max)
+    setOverlap(chunkOverlap!)
+    setRules(rulesConfig.pre_processing_rules)
+    setDefaultConfig(rulesConfig)
+
+    if (isHierarchical) {
+      setParentChildConfig({
+        chunkForContext: rulesConfig.parent_mode || 'paragraph',
+        parent: {
+          delimiter: escape(rulesConfig.segmentation.separator),
+          maxLength: rulesConfig.segmentation.max_tokens,
+        },
+        child: {
+          delimiter: escape(rulesConfig.subchunk_segmentation!.separator),
+          maxLength: rulesConfig.subchunk_segmentation!.max_tokens,
+        },
+      })
+    }
+  }, [setSegmentIdentifier])
+
+  // Get process rule for API
+  const getProcessRule = useCallback((docForm: ChunkingMode): ProcessRule => {
+    if (docForm === ChunkingMode.parentChild) {
+      return {
+        rules: {
+          pre_processing_rules: rules,
+          segmentation: {
+            separator: unescape(parentChildConfig.parent.delimiter),
+            max_tokens: parentChildConfig.parent.maxLength,
+          },
+          parent_mode: parentChildConfig.chunkForContext,
+          subchunk_segmentation: {
+            separator: unescape(parentChildConfig.child.delimiter),
+            max_tokens: parentChildConfig.child.maxLength,
+          },
+        },
+        mode: 'hierarchical',
+      } as ProcessRule
+    }
+
+    return {
+      rules: {
+        pre_processing_rules: rules,
+        segmentation: {
+          separator: unescape(segmentIdentifier),
+          max_tokens: maxChunkLength,
+          chunk_overlap: overlap,
+        },
+      },
+      mode: segmentationType,
+    } as ProcessRule
+  }, [rules, parentChildConfig, segmentIdentifier, maxChunkLength, overlap, segmentationType])
+
+  // Update parent config field
+  const updateParentConfig = useCallback((field: 'delimiter' | 'maxLength', value: string | number) => {
+    setParentChildConfig((prev) => {
+      let newValue: string | number
+      if (field === 'delimiter')
+        newValue = value ? escape(value as string) : ''
+      else
+        newValue = value
+      return {
+        ...prev,
+        parent: { ...prev.parent, [field]: newValue },
+      }
+    })
+  }, [])
+
+  // Update child config field
+  const updateChildConfig = useCallback((field: 'delimiter' | 'maxLength', value: string | number) => {
+    setParentChildConfig((prev) => {
+      let newValue: string | number
+      if (field === 'delimiter')
+        newValue = value ? escape(value as string) : ''
+      else
+        newValue = value
+      return {
+        ...prev,
+        child: { ...prev.child, [field]: newValue },
+      }
+    })
+  }, [])
+
+  // Set chunk for context mode
+  const setChunkForContext = useCallback((mode: ParentMode) => {
+    setParentChildConfig(prev => ({ ...prev, chunkForContext: mode }))
+  }, [])
+
+  return {
+    // General chunking state
+    segmentationType,
+    setSegmentationType,
+    segmentIdentifier,
+    setSegmentIdentifier,
+    maxChunkLength,
+    setMaxChunkLength,
+    limitMaxChunkLength,
+    setLimitMaxChunkLength,
+    overlap,
+    setOverlap,
+
+    // Rules
+    rules,
+    setRules,
+    defaultConfig,
+    setDefaultConfig,
+    toggleRule,
+
+    // Parent-child config
+    parentChildConfig,
+    setParentChildConfig,
+    updateParentConfig,
+    updateChildConfig,
+    setChunkForContext,
+
+    // Actions
+    resetToDefaults,
+    applyConfigFromRules,
+    getProcessRule,
+  }
+}
+
+export type SegmentationState = ReturnType<typeof useSegmentationState>

web/app/components/datasets/create/step-two/types.ts

@@ -0,0 +1,28 @@
+import type { IndexingType } from './hooks'
+import type { DataSourceProvider, NotionPage } from '@/models/common'
+import type { CrawlOptions, CrawlResultItem, createDocumentResponse, CustomFile, DataSourceType, FullDocumentDetail } from '@/models/datasets'
+import type { RETRIEVE_METHOD } from '@/types/app'
+
+export type StepTwoProps = {
+  isSetting?: boolean
+  documentDetail?: FullDocumentDetail
+  isAPIKeySet: boolean
+  onSetting: () => void
+  datasetId?: string
+  indexingType?: IndexingType
+  retrievalMethod?: string
+  dataSourceType: DataSourceType
+  files: CustomFile[]
+  notionPages?: NotionPage[]
+  notionCredentialId: string
+  websitePages?: CrawlResultItem[]
+  crawlOptions?: CrawlOptions
+  websiteCrawlProvider?: DataSourceProvider
+  websiteCrawlJobId?: string
+  onStepChange?: (delta: number) => void
+  updateIndexingTypeCache?: (type: string) => void
+  updateRetrievalMethodCache?: (method: RETRIEVE_METHOD | '') => void
+  updateResultCache?: (res: createDocumentResponse) => void
+  onSave?: () => void
+  onCancel?: () => void
+}

web/app/components/plugins/marketplace/atoms.ts

@@ -0,0 +1,81 @@
+import type { ActivePluginType } from './constants'
+import type { PluginsSort, SearchParamsFromCollection } from './types'
+import { atom, useAtom, useAtomValue, useSetAtom } from 'jotai'
+import { useQueryState } from 'nuqs'
+import { useCallback } from 'react'
+import { DEFAULT_SORT, PLUGIN_CATEGORY_WITH_COLLECTIONS } from './constants'
+import { marketplaceSearchParamsParsers } from './search-params'
+
+const marketplaceSortAtom = atom<PluginsSort>(DEFAULT_SORT)
+export function useMarketplaceSort() {
+  return useAtom(marketplaceSortAtom)
+}
+export function useMarketplaceSortValue() {
+  return useAtomValue(marketplaceSortAtom)
+}
+export function useSetMarketplaceSort() {
+  return useSetAtom(marketplaceSortAtom)
+}
+
+/**
+ * Preserve the state for marketplace
+ */
+export const preserveSearchStateInQueryAtom = atom<boolean>(false)
+
+const searchPluginTextAtom = atom<string>('')
+const activePluginTypeAtom = atom<ActivePluginType>('all')
+const filterPluginTagsAtom = atom<string[]>([])
+
+export function useSearchPluginText() {
+  const preserveSearchStateInQuery = useAtomValue(preserveSearchStateInQueryAtom)
+  const queryState = useQueryState('q', marketplaceSearchParamsParsers.q)
+  const atomState = useAtom(searchPluginTextAtom)
+  return preserveSearchStateInQuery ? queryState : atomState
+}
+export function useActivePluginType() {
+  const preserveSearchStateInQuery = useAtomValue(preserveSearchStateInQueryAtom)
+  const queryState = useQueryState('category', marketplaceSearchParamsParsers.category)
+  const atomState = useAtom(activePluginTypeAtom)
+  return preserveSearchStateInQuery ? queryState : atomState
+}
+export function useFilterPluginTags() {
+  const preserveSearchStateInQuery = useAtomValue(preserveSearchStateInQueryAtom)
+  const queryState = useQueryState('tags', marketplaceSearchParamsParsers.tags)
+  const atomState = useAtom(filterPluginTagsAtom)
+  return preserveSearchStateInQuery ? queryState : atomState
+}
+
+/**
+ * Not all categories have collections, so we need to
+ * force the search mode for those categories.
+ */
+export const searchModeAtom = atom<true | null>(null)
+
+export function useMarketplaceSearchMode() {
+  const [searchPluginText] = useSearchPluginText()
+  const [filterPluginTags] = useFilterPluginTags()
+  const [activePluginType] = useActivePluginType()
+
+  const searchMode = useAtomValue(searchModeAtom)
+  const isSearchMode = !!searchPluginText
+    || filterPluginTags.length > 0
+    || (searchMode ?? (!PLUGIN_CATEGORY_WITH_COLLECTIONS.has(activePluginType)))
+  return isSearchMode
+}
+
+export function useMarketplaceMoreClick() {
+  const [,setQ] = useSearchPluginText()
+  const setSort = useSetAtom(marketplaceSortAtom)
+  const setSearchMode = useSetAtom(searchModeAtom)
+
+  return useCallback((searchParams?: SearchParamsFromCollection) => {
+    if (!searchParams)
+      return
+    setQ(searchParams?.query || '')
+    setSort({
+      sortBy: searchParams?.sort_by || DEFAULT_SORT.sortBy,
+      sortOrder: searchParams?.sort_order || DEFAULT_SORT.sortOrder,
+    })
+    setSearchMode(true)
+  }, [setQ, setSort, setSearchMode])
+}

web/app/components/plugins/marketplace/constants.ts

@@ -1,6 +1,30 @@
+import { PluginCategoryEnum } from '../types'
+
 export const DEFAULT_SORT = {
   sortBy: 'install_count',
   sortOrder: 'DESC',
 }
 
 export const SCROLL_BOTTOM_THRESHOLD = 100
+
+export const PLUGIN_TYPE_SEARCH_MAP = {
+  all: 'all',
+  model: PluginCategoryEnum.model,
+  tool: PluginCategoryEnum.tool,
+  agent: PluginCategoryEnum.agent,
+  extension: PluginCategoryEnum.extension,
+  datasource: PluginCategoryEnum.datasource,
+  trigger: PluginCategoryEnum.trigger,
+  bundle: 'bundle',
+} as const
+
+type ValueOf<T> = T[keyof T]
+
+export type ActivePluginType = ValueOf<typeof PLUGIN_TYPE_SEARCH_MAP>
+
+export const PLUGIN_CATEGORY_WITH_COLLECTIONS = new Set<ActivePluginType>(
+  [
+    PLUGIN_TYPE_SEARCH_MAP.all,
+    PLUGIN_TYPE_SEARCH_MAP.tool,
+  ],
+)

web/app/components/plugins/marketplace/context.tsx

@@ -1,332 +0,0 @@
-'use client'
-
-import type {
-  ReactNode,
-} from 'react'
-import type { TagKey } from '../constants'
-import type { Plugin } from '../types'
-import type {
-  MarketplaceCollection,
-  PluginsSort,
-  SearchParams,
-  SearchParamsFromCollection,
-} from './types'
-import { debounce } from 'es-toolkit/compat'
-import { noop } from 'es-toolkit/function'
-import {
-  useCallback,
-  useEffect,
-  useMemo,
-  useRef,
-  useState,
-} from 'react'
-import {
-  createContext,
-  useContextSelector,
-} from 'use-context-selector'
-import { useMarketplaceFilters } from '@/hooks/use-query-params'
-import { useInstalledPluginList } from '@/service/use-plugins'
-import {
-  getValidCategoryKeys,
-  getValidTagKeys,
-} from '../utils'
-import { DEFAULT_SORT } from './constants'
-import {
-  useMarketplaceCollectionsAndPlugins,
-  useMarketplaceContainerScroll,
-  useMarketplacePlugins,
-} from './hooks'
-import { PLUGIN_TYPE_SEARCH_MAP } from './plugin-type-switch'
-import {
-  getMarketplaceListCondition,
-  getMarketplaceListFilterType,
-} from './utils'
-
-export type MarketplaceContextValue = {
-  searchPluginText: string
-  handleSearchPluginTextChange: (text: string) => void
-  filterPluginTags: string[]
-  handleFilterPluginTagsChange: (tags: string[]) => void
-  activePluginType: string
-  handleActivePluginTypeChange: (type: string) => void
-  page: number
-  handlePageChange: () => void
-  plugins?: Plugin[]
-  pluginsTotal?: number
-  resetPlugins: () => void
-  sort: PluginsSort
-  handleSortChange: (sort: PluginsSort) => void
-  handleQueryPlugins: () => void
-  handleMoreClick: (searchParams: SearchParamsFromCollection) => void
-  marketplaceCollectionsFromClient?: MarketplaceCollection[]
-  setMarketplaceCollectionsFromClient: (collections: MarketplaceCollection[]) => void
-  marketplaceCollectionPluginsMapFromClient?: Record<string, Plugin[]>
-  setMarketplaceCollectionPluginsMapFromClient: (map: Record<string, Plugin[]>) => void
-  isLoading: boolean
-  isSuccessCollections: boolean
-}
-
-export const MarketplaceContext = createContext<MarketplaceContextValue>({
-  searchPluginText: '',
-  handleSearchPluginTextChange: noop,
-  filterPluginTags: [],
-  handleFilterPluginTagsChange: noop,
-  activePluginType: 'all',
-  handleActivePluginTypeChange: noop,
-  page: 1,
-  handlePageChange: noop,
-  plugins: undefined,
-  pluginsTotal: 0,
-  resetPlugins: noop,
-  sort: DEFAULT_SORT,
-  handleSortChange: noop,
-  handleQueryPlugins: noop,
-  handleMoreClick: noop,
-  marketplaceCollectionsFromClient: [],
-  setMarketplaceCollectionsFromClient: noop,
-  marketplaceCollectionPluginsMapFromClient: {},
-  setMarketplaceCollectionPluginsMapFromClient: noop,
-  isLoading: false,
-  isSuccessCollections: false,
-})
-
-type MarketplaceContextProviderProps = {
-  children: ReactNode
-  searchParams?: SearchParams
-  shouldExclude?: boolean
-  scrollContainerId?: string
-  showSearchParams?: boolean
-}
-
-export function useMarketplaceContext(selector: (value: MarketplaceContextValue) => any) {
-  return useContextSelector(MarketplaceContext, selector)
-}
-
-export const MarketplaceContextProvider = ({
-  children,
-  searchParams,
-  shouldExclude,
-  scrollContainerId,
-  showSearchParams,
-}: MarketplaceContextProviderProps) => {
-  // Use nuqs hook for URL-based filter state
-  const [urlFilters, setUrlFilters] = useMarketplaceFilters()
-
-  const { data, isSuccess } = useInstalledPluginList(!shouldExclude)
-  const exclude = useMemo(() => {
-    if (shouldExclude)
-      return data?.plugins.map(plugin => plugin.plugin_id)
-  }, [data?.plugins, shouldExclude])
-
-  // Initialize from URL params (legacy support) or use nuqs state
-  const queryFromSearchParams = searchParams?.q || urlFilters.q
-  const tagsFromSearchParams = getValidTagKeys(urlFilters.tags as TagKey[])
-  const hasValidTags = !!tagsFromSearchParams.length
-  const hasValidCategory = getValidCategoryKeys(urlFilters.category)
-  const categoryFromSearchParams = hasValidCategory || PLUGIN_TYPE_SEARCH_MAP.all
-
-  const [searchPluginText, setSearchPluginText] = useState(queryFromSearchParams)
-  const searchPluginTextRef = useRef(searchPluginText)
-  const [filterPluginTags, setFilterPluginTags] = useState<string[]>(tagsFromSearchParams)
-  const filterPluginTagsRef = useRef(filterPluginTags)
-  const [activePluginType, setActivePluginType] = useState(categoryFromSearchParams)
-  const activePluginTypeRef = useRef(activePluginType)
-  const [sort, setSort] = useState(DEFAULT_SORT)
-  const sortRef = useRef(sort)
-  const {
-    marketplaceCollections: marketplaceCollectionsFromClient,
-    setMarketplaceCollections: setMarketplaceCollectionsFromClient,
-    marketplaceCollectionPluginsMap: marketplaceCollectionPluginsMapFromClient,
-    setMarketplaceCollectionPluginsMap: setMarketplaceCollectionPluginsMapFromClient,
-    queryMarketplaceCollectionsAndPlugins,
-    isLoading,
-    isSuccess: isSuccessCollections,
-  } = useMarketplaceCollectionsAndPlugins()
-  const {
-    plugins,
-    total: pluginsTotal,
-    resetPlugins,
-    queryPlugins,
-    queryPluginsWithDebounced,
-    cancelQueryPluginsWithDebounced,
-    isLoading: isPluginsLoading,
-    fetchNextPage: fetchNextPluginsPage,
-    hasNextPage: hasNextPluginsPage,
-    page: pluginsPage,
-  } = useMarketplacePlugins()
-  const page = Math.max(pluginsPage || 0, 1)
-
-  useEffect(() => {
-    if (queryFromSearchParams || hasValidTags || hasValidCategory) {
-      queryPlugins({
-        query: queryFromSearchParams,
-        category: hasValidCategory,
-        tags: hasValidTags ? tagsFromSearchParams : [],
-        sortBy: sortRef.current.sortBy,
-        sortOrder: sortRef.current.sortOrder,
-        type: getMarketplaceListFilterType(activePluginTypeRef.current),
-      })
-    }
-    else {
-      if (shouldExclude && isSuccess) {
-        queryMarketplaceCollectionsAndPlugins({
-          exclude,
-          type: getMarketplaceListFilterType(activePluginTypeRef.current),
-        })
-      }
-    }
-  }, [queryPlugins, queryMarketplaceCollectionsAndPlugins, isSuccess, exclude])
-
-  const handleQueryMarketplaceCollectionsAndPlugins = useCallback(() => {
-    queryMarketplaceCollectionsAndPlugins({
-      category: activePluginTypeRef.current === PLUGIN_TYPE_SEARCH_MAP.all ? undefined : activePluginTypeRef.current,
-      condition: getMarketplaceListCondition(activePluginTypeRef.current),
-      exclude,
-      type: getMarketplaceListFilterType(activePluginTypeRef.current),
-    })
-    resetPlugins()
-  }, [exclude, queryMarketplaceCollectionsAndPlugins, resetPlugins])
-
-  const applyUrlFilters = useCallback(() => {
-    if (!showSearchParams)
-      return
-    const nextFilters = {
-      q: searchPluginTextRef.current,
-      category: activePluginTypeRef.current,
-      tags: filterPluginTagsRef.current,
-    }
-    const categoryChanged = urlFilters.category !== nextFilters.category
-    setUrlFilters(nextFilters, {
-      history: categoryChanged ? 'push' : 'replace',
-    })
-  }, [setUrlFilters, showSearchParams, urlFilters.category])
-
-  const debouncedUpdateSearchParams = useMemo(() => debounce(() => {
-    applyUrlFilters()
-  }, 500), [applyUrlFilters])
-
-  const handleUpdateSearchParams = useCallback((debounced?: boolean) => {
-    if (debounced) {
-      debouncedUpdateSearchParams()
-    }
-    else {
-      applyUrlFilters()
-    }
-  }, [applyUrlFilters, debouncedUpdateSearchParams])
-
-  const handleQueryPlugins = useCallback((debounced?: boolean) => {
-    handleUpdateSearchParams(debounced)
-    if (debounced) {
-      queryPluginsWithDebounced({
-        query: searchPluginTextRef.current,
-        category: activePluginTypeRef.current === PLUGIN_TYPE_SEARCH_MAP.all ? undefined : activePluginTypeRef.current,
-        tags: filterPluginTagsRef.current,
-        sortBy: sortRef.current.sortBy,
-        sortOrder: sortRef.current.sortOrder,
-        exclude,
-        type: getMarketplaceListFilterType(activePluginTypeRef.current),
-      })
-    }
-    else {
-      queryPlugins({
-        query: searchPluginTextRef.current,
-        category: activePluginTypeRef.current === PLUGIN_TYPE_SEARCH_MAP.all ? undefined : activePluginTypeRef.current,
-        tags: filterPluginTagsRef.current,
-        sortBy: sortRef.current.sortBy,
-        sortOrder: sortRef.current.sortOrder,
-        exclude,
-        type: getMarketplaceListFilterType(activePluginTypeRef.current),
-      })
-    }
-  }, [exclude, queryPluginsWithDebounced, queryPlugins, handleUpdateSearchParams])
-
-  const handleQuery = useCallback((debounced?: boolean) => {
-    if (!searchPluginTextRef.current && !filterPluginTagsRef.current.length) {
-      handleUpdateSearchParams(debounced)
-      cancelQueryPluginsWithDebounced()
-      handleQueryMarketplaceCollectionsAndPlugins()
-      return
-    }
-
-    handleQueryPlugins(debounced)
-  }, [handleQueryMarketplaceCollectionsAndPlugins, handleQueryPlugins, cancelQueryPluginsWithDebounced, handleUpdateSearchParams])
-
-  const handleSearchPluginTextChange = useCallback((text: string) => {
-    setSearchPluginText(text)
-    searchPluginTextRef.current = text
-
-    handleQuery(true)
-  }, [handleQuery])
-
-  const handleFilterPluginTagsChange = useCallback((tags: string[]) => {
-    setFilterPluginTags(tags)
-    filterPluginTagsRef.current = tags
-
-    handleQuery()
-  }, [handleQuery])
-
-  const handleActivePluginTypeChange = useCallback((type: string) => {
-    setActivePluginType(type)
-    activePluginTypeRef.current = type
-
-    handleQuery()
-  }, [handleQuery])
-
-  const handleSortChange = useCallback((sort: PluginsSort) => {
-    setSort(sort)
-    sortRef.current = sort
-
-    handleQueryPlugins()
-  }, [handleQueryPlugins])
-
-  const handlePageChange = useCallback(() => {
-    if (hasNextPluginsPage)
-      fetchNextPluginsPage()
-  }, [fetchNextPluginsPage, hasNextPluginsPage])
-
-  const handleMoreClick = useCallback((searchParams: SearchParamsFromCollection) => {
-    setSearchPluginText(searchParams?.query || '')
-    searchPluginTextRef.current = searchParams?.query || ''
-    setSort({
-      sortBy: searchParams?.sort_by || DEFAULT_SORT.sortBy,
-      sortOrder: searchParams?.sort_order || DEFAULT_SORT.sortOrder,
-    })
-    sortRef.current = {
-      sortBy: searchParams?.sort_by || DEFAULT_SORT.sortBy,
-      sortOrder: searchParams?.sort_order || DEFAULT_SORT.sortOrder,
-    }
-    handleQueryPlugins()
-  }, [handleQueryPlugins])
-
-  useMarketplaceContainerScroll(handlePageChange, scrollContainerId)
-
-  return (
-    <MarketplaceContext.Provider
-      value={{
-        searchPluginText,
-        handleSearchPluginTextChange,
-        filterPluginTags,
-        handleFilterPluginTagsChange,
-        activePluginType,
-        handleActivePluginTypeChange,
-        page,
-        handlePageChange,
-        plugins,
-        pluginsTotal,
-        resetPlugins,
-        sort,
-        handleSortChange,
-        handleQueryPlugins,
-        handleMoreClick,
-        marketplaceCollectionsFromClient,
-        setMarketplaceCollectionsFromClient,
-        marketplaceCollectionPluginsMapFromClient,
-        setMarketplaceCollectionPluginsMapFromClient,
-        isLoading: isLoading || isPluginsLoading,
-        isSuccessCollections,
-      }}
-    >
-      {children}
-    </MarketplaceContext.Provider>
-  )
-}

web/app/components/plugins/marketplace/hooks.ts

@@ -26,6 +26,9 @@ import {
   getMarketplacePluginsByCollectionId,
 } from './utils'
 
+/**
+ * @deprecated Use useMarketplaceCollectionsAndPlugins from query.ts instead
+ */
 export const useMarketplaceCollectionsAndPlugins = () => {
   const [queryParams, setQueryParams] = useState<CollectionsAndPluginsSearchParams>()
   const [marketplaceCollectionsOverride, setMarketplaceCollections] = useState<MarketplaceCollection[]>()
@@ -89,7 +92,9 @@ export const useMarketplacePluginsByCollectionId = (
     isSuccess,
   }
 }
-
+/**
+ * @deprecated Use useMarketplacePlugins from query.ts instead
+ */
 export const useMarketplacePlugins = () => {
   const queryClient = useQueryClient()
   const [queryParams, setQueryParams] = useState<PluginsSearchParams>()

web/app/components/plugins/marketplace/hydration-client.tsx

@@ -0,0 +1,15 @@
+'use client'
+
+import { useHydrateAtoms } from 'jotai/utils'
+import { preserveSearchStateInQueryAtom } from './atoms'
+
+export function HydrateMarketplaceAtoms({
+  preserveSearchStateInQuery,
+  children,
+}: {
+  preserveSearchStateInQuery: boolean
+  children: React.ReactNode
+}) {
+  useHydrateAtoms([[preserveSearchStateInQueryAtom, preserveSearchStateInQuery]])
+  return <>{children}</>
+}

web/app/components/plugins/marketplace/hydration-server.tsx

@@ -0,0 +1,45 @@
+import type { SearchParams } from 'nuqs'
+import { dehydrate, HydrationBoundary } from '@tanstack/react-query'
+import { createLoader } from 'nuqs/server'
+import { getQueryClientServer } from '@/context/query-client-server'
+import { PLUGIN_CATEGORY_WITH_COLLECTIONS } from './constants'
+import { marketplaceKeys } from './query'
+import { marketplaceSearchParamsParsers } from './search-params'
+import { getCollectionsParams, getMarketplaceCollectionsAndPlugins } from './utils'
+
+// The server side logic should move to marketplace's codebase so that we can get rid of Next.js
+
+async function getDehydratedState(searchParams?: Promise<SearchParams>) {
+  if (!searchParams) {
+    return
+  }
+  const loadSearchParams = createLoader(marketplaceSearchParamsParsers)
+  const params = await loadSearchParams(searchParams)
+
+  if (!PLUGIN_CATEGORY_WITH_COLLECTIONS.has(params.category)) {
+    return
+  }
+
+  const queryClient = getQueryClientServer()
+
+  await queryClient.prefetchQuery({
+    queryKey: marketplaceKeys.collections(getCollectionsParams(params.category)),
+    queryFn: () => getMarketplaceCollectionsAndPlugins(getCollectionsParams(params.category)),
+  })
+  return dehydrate(queryClient)
+}
+
+export async function HydrateQueryClient({
+  searchParams,
+  children,
+}: {
+  searchParams: Promise<SearchParams> | undefined
+  children: React.ReactNode
+}) {
+  const dehydratedState = await getDehydratedState(searchParams)
+  return (
+    <HydrationBoundary state={dehydratedState}>
+      {children}
+    </HydrationBoundary>
+  )
+}

web/app/components/plugins/marketplace/index.tsx

@@ -1,55 +1,39 @@
-import type { MarketplaceCollection, SearchParams } from './types'
-import type { Plugin } from '@/app/components/plugins/types'
+import type { SearchParams } from 'nuqs'
 import { TanstackQueryInitializer } from '@/context/query-client'
-import { MarketplaceContextProvider } from './context'
 import Description from './description'
+import { HydrateMarketplaceAtoms } from './hydration-client'
+import { HydrateQueryClient } from './hydration-server'
 import ListWrapper from './list/list-wrapper'
 import StickySearchAndSwitchWrapper from './sticky-search-and-switch-wrapper'
-import { getMarketplaceCollectionsAndPlugins } from './utils'
 
 type MarketplaceProps = {
   showInstallButton?: boolean
-  shouldExclude?: boolean
-  searchParams?: SearchParams
   pluginTypeSwitchClassName?: string
-  scrollContainerId?: string
-  showSearchParams?: boolean
+  /**
+   * Pass the search params from the request to prefetch data on the server
+   * and preserve the search params in the URL.
+   */
+  searchParams?: Promise<SearchParams>
 }
+
 const Marketplace = async ({
   showInstallButton = true,
-  shouldExclude,
-  searchParams,
   pluginTypeSwitchClassName,
-  scrollContainerId,
-  showSearchParams = true,
+  searchParams,
 }: MarketplaceProps) => {
-  let marketplaceCollections: MarketplaceCollection[] = []
-  let marketplaceCollectionPluginsMap: Record<string, Plugin[]> = {}
-  if (!shouldExclude) {
-    const marketplaceCollectionsAndPluginsData = await getMarketplaceCollectionsAndPlugins()
-    marketplaceCollections = marketplaceCollectionsAndPluginsData.marketplaceCollections
-    marketplaceCollectionPluginsMap = marketplaceCollectionsAndPluginsData.marketplaceCollectionPluginsMap
-  }
-
   return (
     <TanstackQueryInitializer>
-      <MarketplaceContextProvider
-        searchParams={searchParams}
-        shouldExclude={shouldExclude}
-        scrollContainerId={scrollContainerId}
-        showSearchParams={showSearchParams}
-      >
-        <Description />
-        <StickySearchAndSwitchWrapper
-          pluginTypeSwitchClassName={pluginTypeSwitchClassName}
-          showSearchParams={showSearchParams}
-        />
-        <ListWrapper
-          marketplaceCollections={marketplaceCollections}
-          marketplaceCollectionPluginsMap={marketplaceCollectionPluginsMap}
-          showInstallButton={showInstallButton}
-        />
-      </MarketplaceContextProvider>
+      <HydrateQueryClient searchParams={searchParams}>
+        <HydrateMarketplaceAtoms preserveSearchStateInQuery={!!searchParams}>
+          <Description />
+          <StickySearchAndSwitchWrapper
+            pluginTypeSwitchClassName={pluginTypeSwitchClassName}
+          />
+          <ListWrapper
+            showInstallButton={showInstallButton}
+          />
+        </HydrateMarketplaceAtoms>
+      </HydrateQueryClient>
     </TanstackQueryInitializer>
   )
 }

web/app/components/plugins/marketplace/list/index.spec.tsx

@@ -1,6 +1,6 @@
 import type { MarketplaceCollection, SearchParamsFromCollection } from '../types'
 import type { Plugin } from '@/app/components/plugins/types'
-import { fireEvent, render, screen, waitFor } from '@testing-library/react'
+import { fireEvent, render, screen } from '@testing-library/react'
 import { beforeEach, describe, expect, it, vi } from 'vitest'
 import { PluginCategoryEnum } from '@/app/components/plugins/types'
 import List from './index'
@@ -30,23 +30,27 @@ vi.mock('#i18n', () => ({
   useLocale: () => 'en-US',
 }))
 
-// Mock useMarketplaceContext with controllable values
-const mockContextValues = {
-  plugins: undefined as Plugin[] | undefined,
-  pluginsTotal: 0,
-  marketplaceCollectionsFromClient: undefined as MarketplaceCollection[] | undefined,
-  marketplaceCollectionPluginsMapFromClient: undefined as Record<string, Plugin[]> | undefined,
-  isLoading: false,
-  isSuccessCollections: false,
-  handleQueryPlugins: vi.fn(),
-  searchPluginText: '',
-  filterPluginTags: [] as string[],
-  page: 1,
-  handleMoreClick: vi.fn(),
-}
+// Mock marketplace state hooks with controllable values
+const { mockMarketplaceData, mockMoreClick } = vi.hoisted(() => {
+  return {
+    mockMarketplaceData: {
+      plugins: undefined as Plugin[] | undefined,
+      pluginsTotal: 0,
+      marketplaceCollections: undefined as MarketplaceCollection[] | undefined,
+      marketplaceCollectionPluginsMap: undefined as Record<string, Plugin[]> | undefined,
+      isLoading: false,
+      page: 1,
+    },
+    mockMoreClick: vi.fn(),
+  }
+})
 
-vi.mock('../context', () => ({
-  useMarketplaceContext: (selector: (v: typeof mockContextValues) => unknown) => selector(mockContextValues),
+vi.mock('../state', () => ({
+  useMarketplaceData: () => mockMarketplaceData,
+}))
+
+vi.mock('../atoms', () => ({
+  useMarketplaceMoreClick: () => mockMoreClick,
 }))
 
 // Mock useLocale context
@@ -578,7 +582,7 @@ describe('ListWithCollection', () => {
   // View More Button Tests
   // ================================
   describe('View More Button', () => {
-    it('should render View More button when collection is searchable and onMoreClick is provided', () => {
+    it('should render View More button when collection is searchable', () => {
       const collections = [createMockCollection({
         name: 'collection-0',
         searchable: true,
@@ -587,14 +591,12 @@ describe('ListWithCollection', () => {
       const pluginsMap: Record<string, Plugin[]> = {
         'collection-0': createMockPluginList(1),
       }
-      const onMoreClick = vi.fn()
 
       render(
         <ListWithCollection
           {...defaultProps}
           marketplaceCollections={collections}
           marketplaceCollectionPluginsMap={pluginsMap}
-          onMoreClick={onMoreClick}
         />,
       )
 
@@ -609,42 +611,19 @@ describe('ListWithCollection', () => {
       const pluginsMap: Record<string, Plugin[]> = {
         'collection-0': createMockPluginList(1),
       }
-      const onMoreClick = vi.fn()
 
       render(
         <ListWithCollection
           {...defaultProps}
           marketplaceCollections={collections}
           marketplaceCollectionPluginsMap={pluginsMap}
-          onMoreClick={onMoreClick}
         />,
       )
 
       expect(screen.queryByText('View More')).not.toBeInTheDocument()
     })
 
-    it('should not render View More button when onMoreClick is not provided', () => {
-      const collections = [createMockCollection({
-        name: 'collection-0',
-        searchable: true,
-      })]
-      const pluginsMap: Record<string, Plugin[]> = {
-        'collection-0': createMockPluginList(1),
-      }
-
-      render(
-        <ListWithCollection
-          {...defaultProps}
-          marketplaceCollections={collections}
-          marketplaceCollectionPluginsMap={pluginsMap}
-          onMoreClick={undefined}
-        />,
-      )
-
-      expect(screen.queryByText('View More')).not.toBeInTheDocument()
-    })
-
-    it('should call onMoreClick with search_params when View More is clicked', () => {
+    it('should call moreClick hook with search_params when View More is clicked', () => {
       const searchParams: SearchParamsFromCollection = { query: 'test-query', sort_by: 'install_count' }
       const collections = [createMockCollection({
         name: 'collection-0',
@@ -654,21 +633,19 @@ describe('ListWithCollection', () => {
       const pluginsMap: Record<string, Plugin[]> = {
         'collection-0': createMockPluginList(1),
       }
-      const onMoreClick = vi.fn()
 
       render(
         <ListWithCollection
           {...defaultProps}
           marketplaceCollections={collections}
           marketplaceCollectionPluginsMap={pluginsMap}
-          onMoreClick={onMoreClick}
         />,
       )
 
       fireEvent.click(screen.getByText('View More'))
 
-      expect(onMoreClick).toHaveBeenCalledTimes(1)
-      expect(onMoreClick).toHaveBeenCalledWith(searchParams)
+      expect(mockMoreClick).toHaveBeenCalledTimes(1)
+      expect(mockMoreClick).toHaveBeenCalledWith(searchParams)
     })
   })
 
@@ -802,24 +779,15 @@ describe('ListWithCollection', () => {
 // ListWrapper Component Tests
 // ================================
 describe('ListWrapper', () => {
-  const defaultProps = {
-    marketplaceCollections: [] as MarketplaceCollection[],
-    marketplaceCollectionPluginsMap: {} as Record<string, Plugin[]>,
-    showInstallButton: false,
-  }
-
   beforeEach(() => {
     vi.clearAllMocks()
-    // Reset context values
-    mockContextValues.plugins = undefined
-    mockContextValues.pluginsTotal = 0
-    mockContextValues.marketplaceCollectionsFromClient = undefined
-    mockContextValues.marketplaceCollectionPluginsMapFromClient = undefined
-    mockContextValues.isLoading = false
-    mockContextValues.isSuccessCollections = false
-    mockContextValues.searchPluginText = ''
-    mockContextValues.filterPluginTags = []
-    mockContextValues.page = 1
+    // Reset mock data
+    mockMarketplaceData.plugins = undefined
+    mockMarketplaceData.pluginsTotal = 0
+    mockMarketplaceData.marketplaceCollections = undefined
+    mockMarketplaceData.marketplaceCollectionPluginsMap = undefined
+    mockMarketplaceData.isLoading = false
+    mockMarketplaceData.page = 1
   })
 
   // ================================
@@ -827,32 +795,32 @@ describe('ListWrapper', () => {
   // ================================
   describe('Rendering', () => {
     it('should render without crashing', () => {
-      render(<ListWrapper {...defaultProps} />)
+      render(<ListWrapper />)
 
       expect(document.body).toBeInTheDocument()
     })
 
     it('should render with scrollbarGutter style', () => {
-      const { container } = render(<ListWrapper {...defaultProps} />)
+      const { container } = render(<ListWrapper />)
 
       const wrapper = container.firstChild as HTMLElement
       expect(wrapper).toHaveStyle({ scrollbarGutter: 'stable' })
     })
 
     it('should render Loading component when isLoading is true and page is 1', () => {
-      mockContextValues.isLoading = true
-      mockContextValues.page = 1
+      mockMarketplaceData.isLoading = true
+      mockMarketplaceData.page = 1
 
-      render(<ListWrapper {...defaultProps} />)
+      render(<ListWrapper />)
 
       expect(screen.getByTestId('loading-component')).toBeInTheDocument()
     })
 
     it('should not render Loading component when page > 1', () => {
-      mockContextValues.isLoading = true
-      mockContextValues.page = 2
+      mockMarketplaceData.isLoading = true
+      mockMarketplaceData.page = 2
 
-      render(<ListWrapper {...defaultProps} />)
+      render(<ListWrapper />)
 
       expect(screen.queryByTestId('loading-component')).not.toBeInTheDocument()
     })
@@ -863,26 +831,26 @@ describe('ListWrapper', () => {
   // ================================
   describe('Plugins Header', () => {
     it('should render plugins result count when plugins are present', () => {
-      mockContextValues.plugins = createMockPluginList(5)
-      mockContextValues.pluginsTotal = 5
+      mockMarketplaceData.plugins = createMockPluginList(5)
+      mockMarketplaceData.pluginsTotal = 5
 
-      render(<ListWrapper {...defaultProps} />)
+      render(<ListWrapper />)
 
       expect(screen.getByText('5 plugins found')).toBeInTheDocument()
     })
 
     it('should render SortDropdown when plugins are present', () => {
-      mockContextValues.plugins = createMockPluginList(1)
+      mockMarketplaceData.plugins = createMockPluginList(1)
 
-      render(<ListWrapper {...defaultProps} />)
+      render(<ListWrapper />)
 
       expect(screen.getByTestId('sort-dropdown')).toBeInTheDocument()
     })
 
     it('should not render plugins header when plugins is undefined', () => {
-      mockContextValues.plugins = undefined
+      mockMarketplaceData.plugins = undefined
 
-      render(<ListWrapper {...defaultProps} />)
+      render(<ListWrapper />)
 
       expect(screen.queryByTestId('sort-dropdown')).not.toBeInTheDocument()
     })
@@ -892,197 +860,60 @@ describe('ListWrapper', () => {
   // List Rendering Logic Tests
   // ================================
   describe('List Rendering Logic', () => {
-    it('should render List when not loading', () => {
-      mockContextValues.isLoading = false
-      const collections = createMockCollectionList(1)
-      const pluginsMap: Record<string, Plugin[]> = {
+    it('should render collections when not loading', () => {
+      mockMarketplaceData.isLoading = false
+      mockMarketplaceData.marketplaceCollections = createMockCollectionList(1)
+      mockMarketplaceData.marketplaceCollectionPluginsMap = {
         'collection-0': createMockPluginList(1),
       }
 
-      render(
-        <ListWrapper
-          {...defaultProps}
-          marketplaceCollections={collections}
-          marketplaceCollectionPluginsMap={pluginsMap}
-        />,
-      )
+      render(<ListWrapper />)
 
       expect(screen.getByText('Collection 0')).toBeInTheDocument()
     })
 
     it('should render List when loading but page > 1', () => {
-      mockContextValues.isLoading = true
-      mockContextValues.page = 2
-      const collections = createMockCollectionList(1)
-      const pluginsMap: Record<string, Plugin[]> = {
+      mockMarketplaceData.isLoading = true
+      mockMarketplaceData.page = 2
+      mockMarketplaceData.marketplaceCollections = createMockCollectionList(1)
+      mockMarketplaceData.marketplaceCollectionPluginsMap = {
         'collection-0': createMockPluginList(1),
       }
 
-      render(
-        <ListWrapper
-          {...defaultProps}
-          marketplaceCollections={collections}
-          marketplaceCollectionPluginsMap={pluginsMap}
-        />,
-      )
+      render(<ListWrapper />)
 
       expect(screen.getByText('Collection 0')).toBeInTheDocument()
     })
-
-    it('should use client collections when available', () => {
-      const serverCollections = createMockCollectionList(1)
-      serverCollections[0].label = { 'en-US': 'Server Collection' }
-      const clientCollections = createMockCollectionList(1)
-      clientCollections[0].label = { 'en-US': 'Client Collection' }
-
-      const serverPluginsMap: Record<string, Plugin[]> = {
-        'collection-0': createMockPluginList(1),
-      }
-      const clientPluginsMap: Record<string, Plugin[]> = {
-        'collection-0': createMockPluginList(1),
-      }
-
-      mockContextValues.marketplaceCollectionsFromClient = clientCollections
-      mockContextValues.marketplaceCollectionPluginsMapFromClient = clientPluginsMap
-
-      render(
-        <ListWrapper
-          {...defaultProps}
-          marketplaceCollections={serverCollections}
-          marketplaceCollectionPluginsMap={serverPluginsMap}
-        />,
-      )
-
-      expect(screen.getByText('Client Collection')).toBeInTheDocument()
-      expect(screen.queryByText('Server Collection')).not.toBeInTheDocument()
-    })
-
-    it('should use server collections when client collections are not available', () => {
-      const serverCollections = createMockCollectionList(1)
-      serverCollections[0].label = { 'en-US': 'Server Collection' }
-      const serverPluginsMap: Record<string, Plugin[]> = {
-        'collection-0': createMockPluginList(1),
-      }
-
-      mockContextValues.marketplaceCollectionsFromClient = undefined
-      mockContextValues.marketplaceCollectionPluginsMapFromClient = undefined
-
-      render(
-        <ListWrapper
-          {...defaultProps}
-          marketplaceCollections={serverCollections}
-          marketplaceCollectionPluginsMap={serverPluginsMap}
-        />,
-      )
-
-      expect(screen.getByText('Server Collection')).toBeInTheDocument()
-    })
   })
 
   // ================================
-  // Context Integration Tests
+  // Data Integration Tests
   // ================================
-  describe('Context Integration', () => {
-    it('should pass plugins from context to List', () => {
-      const plugins = createMockPluginList(2)
-      mockContextValues.plugins = plugins
+  describe('Data Integration', () => {
+    it('should pass plugins from state to List', () => {
+      mockMarketplaceData.plugins = createMockPluginList(2)
 
-      render(<ListWrapper {...defaultProps} />)
+      render(<ListWrapper />)
 
       expect(screen.getByTestId('card-plugin-0')).toBeInTheDocument()
       expect(screen.getByTestId('card-plugin-1')).toBeInTheDocument()
     })
 
-    it('should pass handleMoreClick from context to List', () => {
-      const mockHandleMoreClick = vi.fn()
-      mockContextValues.handleMoreClick = mockHandleMoreClick
-
-      const collections = [createMockCollection({
+    it('should show View More button and call moreClick hook', () => {
+      mockMarketplaceData.marketplaceCollections = [createMockCollection({
         name: 'collection-0',
         searchable: true,
         search_params: { query: 'test' },
       })]
-      const pluginsMap: Record<string, Plugin[]> = {
+      mockMarketplaceData.marketplaceCollectionPluginsMap = {
         'collection-0': createMockPluginList(1),
       }
 
-      render(
-        <ListWrapper
-          {...defaultProps}
-          marketplaceCollections={collections}
-          marketplaceCollectionPluginsMap={pluginsMap}
-        />,
-      )
+      render(<ListWrapper />)
 
       fireEvent.click(screen.getByText('View More'))
 
-      expect(mockHandleMoreClick).toHaveBeenCalled()
-    })
-  })
-
-  // ================================
-  // Effect Tests (handleQueryPlugins)
-  // ================================
-  describe('handleQueryPlugins Effect', () => {
-    it('should call handleQueryPlugins when conditions are met', async () => {
-      const mockHandleQueryPlugins = vi.fn()
-      mockContextValues.handleQueryPlugins = mockHandleQueryPlugins
-      mockContextValues.isSuccessCollections = true
-      mockContextValues.marketplaceCollectionsFromClient = undefined
-      mockContextValues.searchPluginText = ''
-      mockContextValues.filterPluginTags = []
-
-      render(<ListWrapper {...defaultProps} />)
-
-      await waitFor(() => {
-        expect(mockHandleQueryPlugins).toHaveBeenCalled()
-      })
-    })
-
-    it('should not call handleQueryPlugins when client collections exist', async () => {
-      const mockHandleQueryPlugins = vi.fn()
-      mockContextValues.handleQueryPlugins = mockHandleQueryPlugins
-      mockContextValues.isSuccessCollections = true
-      mockContextValues.marketplaceCollectionsFromClient = createMockCollectionList(1)
-      mockContextValues.searchPluginText = ''
-      mockContextValues.filterPluginTags = []
-
-      render(<ListWrapper {...defaultProps} />)
-
-      // Give time for effect to run
-      await waitFor(() => {
-        expect(mockHandleQueryPlugins).not.toHaveBeenCalled()
-      })
-    })
-
-    it('should not call handleQueryPlugins when search text exists', async () => {
-      const mockHandleQueryPlugins = vi.fn()
-      mockContextValues.handleQueryPlugins = mockHandleQueryPlugins
-      mockContextValues.isSuccessCollections = true
-      mockContextValues.marketplaceCollectionsFromClient = undefined
-      mockContextValues.searchPluginText = 'search text'
-      mockContextValues.filterPluginTags = []
-
-      render(<ListWrapper {...defaultProps} />)
-
-      await waitFor(() => {
-        expect(mockHandleQueryPlugins).not.toHaveBeenCalled()
-      })
-    })
-
-    it('should not call handleQueryPlugins when filter tags exist', async () => {
-      const mockHandleQueryPlugins = vi.fn()
-      mockContextValues.handleQueryPlugins = mockHandleQueryPlugins
-      mockContextValues.isSuccessCollections = true
-      mockContextValues.marketplaceCollectionsFromClient = undefined
-      mockContextValues.searchPluginText = ''
-      mockContextValues.filterPluginTags = ['tag1']
-
-      render(<ListWrapper {...defaultProps} />)
-
-      await waitFor(() => {
-        expect(mockHandleQueryPlugins).not.toHaveBeenCalled()
-      })
+      expect(mockMoreClick).toHaveBeenCalled()
     })
   })
 
@@ -1090,32 +921,32 @@ describe('ListWrapper', () => {
   // Edge Cases Tests
   // ================================
   describe('Edge Cases', () => {
-    it('should handle empty plugins array from context', () => {
-      mockContextValues.plugins = []
-      mockContextValues.pluginsTotal = 0
+    it('should handle empty plugins array', () => {
+      mockMarketplaceData.plugins = []
+      mockMarketplaceData.pluginsTotal = 0
 
-      render(<ListWrapper {...defaultProps} />)
+      render(<ListWrapper />)
 
       expect(screen.getByText('0 plugins found')).toBeInTheDocument()
       expect(screen.getByTestId('empty-component')).toBeInTheDocument()
     })
 
     it('should handle large pluginsTotal', () => {
-      mockContextValues.plugins = createMockPluginList(10)
-      mockContextValues.pluginsTotal = 10000
+      mockMarketplaceData.plugins = createMockPluginList(10)
+      mockMarketplaceData.pluginsTotal = 10000
 
-      render(<ListWrapper {...defaultProps} />)
+      render(<ListWrapper />)
 
       expect(screen.getByText('10000 plugins found')).toBeInTheDocument()
     })
 
     it('should handle both loading and has plugins', () => {
-      mockContextValues.isLoading = true
-      mockContextValues.page = 2
-      mockContextValues.plugins = createMockPluginList(5)
-      mockContextValues.pluginsTotal = 50
+      mockMarketplaceData.isLoading = true
+      mockMarketplaceData.page = 2
+      mockMarketplaceData.plugins = createMockPluginList(5)
+      mockMarketplaceData.pluginsTotal = 50
 
-      render(<ListWrapper {...defaultProps} />)
+      render(<ListWrapper />)
 
       // Should show plugins header and list
       expect(screen.getByText('50 plugins found')).toBeInTheDocument()
@@ -1428,106 +1259,72 @@ describe('CardWrapper (via List integration)', () => {
 describe('Combined Workflows', () => {
   beforeEach(() => {
     vi.clearAllMocks()
-    mockContextValues.plugins = undefined
-    mockContextValues.pluginsTotal = 0
-    mockContextValues.isLoading = false
-    mockContextValues.page = 1
-    mockContextValues.marketplaceCollectionsFromClient = undefined
-    mockContextValues.marketplaceCollectionPluginsMapFromClient = undefined
+    mockMarketplaceData.plugins = undefined
+    mockMarketplaceData.pluginsTotal = 0
+    mockMarketplaceData.isLoading = false
+    mockMarketplaceData.page = 1
+    mockMarketplaceData.marketplaceCollections = undefined
+    mockMarketplaceData.marketplaceCollectionPluginsMap = undefined
   })
 
   it('should transition from loading to showing collections', async () => {
-    mockContextValues.isLoading = true
-    mockContextValues.page = 1
+    mockMarketplaceData.isLoading = true
+    mockMarketplaceData.page = 1
 
-    const { rerender } = render(
-      <ListWrapper
-        marketplaceCollections={[]}
-        marketplaceCollectionPluginsMap={{}}
-      />,
-    )
+    const { rerender } = render(<ListWrapper />)
 
     expect(screen.getByTestId('loading-component')).toBeInTheDocument()
 
     // Simulate loading complete
-    mockContextValues.isLoading = false
-    const collections = createMockCollectionList(1)
-    const pluginsMap: Record<string, Plugin[]> = {
+    mockMarketplaceData.isLoading = false
+    mockMarketplaceData.marketplaceCollections = createMockCollectionList(1)
+    mockMarketplaceData.marketplaceCollectionPluginsMap = {
       'collection-0': createMockPluginList(1),
     }
-    mockContextValues.marketplaceCollectionsFromClient = collections
-    mockContextValues.marketplaceCollectionPluginsMapFromClient = pluginsMap
 
-    rerender(
-      <ListWrapper
-        marketplaceCollections={[]}
-        marketplaceCollectionPluginsMap={{}}
-      />,
-    )
+    rerender(<ListWrapper />)
 
     expect(screen.queryByTestId('loading-component')).not.toBeInTheDocument()
     expect(screen.getByText('Collection 0')).toBeInTheDocument()
   })
 
   it('should transition from collections to search results', async () => {
-    const collections = createMockCollectionList(1)
-    const pluginsMap: Record<string, Plugin[]> = {
+    mockMarketplaceData.marketplaceCollections = createMockCollectionList(1)
+    mockMarketplaceData.marketplaceCollectionPluginsMap = {
       'collection-0': createMockPluginList(1),
     }
-    mockContextValues.marketplaceCollectionsFromClient = collections
-    mockContextValues.marketplaceCollectionPluginsMapFromClient = pluginsMap
 
-    const { rerender } = render(
-      <ListWrapper
-        marketplaceCollections={[]}
-        marketplaceCollectionPluginsMap={{}}
-      />,
-    )
+    const { rerender } = render(<ListWrapper />)
 
     expect(screen.getByText('Collection 0')).toBeInTheDocument()
 
     // Simulate search results
-    mockContextValues.plugins = createMockPluginList(5)
-    mockContextValues.pluginsTotal = 5
+    mockMarketplaceData.plugins = createMockPluginList(5)
+    mockMarketplaceData.pluginsTotal = 5
 
-    rerender(
-      <ListWrapper
-        marketplaceCollections={[]}
-        marketplaceCollectionPluginsMap={{}}
-      />,
-    )
+    rerender(<ListWrapper />)
 
     expect(screen.queryByText('Collection 0')).not.toBeInTheDocument()
     expect(screen.getByText('5 plugins found')).toBeInTheDocument()
   })
 
   it('should handle empty search results', () => {
-    mockContextValues.plugins = []
-    mockContextValues.pluginsTotal = 0
+    mockMarketplaceData.plugins = []
+    mockMarketplaceData.pluginsTotal = 0
 
-    render(
-      <ListWrapper
-        marketplaceCollections={[]}
-        marketplaceCollectionPluginsMap={{}}
-      />,
-    )
+    render(<ListWrapper />)
 
     expect(screen.getByTestId('empty-component')).toBeInTheDocument()
     expect(screen.getByText('0 plugins found')).toBeInTheDocument()
   })
 
   it('should support pagination (page > 1)', () => {
-    mockContextValues.plugins = createMockPluginList(40)
-    mockContextValues.pluginsTotal = 80
-    mockContextValues.isLoading = true
-    mockContextValues.page = 2
+    mockMarketplaceData.plugins = createMockPluginList(40)
+    mockMarketplaceData.pluginsTotal = 80
+    mockMarketplaceData.isLoading = true
+    mockMarketplaceData.page = 2
 
-    render(
-      <ListWrapper
-        marketplaceCollections={[]}
-        marketplaceCollectionPluginsMap={{}}
-      />,
-    )
+    render(<ListWrapper />)
 
     // Should show existing results while loading more
     expect(screen.getByText('80 plugins found')).toBeInTheDocument()
@@ -1542,9 +1339,9 @@ describe('Combined Workflows', () => {
 describe('Accessibility', () => {
   beforeEach(() => {
     vi.clearAllMocks()
-    mockContextValues.plugins = undefined
-    mockContextValues.isLoading = false
-    mockContextValues.page = 1
+    mockMarketplaceData.plugins = undefined
+    mockMarketplaceData.isLoading = false
+    mockMarketplaceData.page = 1
   })
 
   it('should have semantic structure with collections', () => {
@@ -1573,13 +1370,11 @@ describe('Accessibility', () => {
     const pluginsMap: Record<string, Plugin[]> = {
       'collection-0': createMockPluginList(1),
     }
-    const onMoreClick = vi.fn()
 
     render(
       <ListWithCollection
         marketplaceCollections={collections}
         marketplaceCollectionPluginsMap={pluginsMap}
-        onMoreClick={onMoreClick}
       />,
     )
 

web/app/components/plugins/marketplace/list/index.tsx

@@ -13,7 +13,6 @@ type ListProps = {
   showInstallButton?: boolean
   cardContainerClassName?: string
   cardRender?: (plugin: Plugin) => React.JSX.Element | null
-  onMoreClick?: () => void
   emptyClassName?: string
 }
 const List = ({
@@ -23,7 +22,6 @@ const List = ({
   showInstallButton,
   cardContainerClassName,
   cardRender,
-  onMoreClick,
   emptyClassName,
 }: ListProps) => {
   return (
@@ -36,7 +34,6 @@ const List = ({
             showInstallButton={showInstallButton}
             cardContainerClassName={cardContainerClassName}
             cardRender={cardRender}
-            onMoreClick={onMoreClick}
           />
         )
       }

web/app/components/plugins/marketplace/list/list-with-collection.tsx

@@ -1,12 +1,12 @@
 'use client'
 
 import type { MarketplaceCollection } from '../types'
-import type { SearchParamsFromCollection } from '@/app/components/plugins/marketplace/types'
 import type { Plugin } from '@/app/components/plugins/types'
 import { useLocale, useTranslation } from '#i18n'
 import { RiArrowRightSLine } from '@remixicon/react'
 import { getLanguage } from '@/i18n-config/language'
 import { cn } from '@/utils/classnames'
+import { useMarketplaceMoreClick } from '../atoms'
 import CardWrapper from './card-wrapper'
 
 type ListWithCollectionProps = {
@@ -15,7 +15,6 @@ type ListWithCollectionProps = {
   showInstallButton?: boolean
   cardContainerClassName?: string
   cardRender?: (plugin: Plugin) => React.JSX.Element | null
-  onMoreClick?: (searchParams?: SearchParamsFromCollection) => void
 }
 const ListWithCollection = ({
   marketplaceCollections,
@@ -23,10 +22,10 @@ const ListWithCollection = ({
   showInstallButton,
   cardContainerClassName,
   cardRender,
-  onMoreClick,
 }: ListWithCollectionProps) => {
   const { t } = useTranslation()
   const locale = useLocale()
+  const onMoreClick = useMarketplaceMoreClick()
 
   return (
     <>
@@ -44,10 +43,10 @@ const ListWithCollection = ({
                 <div className="system-xs-regular text-text-tertiary">{collection.description[getLanguage(locale)]}</div>
               </div>
               {
-                collection.searchable && onMoreClick && (
+                collection.searchable && (
                   <div
                     className="system-xs-medium flex cursor-pointer items-center text-text-accent "
-                    onClick={() => onMoreClick?.(collection.search_params)}
+                    onClick={() => onMoreClick(collection.search_params)}
                   >
                     {t('marketplace.viewMore', { ns: 'plugin' })}
                     <RiArrowRightSLine className="h-4 w-4" />

web/app/components/plugins/marketplace/list/list-wrapper.tsx

@@ -1,46 +1,26 @@
 'use client'
-import type { Plugin } from '../../types'
-import type { MarketplaceCollection } from '../types'
 import { useTranslation } from '#i18n'
-import { useEffect } from 'react'
 import Loading from '@/app/components/base/loading'
-import { useMarketplaceContext } from '../context'
 import SortDropdown from '../sort-dropdown'
+import { useMarketplaceData } from '../state'
 import List from './index'
 
 type ListWrapperProps = {
-  marketplaceCollections: MarketplaceCollection[]
-  marketplaceCollectionPluginsMap: Record<string, Plugin[]>
   showInstallButton?: boolean
 }
 const ListWrapper = ({
-  marketplaceCollections,
-  marketplaceCollectionPluginsMap,
   showInstallButton,
 }: ListWrapperProps) => {
   const { t } = useTranslation()
-  const plugins = useMarketplaceContext(v => v.plugins)
-  const pluginsTotal = useMarketplaceContext(v => v.pluginsTotal)
-  const marketplaceCollectionsFromClient = useMarketplaceContext(v => v.marketplaceCollectionsFromClient)
-  const marketplaceCollectionPluginsMapFromClient = useMarketplaceContext(v => v.marketplaceCollectionPluginsMapFromClient)
-  const isLoading = useMarketplaceContext(v => v.isLoading)
-  const isSuccessCollections = useMarketplaceContext(v => v.isSuccessCollections)
-  const handleQueryPlugins = useMarketplaceContext(v => v.handleQueryPlugins)
-  const searchPluginText = useMarketplaceContext(v => v.searchPluginText)
-  const filterPluginTags = useMarketplaceContext(v => v.filterPluginTags)
-  const page = useMarketplaceContext(v => v.page)
-  const handleMoreClick = useMarketplaceContext(v => v.handleMoreClick)
 
-  useEffect(() => {
-    if (
-      !marketplaceCollectionsFromClient?.length
-      && isSuccessCollections
-      && !searchPluginText
-      && !filterPluginTags.length
-    ) {
-      handleQueryPlugins()
-    }
-  }, [handleQueryPlugins, marketplaceCollections, marketplaceCollectionsFromClient, isSuccessCollections, searchPluginText, filterPluginTags])
+  const {
+    plugins,
+    pluginsTotal,
+    marketplaceCollections,
+    marketplaceCollectionPluginsMap,
+    isLoading,
+    page,
+  } = useMarketplaceData()
 
   return (
     <div
@@ -66,11 +46,10 @@ const ListWrapper = ({
       {
         (!isLoading || page > 1) && (
           <List
-            marketplaceCollections={marketplaceCollectionsFromClient || marketplaceCollections}
-            marketplaceCollectionPluginsMap={marketplaceCollectionPluginsMapFromClient || marketplaceCollectionPluginsMap}
+            marketplaceCollections={marketplaceCollections || []}
+            marketplaceCollectionPluginsMap={marketplaceCollectionPluginsMap || {}}
             plugins={plugins}
             showInstallButton={showInstallButton}
-            onMoreClick={handleMoreClick}
           />
         )
       }

web/app/components/plugins/marketplace/plugin-type-switch.tsx

@@ -1,4 +1,5 @@
 'use client'
+import type { ActivePluginType } from './constants'
 import { useTranslation } from '#i18n'
 import {
   RiArchive2Line,
@@ -8,35 +9,27 @@ import {
   RiPuzzle2Line,
   RiSpeakAiLine,
 } from '@remixicon/react'
-import { useCallback, useEffect } from 'react'
+import { useSetAtom } from 'jotai'
 import { Trigger as TriggerIcon } from '@/app/components/base/icons/src/vender/plugin'
 import { cn } from '@/utils/classnames'
-import { PluginCategoryEnum } from '../types'
-import { useMarketplaceContext } from './context'
+import { searchModeAtom, useActivePluginType } from './atoms'
+import { PLUGIN_CATEGORY_WITH_COLLECTIONS, PLUGIN_TYPE_SEARCH_MAP } from './constants'
 
-export const PLUGIN_TYPE_SEARCH_MAP = {
-  all: 'all',
-  model: PluginCategoryEnum.model,
-  tool: PluginCategoryEnum.tool,
-  agent: PluginCategoryEnum.agent,
-  extension: PluginCategoryEnum.extension,
-  datasource: PluginCategoryEnum.datasource,
-  trigger: PluginCategoryEnum.trigger,
-  bundle: 'bundle',
-}
 type PluginTypeSwitchProps = {
   className?: string
-  showSearchParams?: boolean
 }
 const PluginTypeSwitch = ({
   className,
-  showSearchParams,
 }: PluginTypeSwitchProps) => {
   const { t } = useTranslation()
-  const activePluginType = useMarketplaceContext(s => s.activePluginType)
-  const handleActivePluginTypeChange = useMarketplaceContext(s => s.handleActivePluginTypeChange)
+  const [activePluginType, handleActivePluginTypeChange] = useActivePluginType()
+  const setSearchMode = useSetAtom(searchModeAtom)
 
-  const options = [
+  const options: Array<{
+    value: ActivePluginType
+    text: string
+    icon: React.ReactNode | null
+  }> = [
     {
       value: PLUGIN_TYPE_SEARCH_MAP.all,
       text: t('category.all', { ns: 'plugin' }),
@@ -79,23 +72,6 @@ const PluginTypeSwitch = ({
     },
   ]
 
-  const handlePopState = useCallback(() => {
-    if (!showSearchParams)
-      return
-    // nuqs handles popstate automatically
-    const url = new URL(window.location.href)
-    const category = url.searchParams.get('category') || PLUGIN_TYPE_SEARCH_MAP.all
-    handleActivePluginTypeChange(category)
-  }, [showSearchParams, handleActivePluginTypeChange])
-
-  useEffect(() => {
-    // nuqs manages popstate internally, but we keep this for URL sync
-    window.addEventListener('popstate', handlePopState)
-    return () => {
-      window.removeEventListener('popstate', handlePopState)
-    }
-  }, [handlePopState])
-
   return (
     <div className={cn(
       'flex shrink-0 items-center justify-center space-x-2 bg-background-body py-3',
@@ -112,6 +88,9 @@ const PluginTypeSwitch = ({
             )}
             onClick={() => {
               handleActivePluginTypeChange(option.value)
+              if (PLUGIN_CATEGORY_WITH_COLLECTIONS.has(option.value)) {
+                setSearchMode(null)
+              }
             }}
           >
             {option.icon}

web/app/components/plugins/marketplace/query.ts

@@ -0,0 +1,38 @@
+import type { CollectionsAndPluginsSearchParams, PluginsSearchParams } from './types'
+import { useInfiniteQuery, useQuery } from '@tanstack/react-query'
+import { getMarketplaceCollectionsAndPlugins, getMarketplacePlugins } from './utils'
+
+// TODO: Avoid manual maintenance of query keys and better service management,
+// https://github.com/langgenius/dify/issues/30342
+
+export const marketplaceKeys = {
+  all: ['marketplace'] as const,
+  collections: (params?: CollectionsAndPluginsSearchParams) => [...marketplaceKeys.all, 'collections', params] as const,
+  collectionPlugins: (collectionId: string, params?: CollectionsAndPluginsSearchParams) => [...marketplaceKeys.all, 'collectionPlugins', collectionId, params] as const,
+  plugins: (params?: PluginsSearchParams) => [...marketplaceKeys.all, 'plugins', params] as const,
+}
+
+export function useMarketplaceCollectionsAndPlugins(
+  collectionsParams: CollectionsAndPluginsSearchParams,
+) {
+  return useQuery({
+    queryKey: marketplaceKeys.collections(collectionsParams),
+    queryFn: ({ signal }) => getMarketplaceCollectionsAndPlugins(collectionsParams, { signal }),
+  })
+}
+
+export function useMarketplacePlugins(
+  queryParams: PluginsSearchParams | undefined,
+) {
+  return useInfiniteQuery({
+    queryKey: marketplaceKeys.plugins(queryParams),
+    queryFn: ({ pageParam = 1, signal }) => getMarketplacePlugins(queryParams, pageParam, signal),
+    getNextPageParam: (lastPage) => {
+      const nextPage = lastPage.page + 1
+      const loaded = lastPage.page * lastPage.pageSize
+      return loaded < (lastPage.total || 0) ? nextPage : undefined
+    },
+    initialPageParam: 1,
+    enabled: queryParams !== undefined,
+  })
+}

web/app/components/plugins/marketplace/search-box/index.spec.tsx

@@ -26,16 +26,19 @@ vi.mock('#i18n', () => ({
   }),
 }))
 
-// Mock useMarketplaceContext
-const mockContextValues = {
-  searchPluginText: '',
-  handleSearchPluginTextChange: vi.fn(),
-  filterPluginTags: [] as string[],
-  handleFilterPluginTagsChange: vi.fn(),
-}
+// Mock marketplace state hooks
+const { mockSearchPluginText, mockHandleSearchPluginTextChange, mockFilterPluginTags, mockHandleFilterPluginTagsChange } = vi.hoisted(() => {
+  return {
+    mockSearchPluginText: '',
+    mockHandleSearchPluginTextChange: vi.fn(),
+    mockFilterPluginTags: [] as string[],
+    mockHandleFilterPluginTagsChange: vi.fn(),
+  }
+})
 
-vi.mock('../context', () => ({
-  useMarketplaceContext: (selector: (v: typeof mockContextValues) => unknown) => selector(mockContextValues),
+vi.mock('../atoms', () => ({
+  useSearchPluginText: () => [mockSearchPluginText, mockHandleSearchPluginTextChange],
+  useFilterPluginTags: () => [mockFilterPluginTags, mockHandleFilterPluginTagsChange],
 }))
 
 // Mock useTags hook
@@ -430,9 +433,6 @@ describe('SearchBoxWrapper', () => {
   beforeEach(() => {
     vi.clearAllMocks()
     mockPortalOpenState = false
-    // Reset context values
-    mockContextValues.searchPluginText = ''
-    mockContextValues.filterPluginTags = []
   })
 
   describe('Rendering', () => {
@@ -456,28 +456,14 @@ describe('SearchBoxWrapper', () => {
     })
   })
 
-  describe('Context Integration', () => {
-    it('should use searchPluginText from context', () => {
-      mockContextValues.searchPluginText = 'context search'
-      render(<SearchBoxWrapper />)
-
-      expect(screen.getByDisplayValue('context search')).toBeInTheDocument()
-    })
-
+  describe('Hook Integration', () => {
     it('should call handleSearchPluginTextChange when search changes', () => {
       render(<SearchBoxWrapper />)
 
       const input = screen.getByRole('textbox')
       fireEvent.change(input, { target: { value: 'new search' } })
 
-      expect(mockContextValues.handleSearchPluginTextChange).toHaveBeenCalledWith('new search')
-    })
-
-    it('should use filterPluginTags from context', () => {
-      mockContextValues.filterPluginTags = ['agent', 'rag']
-      render(<SearchBoxWrapper />)
-
-      expect(screen.getByTestId('portal-elem')).toBeInTheDocument()
+      expect(mockHandleSearchPluginTextChange).toHaveBeenCalledWith('new search')
     })
   })
 

web/app/components/plugins/marketplace/search-box/search-box-wrapper.tsx

@@ -1,15 +1,13 @@
 'use client'
 
 import { useTranslation } from '#i18n'
-import { useMarketplaceContext } from '../context'
+import { useFilterPluginTags, useSearchPluginText } from '../atoms'
 import SearchBox from './index'
 
 const SearchBoxWrapper = () => {
   const { t } = useTranslation()
-  const searchPluginText = useMarketplaceContext(v => v.searchPluginText)
-  const handleSearchPluginTextChange = useMarketplaceContext(v => v.handleSearchPluginTextChange)
-  const filterPluginTags = useMarketplaceContext(v => v.filterPluginTags)
-  const handleFilterPluginTagsChange = useMarketplaceContext(v => v.handleFilterPluginTagsChange)
+  const [searchPluginText, handleSearchPluginTextChange] = useSearchPluginText()
+  const [filterPluginTags, handleFilterPluginTagsChange] = useFilterPluginTags()
 
   return (
     <SearchBox

web/app/components/plugins/marketplace/search-params.ts

@@ -0,0 +1,9 @@
+import type { ActivePluginType } from './constants'
+import { parseAsArrayOf, parseAsString, parseAsStringEnum } from 'nuqs/server'
+import { PLUGIN_TYPE_SEARCH_MAP } from './constants'
+
+export const marketplaceSearchParamsParsers = {
+  category: parseAsStringEnum<ActivePluginType>(Object.values(PLUGIN_TYPE_SEARCH_MAP) as ActivePluginType[]).withDefault('all').withOptions({ history: 'replace', clearOnDefault: false }),
+  q: parseAsString.withDefault('').withOptions({ history: 'replace' }),
+  tags: parseAsArrayOf(parseAsString).withDefault([]).withOptions({ history: 'replace' }),
+}

web/app/components/plugins/marketplace/sort-dropdown/index.spec.tsx

@@ -1,4 +1,3 @@
-import type { MarketplaceContextValue } from '../context'
 import { fireEvent, render, screen, within } from '@testing-library/react'
 import userEvent from '@testing-library/user-event'
 import { beforeEach, describe, expect, it, vi } from 'vitest'
@@ -28,18 +27,12 @@ vi.mock('#i18n', () => ({
   }),
 }))
 
-// Mock marketplace context with controllable values
-let mockSort = { sortBy: 'install_count', sortOrder: 'DESC' }
+// Mock marketplace atoms with controllable values
+let mockSort: { sortBy: string, sortOrder: string } = { sortBy: 'install_count', sortOrder: 'DESC' }
 const mockHandleSortChange = vi.fn()
 
-vi.mock('../context', () => ({
-  useMarketplaceContext: (selector: (value: MarketplaceContextValue) => unknown) => {
-    const contextValue = {
-      sort: mockSort,
-      handleSortChange: mockHandleSortChange,
-    } as unknown as MarketplaceContextValue
-    return selector(contextValue)
-  },
+vi.mock('../atoms', () => ({
+  useMarketplaceSort: () => [mockSort, mockHandleSortChange],
 }))
 
 // Mock portal component with controllable open state

web/app/components/plugins/marketplace/sort-dropdown/index.tsx

@@ -10,7 +10,7 @@ import {
   PortalToFollowElemContent,
   PortalToFollowElemTrigger,
 } from '@/app/components/base/portal-to-follow-elem'
-import { useMarketplaceContext } from '../context'
+import { useMarketplaceSort } from '../atoms'
 
 const SortDropdown = () => {
   const { t } = useTranslation()
@@ -36,8 +36,7 @@ const SortDropdown = () => {
       text: t('marketplace.sortOption.firstReleased', { ns: 'plugin' }),
     },
   ]
-  const sort = useMarketplaceContext(v => v.sort)
-  const handleSortChange = useMarketplaceContext(v => v.handleSortChange)
+  const [sort, handleSortChange] = useMarketplaceSort()
   const [open, setOpen] = useState(false)
   const selectedOption = options.find(option => option.value === sort.sortBy && option.order === sort.sortOrder) ?? options[0]
 

web/app/components/plugins/marketplace/state.ts

@@ -0,0 +1,54 @@
+import type { PluginsSearchParams } from './types'
+import { useDebounce } from 'ahooks'
+import { useCallback, useMemo } from 'react'
+import { useActivePluginType, useFilterPluginTags, useMarketplaceSearchMode, useMarketplaceSortValue, useSearchPluginText } from './atoms'
+import { PLUGIN_TYPE_SEARCH_MAP } from './constants'
+import { useMarketplaceContainerScroll } from './hooks'
+import { useMarketplaceCollectionsAndPlugins, useMarketplacePlugins } from './query'
+import { getCollectionsParams, getMarketplaceListFilterType } from './utils'
+
+export function useMarketplaceData() {
+  const [searchPluginTextOriginal] = useSearchPluginText()
+  const searchPluginText = useDebounce(searchPluginTextOriginal, { wait: 500 })
+  const [filterPluginTags] = useFilterPluginTags()
+  const [activePluginType] = useActivePluginType()
+
+  const collectionsQuery = useMarketplaceCollectionsAndPlugins(
+    getCollectionsParams(activePluginType),
+  )
+
+  const sort = useMarketplaceSortValue()
+  const isSearchMode = useMarketplaceSearchMode()
+  const queryParams = useMemo((): PluginsSearchParams | undefined => {
+    if (!isSearchMode)
+      return undefined
+    return {
+      query: searchPluginText,
+      category: activePluginType === PLUGIN_TYPE_SEARCH_MAP.all ? undefined : activePluginType,
+      tags: filterPluginTags,
+      sortBy: sort.sortBy,
+      sortOrder: sort.sortOrder,
+      type: getMarketplaceListFilterType(activePluginType),
+    }
+  }, [isSearchMode, searchPluginText, activePluginType, filterPluginTags, sort])
+
+  const pluginsQuery = useMarketplacePlugins(queryParams)
+  const { hasNextPage, fetchNextPage, isFetching } = pluginsQuery
+
+  const handlePageChange = useCallback(() => {
+    if (hasNextPage && !isFetching)
+      fetchNextPage()
+  }, [fetchNextPage, hasNextPage, isFetching])
+
+  // Scroll pagination
+  useMarketplaceContainerScroll(handlePageChange)
+
+  return {
+    marketplaceCollections: collectionsQuery.data?.marketplaceCollections,
+    marketplaceCollectionPluginsMap: collectionsQuery.data?.marketplaceCollectionPluginsMap,
+    plugins: pluginsQuery.data?.pages.flatMap(page => page.plugins),
+    pluginsTotal: pluginsQuery.data?.pages[0]?.total,
+    page: pluginsQuery.data?.pages.length || 1,
+    isLoading: collectionsQuery.isLoading || pluginsQuery.isLoading,
+  }
+}

web/app/components/plugins/marketplace/sticky-search-and-switch-wrapper.tsx

@@ -6,12 +6,10 @@ import SearchBoxWrapper from './search-box/search-box-wrapper'
 
 type StickySearchAndSwitchWrapperProps = {
   pluginTypeSwitchClassName?: string
-  showSearchParams?: boolean
 }
 
 const StickySearchAndSwitchWrapper = ({
   pluginTypeSwitchClassName,
-  showSearchParams,
 }: StickySearchAndSwitchWrapperProps) => {
   const hasCustomTopClass = pluginTypeSwitchClassName?.includes('top-')
 
@@ -24,9 +22,7 @@ const StickySearchAndSwitchWrapper = ({
       )}
     >
       <SearchBoxWrapper />
-      <PluginTypeSwitch
-        showSearchParams={showSearchParams}
-      />
+      <PluginTypeSwitch />
     </div>
   )
 }

web/app/components/plugins/marketplace/utils.ts

@@ -1,16 +1,19 @@
+import type { ActivePluginType } from './constants'
 import type {
   CollectionsAndPluginsSearchParams,
   MarketplaceCollection,
+  PluginsSearchParams,
 } from '@/app/components/plugins/marketplace/types'
-import type { Plugin } from '@/app/components/plugins/types'
+import type { Plugin, PluginsFromMarketplaceResponse } from '@/app/components/plugins/types'
 import { PluginCategoryEnum } from '@/app/components/plugins/types'
 import {
   APP_VERSION,
   IS_MARKETPLACE,
   MARKETPLACE_API_PREFIX,
 } from '@/config'
+import { postMarketplace } from '@/service/base'
 import { getMarketplaceUrl } from '@/utils/var'
-import { PLUGIN_TYPE_SEARCH_MAP } from './plugin-type-switch'
+import { PLUGIN_TYPE_SEARCH_MAP } from './constants'
 
 type MarketplaceFetchOptions = {
   signal?: AbortSignal
@@ -26,12 +29,13 @@ export const getPluginIconInMarketplace = (plugin: Plugin) => {
   return `${MARKETPLACE_API_PREFIX}/plugins/${plugin.org}/${plugin.name}/icon`
 }
 
-export const getFormattedPlugin = (bundle: any) => {
+export const getFormattedPlugin = (bundle: Plugin): Plugin => {
   if (bundle.type === 'bundle') {
     return {
       ...bundle,
       icon: getPluginIconInMarketplace(bundle),
       brief: bundle.description,
+      // @ts-expect-error I do not have enough information
       label: bundle.labels,
     }
   }
@@ -129,6 +133,64 @@ export const getMarketplaceCollectionsAndPlugins = async (
   }
 }
 
+export const getMarketplacePlugins = async (
+  queryParams: PluginsSearchParams | undefined,
+  pageParam: number,
+  signal?: AbortSignal,
+) => {
+  if (!queryParams) {
+    return {
+      plugins: [] as Plugin[],
+      total: 0,
+      page: 1,
+      pageSize: 40,
+    }
+  }
+
+  const {
+    query,
+    sortBy,
+    sortOrder,
+    category,
+    tags,
+    type,
+    pageSize = 40,
+  } = queryParams
+  const pluginOrBundle = type === 'bundle' ? 'bundles' : 'plugins'
+
+  try {
+    const res = await postMarketplace<{ data: PluginsFromMarketplaceResponse }>(`/${pluginOrBundle}/search/advanced`, {
+      body: {
+        page: pageParam,
+        page_size: pageSize,
+        query,
+        sort_by: sortBy,
+        sort_order: sortOrder,
+        category: category !== 'all' ? category : '',
+        tags,
+        type,
+      },
+      signal,
+    })
+    const resPlugins = res.data.bundles || res.data.plugins || []
+
+    return {
+      plugins: resPlugins.map(plugin => getFormattedPlugin(plugin)),
+      total: res.data.total,
+      page: pageParam,
+      pageSize,
+    }
+  }
+  catch {
+    return {
+      plugins: [],
+      total: 0,
+      page: pageParam,
+      pageSize,
+    }
+  }
+}
+
 export const getMarketplaceListCondition = (pluginType: string) => {
   if ([PluginCategoryEnum.tool, PluginCategoryEnum.agent, PluginCategoryEnum.model, PluginCategoryEnum.datasource, PluginCategoryEnum.trigger].includes(pluginType as PluginCategoryEnum))
     return `category=${pluginType}`
@@ -142,7 +204,7 @@ export const getMarketplaceListCondition = (pluginType: string) => {
   return ''
 }
 
-export const getMarketplaceListFilterType = (category: string) => {
+export const getMarketplaceListFilterType = (category: ActivePluginType) => {
   if (category === PLUGIN_TYPE_SEARCH_MAP.all)
     return undefined
 
@@ -151,3 +213,14 @@ export const getMarketplaceListFilterType = (category: string) => {
 
   return 'plugin'
 }
+
+export function getCollectionsParams(category: ActivePluginType): CollectionsAndPluginsSearchParams {
+  if (category === PLUGIN_TYPE_SEARCH_MAP.all) {
+    return {}
+  }
+  return {
+    category,
+    condition: getMarketplaceListCondition(category),
+    type: getMarketplaceListFilterType(category),
+  }
+}

web/app/components/plugins/plugin-page/index.tsx

@@ -27,7 +27,7 @@ import { cn } from '@/utils/classnames'
 import { PLUGIN_PAGE_TABS_MAP } from '../hooks'
 import InstallFromLocalPackage from '../install-plugin/install-from-local-package'
 import InstallFromMarketplace from '../install-plugin/install-from-marketplace'
-import { PLUGIN_TYPE_SEARCH_MAP } from '../marketplace/plugin-type-switch'
+import { PLUGIN_TYPE_SEARCH_MAP } from '../marketplace/constants'
 import {
   PluginPageContextProvider,
   usePluginPageContext,

web/app/components/plugins/reference-setting-modal/auto-update-setting/index.spec.tsx

@@ -262,7 +262,7 @@ vi.mock('@/app/components/base/icons/src/vender/other', () => ({
 }))
 
 // Mock PLUGIN_TYPE_SEARCH_MAP
-vi.mock('../../marketplace/plugin-type-switch', () => ({
+vi.mock('../../marketplace/constants', () => ({
   PLUGIN_TYPE_SEARCH_MAP: {
     all: 'all',
     model: 'model',

web/app/components/plugins/reference-setting-modal/auto-update-setting/tool-picker.tsx

@@ -1,5 +1,6 @@
 'use client'
 import type { FC } from 'react'
+import type { ActivePluginType } from '../../marketplace/constants'
 import * as React from 'react'
 import { useCallback, useMemo, useState } from 'react'
 import { useTranslation } from 'react-i18next'
@@ -12,7 +13,7 @@ import {
 import SearchBox from '@/app/components/plugins/marketplace/search-box'
 import { useInstalledPluginList } from '@/service/use-plugins'
 import { cn } from '@/utils/classnames'
-import { PLUGIN_TYPE_SEARCH_MAP } from '../../marketplace/plugin-type-switch'
+import { PLUGIN_TYPE_SEARCH_MAP } from '../../marketplace/constants'
 import { PluginSource } from '../../types'
 import NoDataPlaceholder from './no-data-placeholder'
 import ToolItem from './tool-item'
@@ -73,7 +74,7 @@ const ToolPicker: FC<Props> = ({
     },
   ]
 
-  const [pluginType, setPluginType] = useState(PLUGIN_TYPE_SEARCH_MAP.all)
+  const [pluginType, setPluginType] = useState<ActivePluginType>(PLUGIN_TYPE_SEARCH_MAP.all)
   const [query, setQuery] = useState('')
   const [tags, setTags] = useState<string[]>([])
   const { data, isLoading } = useInstalledPluginList()

web/app/components/share/text-generation/run-once/index.tsx

@@ -195,7 +195,7 @@ const RunOnce: FC<IRunOnceProps> = ({
                         noWrapper
                         className="bg h-[80px] overflow-y-auto rounded-[10px] bg-components-input-bg-normal p-1"
                         placeholder={
-                          <div className="whitespace-pre">{item.json_schema}</div>
+                          <div className="whitespace-pre">{typeof item.json_schema === 'string' ? item.json_schema : JSON.stringify(item.json_schema || '', null, 2)}</div>
                         }
                       />
                     )}

web/app/components/workflow/nodes/_base/components/before-run-form/form-item.tsx

@@ -48,6 +48,12 @@ const FormItem: FC<Props> = ({
   const { t } = useTranslation()
   const { type } = payload
   const fileSettings = useHooksStore(s => s.configsMap?.fileSettings)
+  const jsonSchemaPlaceholder = React.useMemo(() => {
+    const schema = (payload as any)?.json_schema
+    if (!schema)
+      return ''
+    return typeof schema === 'string' ? schema : JSON.stringify(schema, null, 2)
+  }, [payload])
 
   const handleArrayItemChange = useCallback((index: number) => {
     return (newValue: any) => {
@@ -211,7 +217,7 @@ const FormItem: FC<Props> = ({
             noWrapper
             className="bg h-[80px] overflow-y-auto rounded-[10px] bg-components-input-bg-normal p-1"
             placeholder={
-              <div className="whitespace-pre">{payload.json_schema}</div>
+              <div className="whitespace-pre">{jsonSchemaPlaceholder}</div>
             }
           />
         )}

web/app/components/workflow/nodes/_base/components/variable/utils.ts

@@ -353,7 +353,7 @@ const formatItem = (
         try {
           if (type === VarType.object && v.json_schema) {
             varRes.children = {
-              schema: JSON.parse(v.json_schema),
+              schema: typeof v.json_schema === 'string' ? JSON.parse(v.json_schema) : v.json_schema,
             }
           }
         }

web/app/components/workflow/types.ts

@@ -226,7 +226,7 @@ export type InputVar = {
   getVarValueFromDependent?: boolean
   hide?: boolean
   isFileItem?: boolean
-  json_schema?: string // for jsonObject type
+  json_schema?: string | Record<string, any> // for jsonObject type
 } & Partial<UploadFileSetting>
 
 export type ModelConfig = {

web/context/query-client-server.ts

@@ -0,0 +1,16 @@
+import { QueryClient } from '@tanstack/react-query'
+import { cache } from 'react'
+
+const STALE_TIME = 1000 * 60 * 30 // 30 minutes
+
+export function makeQueryClient() {
+  return new QueryClient({
+    defaultOptions: {
+      queries: {
+        staleTime: STALE_TIME,
+      },
+    },
+  })
+}
+
+export const getQueryClientServer = cache(makeQueryClient)

web/context/query-client.tsx

@@ -1,23 +1,27 @@
 'use client'
 
+import type { QueryClient } from '@tanstack/react-query'
 import type { FC, PropsWithChildren } from 'react'
-import { QueryClient, QueryClientProvider } from '@tanstack/react-query'
+import { QueryClientProvider } from '@tanstack/react-query'
+import { useState } from 'react'
 import { TanStackDevtoolsLoader } from '@/app/components/devtools/tanstack/loader'
+import { makeQueryClient } from './query-client-server'
 
-const STALE_TIME = 1000 * 60 * 30 // 30 minutes
+let browserQueryClient: QueryClient | undefined
 
-const client = new QueryClient({
-  defaultOptions: {
-    queries: {
-      staleTime: STALE_TIME,
-    },
-  },
-})
+function getQueryClient() {
+  if (typeof window === 'undefined') {
+    return makeQueryClient()
+  }
+  if (!browserQueryClient)
+    browserQueryClient = makeQueryClient()
+  return browserQueryClient
+}
 
-export const TanstackQueryInitializer: FC<PropsWithChildren> = (props) => {
-  const { children } = props
+export const TanstackQueryInitializer: FC<PropsWithChildren> = ({ children }) => {
+  const [queryClient] = useState(getQueryClient)
   return (
-    <QueryClientProvider client={client}>
+    <QueryClientProvider client={queryClient}>
       {children}
       <TanStackDevtoolsLoader />
     </QueryClientProvider>

web/hooks/use-query-params.spec.tsx

@@ -8,7 +8,6 @@ import {
   PRICING_MODAL_QUERY_PARAM,
   PRICING_MODAL_QUERY_VALUE,
   useAccountSettingModal,
-  useMarketplaceFilters,
   usePluginInstallation,
   usePricingModal,
 } from './use-query-params'
@@ -302,174 +301,6 @@ describe('useQueryParams hooks', () => {
     })
   })
 
-  // Marketplace filters query behavior.
-  describe('useMarketplaceFilters', () => {
-    it('should return default filters when query params are missing', () => {
-      // Arrange
-      const { result } = renderWithAdapter(() => useMarketplaceFilters())
-
-      // Act
-      const [filters] = result.current
-
-      // Assert
-      expect(filters.q).toBe('')
-      expect(filters.category).toBe('all')
-      expect(filters.tags).toEqual([])
-    })
-
-    it('should parse filters when query params are present', () => {
-      // Arrange
-      const { result } = renderWithAdapter(
-        () => useMarketplaceFilters(),
-        '?q=prompt&category=tool&tags=ai,ml',
-      )
-
-      // Act
-      const [filters] = result.current
-
-      // Assert
-      expect(filters.q).toBe('prompt')
-      expect(filters.category).toBe('tool')
-      expect(filters.tags).toEqual(['ai', 'ml'])
-    })
-
-    it('should treat empty tags param as empty array', () => {
-      // Arrange
-      const { result } = renderWithAdapter(
-        () => useMarketplaceFilters(),
-        '?tags=',
-      )
-
-      // Act
-      const [filters] = result.current
-
-      // Assert
-      expect(filters.tags).toEqual([])
-    })
-
-    it('should preserve other filters when updating a single field', async () => {
-      // Arrange
-      const { result } = renderWithAdapter(
-        () => useMarketplaceFilters(),
-        '?category=tool&tags=ai,ml',
-      )
-
-      // Act
-      act(() => {
-        result.current[1]({ q: 'search' })
-      })
-
-      // Assert
-      await waitFor(() => expect(result.current[0].q).toBe('search'))
-      expect(result.current[0].category).toBe('tool')
-      expect(result.current[0].tags).toEqual(['ai', 'ml'])
-    })
-
-    it('should clear q param when q is empty', async () => {
-      // Arrange
-      const { result, onUrlUpdate } = renderWithAdapter(
-        () => useMarketplaceFilters(),
-        '?q=search',
-      )
-
-      // Act
-      act(() => {
-        result.current[1]({ q: '' })
-      })
-
-      // Assert
-      await waitFor(() => expect(onUrlUpdate).toHaveBeenCalled())
-      const update = onUrlUpdate.mock.calls[onUrlUpdate.mock.calls.length - 1][0]
-      expect(update.searchParams.has('q')).toBe(false)
-    })
-
-    it('should serialize tags as comma-separated values', async () => {
-      // Arrange
-      const { result, onUrlUpdate } = renderWithAdapter(() => useMarketplaceFilters())
-
-      // Act
-      act(() => {
-        result.current[1]({ tags: ['ai', 'ml'] })
-      })
-
-      // Assert
-      await waitFor(() => expect(onUrlUpdate).toHaveBeenCalled())
-      const update = onUrlUpdate.mock.calls[onUrlUpdate.mock.calls.length - 1][0]
-      expect(update.searchParams.get('tags')).toBe('ai,ml')
-    })
-
-    it('should remove tags param when list is empty', async () => {
-      // Arrange
-      const { result, onUrlUpdate } = renderWithAdapter(
-        () => useMarketplaceFilters(),
-        '?tags=ai,ml',
-      )
-
-      // Act
-      act(() => {
-        result.current[1]({ tags: [] })
-      })
-
-      // Assert
-      await waitFor(() => expect(onUrlUpdate).toHaveBeenCalled())
-      const update = onUrlUpdate.mock.calls[onUrlUpdate.mock.calls.length - 1][0]
-      expect(update.searchParams.has('tags')).toBe(false)
-    })
-
-    it('should keep category in the URL when set to default', async () => {
-      // Arrange
-      const { result, onUrlUpdate } = renderWithAdapter(
-        () => useMarketplaceFilters(),
-        '?category=tool',
-      )
-
-      // Act
-      act(() => {
-        result.current[1]({ category: 'all' })
-      })
-
-      // Assert
-      await waitFor(() => expect(onUrlUpdate).toHaveBeenCalled())
-      const update = onUrlUpdate.mock.calls[onUrlUpdate.mock.calls.length - 1][0]
-      expect(update.searchParams.get('category')).toBe('all')
-    })
-
-    it('should clear all marketplace filters when set to null', async () => {
-      // Arrange
-      const { result, onUrlUpdate } = renderWithAdapter(
-        () => useMarketplaceFilters(),
-        '?q=search&category=tool&tags=ai,ml',
-      )
-
-      // Act
-      act(() => {
-        result.current[1](null)
-      })
-
-      // Assert
-      await waitFor(() => expect(onUrlUpdate).toHaveBeenCalled())
-      const update = onUrlUpdate.mock.calls[onUrlUpdate.mock.calls.length - 1][0]
-      expect(update.searchParams.has('q')).toBe(false)
-      expect(update.searchParams.has('category')).toBe(false)
-      expect(update.searchParams.has('tags')).toBe(false)
-    })
-
-    it('should use replace history when updating filters', async () => {
-      // Arrange
-      const { result, onUrlUpdate } = renderWithAdapter(() => useMarketplaceFilters())
-
-      // Act
-      act(() => {
-        result.current[1]({ q: 'search' })
-      })
-
-      // Assert
-      await waitFor(() => expect(onUrlUpdate).toHaveBeenCalled())
-      const update = onUrlUpdate.mock.calls[onUrlUpdate.mock.calls.length - 1][0]
-      expect(update.options.history).toBe('replace')
-    })
-  })
-
   // Plugin installation query behavior.
   describe('usePluginInstallation', () => {
     it('should parse package ids from JSON arrays', () => {

web/hooks/use-query-params.ts

@@ -15,7 +15,6 @@
 
 import {
   createParser,
-  parseAsArrayOf,
   parseAsString,
   useQueryState,
   useQueryStates,
@@ -93,39 +92,6 @@ export function useAccountSettingModal<T extends string = string>() {
   return [{ isOpen, payload: currentTab }, setState] as const
 }
 
-/**
- * Marketplace Search Query Parameters
- */
-export type MarketplaceFilters = {
-  q: string // search query
-  category: string // plugin category
-  tags: string[] // array of tags
-}
-
-/**
- * Hook to manage marketplace search/filter state via URL
- * Provides atomic updates - all params update together
- *
- * @example
- * const [filters, setFilters] = useMarketplaceFilters()
- * setFilters({ q: 'search', category: 'tool', tags: ['ai'] }) // Updates all at once
- * setFilters({ q: '' }) // Only updates q, keeps others
- * setFilters(null) // Clears all marketplace params
- */
-export function useMarketplaceFilters() {
-  return useQueryStates(
-    {
-      q: parseAsString.withDefault(''),
-      category: parseAsString.withDefault('all').withOptions({ clearOnDefault: false }),
-      tags: parseAsArrayOf(parseAsString).withDefault([]),
-    },
-    {
-      // Update URL without pushing to history (replaceState behavior)
-      history: 'replace',
-    },
-  )
-}
-
 /**
  * Plugin Installation Query Parameters
  */

web/models/debug.ts

@@ -62,7 +62,7 @@ export type PromptVariable = {
   icon?: string
   icon_background?: string
   hide?: boolean // used in frontend to hide variable
-  json_schema?: string
+  json_schema?: string | Record<string, any>
 }
 
 export type CompletionParams = {

web/service/use-apps.ts

@@ -10,13 +10,14 @@ import type {
   AppVoicesListResponse,
   WorkflowDailyConversationsResponse,
 } from '@/models/app'
-import type { App, AppModeEnum } from '@/types/app'
+import type { App } from '@/types/app'
 import {
   keepPreviousData,
   useInfiniteQuery,
   useQuery,
   useQueryClient,
 } from '@tanstack/react-query'
+import { AppModeEnum } from '@/types/app'
 import { get, post } from './base'
 import { useInvalid } from './use-base'
 
@@ -36,6 +37,16 @@ type DateRangeParams = {
   end?: string
 }
 
+// Allowed app modes for filtering; defined at module scope to avoid re-creating on every call
+const allowedModes = new Set<AppModeEnum | 'all'>([
+  'all',
+  AppModeEnum.WORKFLOW,
+  AppModeEnum.ADVANCED_CHAT,
+  AppModeEnum.CHAT,
+  AppModeEnum.AGENT_CHAT,
+  AppModeEnum.COMPLETION,
+])
+
 const normalizeAppListParams = (params: AppListParams) => {
   const {
     page = 1,
@@ -46,11 +57,13 @@ const normalizeAppListParams = (params: AppListParams) => {
     is_created_by_me,
   } = params
 
+  const safeMode = allowedModes.has((mode as any)) ? mode : undefined
+
   return {
     page,
     limit,
     name,
-    ...(mode && mode !== 'all' ? { mode } : {}),
+    ...(safeMode && safeMode !== 'all' ? { mode: safeMode } : {}),
     ...(tag_ids?.length ? { tag_ids } : {}),
     ...(is_created_by_me ? { is_created_by_me } : {}),
   }

web/service/workflow-payload.ts

@@ -66,7 +66,30 @@ export const sanitizeWorkflowDraftPayload = (params: WorkflowDraftSyncParams): W
   if (!graph?.nodes?.length)
     return params
 
-  const sanitizedNodes = graph.nodes.map(node => sanitizeTriggerPluginNode(node as Node<TriggerPluginNodePayload>))
+  const sanitizedNodes = graph.nodes.map((node) => {
+    // First sanitize known node types (TriggerPlugin)
+    const n = sanitizeTriggerPluginNode(node as Node<TriggerPluginNodePayload>) as Node<any>
+
+    // Normalize Start node variable json_schema: ensure dict, not string
+    if ((n.data as any)?.type === BlockEnum.Start && Array.isArray((n.data as any).variables)) {
+      const next = { ...n, data: { ...n.data } }
+      next.data.variables = (n.data as any).variables.map((v: any) => {
+        if (v && v.type === 'json_object' && typeof v.json_schema === 'string') {
+          try {
+            const obj = JSON.parse(v.json_schema)
+            return { ...v, json_schema: obj }
+          }
+          catch {
+            return v
+          }
+        }
+        return v
+      })
+      return next
+    }
+
+    return n
+  })
 
   return {
     ...params,
@@ -126,7 +149,25 @@ export const hydrateWorkflowDraftResponse = (draft: FetchWorkflowDraftResponse):
           if (node.data)
             removeTempProperties(node.data as Record<string, unknown>)
 
-          return hydrateTriggerPluginNode(node)
+          let n = hydrateTriggerPluginNode(node)
+          // Normalize Start node variable json_schema to object when loading
+          if ((n.data as any)?.type === BlockEnum.Start && Array.isArray((n.data as any).variables)) {
+            const next = { ...n, data: { ...n.data } } as Node<any>
+            next.data.variables = (n.data as any).variables.map((v: any) => {
+              if (v && v.type === 'json_object' && typeof v.json_schema === 'string') {
+                try {
+                  const obj = JSON.parse(v.json_schema)
+                  return { ...v, json_schema: obj }
+                }
+                catch {
+                  return v
+                }
+              }
+              return v
+            })
+            n = next
+          }
+          return n
         })
     }
 

web/service/workflow.ts

@@ -9,6 +9,7 @@ import type {
 } from '@/types/workflow'
 import { get, post } from './base'
 import { getFlowPrefix } from './utils'
+import { sanitizeWorkflowDraftPayload } from './workflow-payload'
 
 export const fetchWorkflowDraft = (url: string) => {
   return get(url, {}, { silent: true }) as Promise<FetchWorkflowDraftResponse>
@@ -18,7 +19,8 @@ export const syncWorkflowDraft = ({ url, params }: {
   url: string
   params: Pick<FetchWorkflowDraftResponse, 'graph' | 'features' | 'environment_variables' | 'conversation_variables'>
 }) => {
-  return post<CommonResponse & { updated_at: number, hash: string }>(url, { body: params }, { silent: true })
+  const sanitized = sanitizeWorkflowDraftPayload(params)
+  return post<CommonResponse & { updated_at: number, hash: string }>(url, { body: sanitized }, { silent: true })
 }
 
 export const fetchNodesDefaultConfigs = (url: string) => {