feat(multimodal): support document image extraction and inline vision processing

Added document image extraction capability for PDF and DOCX files, including page/index metadata and storage integration. Extended `process_files` with `document_image_recognition` flag to conditionally enable vision-based image processing when model supports it. Updated knowledge repository and workflow node logic to enforce status=1 checks. Added PyMuPDF dependency.

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

@@ -1,12 +1,15 @@
 import logging
+import uuid
 from typing import Any
 
+from app.core.config import settings
 from app.core.workflow.engine.state_manager import WorkflowState
 from app.core.workflow.engine.variable_pool import VariablePool
 from app.core.workflow.nodes.base_node import BaseNode
 from app.core.workflow.nodes.document_extractor.config import DocExtractorNodeConfig
 from app.core.workflow.variable.base_variable import VariableType, FileObject
 from app.db import get_db_read
+from app.models.file_metadata_model import FileMetadata
 from app.schemas.app_schema import FileInput, FileType, TransferMethod
 
 logger = logging.getLogger(__name__)
@@ -15,7 +18,6 @@ logger = logging.getLogger(__name__)
 def _file_object_to_file_input(f: FileObject) -> FileInput:
     """Convert workflow FileObject to multimodal FileInput."""
     file_type = f.origin_file_type or ""
-    # Prefer mime_type for more accurate type detection
     if not file_type and f.mime_type:
         file_type = f.mime_type
     resolved_type = FileType.trans(f.type) if isinstance(f.type, str) else f.type
@@ -51,21 +53,68 @@ def _normalise_files(val: Any) -> list[FileObject]:
     return []
 
 
+async def _save_image_to_storage(
+    img_bytes: bytes,
+    ext: str,
+    tenant_id: uuid.UUID,
+    workspace_id: uuid.UUID,
+) -> tuple[uuid.UUID, str]:
+    """
+    将图片字节保存到存储后端，写入 FileMetadata，返回 (file_id, url)。
+    """
+    from app.services.file_storage_service import FileStorageService, generate_file_key
+
+    file_id = uuid.uuid4()
+    file_ext = f".{ext}" if not ext.startswith(".") else ext
+    content_type = f"image/{ext}"
+
+    file_key = generate_file_key(
+        tenant_id=tenant_id,
+        workspace_id=workspace_id,
+        file_id=file_id,
+        file_ext=file_ext,
+    )
+
+    storage_svc = FileStorageService()
+    await storage_svc.storage.upload(file_key, img_bytes, content_type)
+
+    with get_db_read() as db:
+        meta = FileMetadata(
+            id=file_id,
+            tenant_id=tenant_id,
+            workspace_id=workspace_id,
+            file_key=file_key,
+            file_name=f"doc_image_{file_id}{file_ext}",
+            file_ext=file_ext,
+            file_size=len(img_bytes),
+            content_type=content_type,
+            status="completed",
+        )
+        db.add(meta)
+        db.commit()
+
+    url = f"{settings.FILE_LOCAL_SERVER_URL}/storage/permanent/{file_id}"
+    return file_id, url
+
+
 class DocExtractorNode(BaseNode):
     """Document Extractor Node.
 
     Reads one or more file variables and extracts their text content
-    by delegating to MultimodalService._extract_document_text.
+    and embedded images.
 
     Outputs:
-        text   (string)        – full concatenated text of all input files
-        chunks (array[string]) – per-file extracted text
+        text   (string)        – full text with image placeholders like [图片 第N页 第M张]
+        chunks (array[string]) – per-file extracted text (with placeholders)
+        images (array[file])   – extracted images as FileObject list, each with
+                                 name encoding position: "p{page}_i{index}"
     """
 
     def _output_types(self) -> dict[str, VariableType]:
         return {
             "text": VariableType.STRING,
             "chunks": VariableType.ARRAY_STRING,
+            "images": VariableType.ARRAY_FILE,
         }
 
     def _extract_output(self, business_result: Any) -> Any:
@@ -80,13 +129,18 @@ class DocExtractorNode(BaseNode):
         raw_val = self.get_variable(config.file_selector, variable_pool, strict=False)
         if raw_val is None:
             logger.warning(f"Node {self.node_id}: file variable '{config.file_selector}' is empty")
-            return {"text": "", "chunks": []}
+            return {"text": "", "chunks": [], "images": []}
 
         files = _normalise_files(raw_val)
         if not files:
-            return {"text": "", "chunks": []}
+            return {"text": "", "chunks": [], "images": []}
+
+        tenant_id = uuid.UUID(self.get_variable("sys.tenant_id", variable_pool, strict=False) or str(uuid.uuid4()))
+        workspace_id = uuid.UUID(self.get_variable("sys.workspace_id", variable_pool))
 
         chunks: list[str] = []
+        image_file_objects: list[dict] = []
+
         with get_db_read() as db:
             from app.services.multimodal_service import MultimodalService
             svc = MultimodalService(db)
@@ -94,13 +148,44 @@ class DocExtractorNode(BaseNode):
                 label = f.name or f.url or f.file_id
                 try:
                     file_input = _file_object_to_file_input(f)
-                    # Ensure URL is populated for local files
                     if not file_input.url:
                         file_input.url = await svc.get_file_url(file_input)
-                    # Reuse cached bytes if already fetched
                     if f.get_content():
                         file_input.set_content(f.get_content())
+
                     text = await svc.extract_document_text(file_input)
+
+                    # 从工作流 features 读取 document_image_recognition 开关
+                    fu_config = self.workflow_config.get("features", {}).get("file_upload", {})
+                    image_recognition = isinstance(fu_config, dict) and fu_config.get("document_image_recognition", False)
+                    if image_recognition:
+                        img_infos = await svc.extract_document_images(file_input)
+                        for img_info in img_infos:
+                            page = img_info["page"]
+                            index = img_info["index"]
+                            ext = img_info.get("ext", "png")
+                            placeholder = f"[图片 第{page}页 第{index + 1}张]" if page > 0 else f"[图片 第{index + 1}张]"
+                            try:
+                                file_id, url = await _save_image_to_storage(
+                                    img_bytes=img_info["bytes"],
+                                    ext=ext,
+                                    tenant_id=tenant_id,
+                                    workspace_id=workspace_id,
+                                )
+                                image_file_objects.append(FileObject(
+                                    type=FileType.IMAGE,
+                                    url=url,
+                                    transfer_method=TransferMethod.REMOTE_URL,
+                                    origin_file_type=f"image/{ext}",
+                                    file_id=str(file_id),
+                                    name=f"p{page}_i{index}",
+                                    mime_type=f"image/{ext}",
+                                    is_file=True,
+                                ).model_dump())
+                                text = text + f"\n{placeholder}: {url}"
+                            except Exception as e:
+                                logger.error(f"Node {self.node_id}: failed to save image {placeholder}: {e}")
+
                     chunks.append(text)
                 except Exception as e:
                     logger.error(
@@ -110,5 +195,8 @@ class DocExtractorNode(BaseNode):
                     chunks.append("")
 
         full_text = "\n\n".join(c for c in chunks if c)
-        logger.info(f"Node {self.node_id}: extracted {len(files)} file(s), total chars={len(full_text)}")
-        return {"text": full_text, "chunks": chunks}
+        logger.info(
+            f"Node {self.node_id}: extracted {len(files)} file(s), "
+            f"total chars={len(full_text)}, images={len(image_file_objects)}"
+        )
+        return {"text": full_text, "chunks": chunks, "images": image_file_objects}

api/app/core/workflow/nodes/knowledge/node.py

@@ -333,7 +333,7 @@ class KnowledgeRetrievalNode(BaseNode):
             tasks = []
             for kb_config in knowledge_bases:
                 db_knowledge = knowledge_repository.get_knowledge_by_id(db=db, knowledge_id=kb_config.kb_id)
-                if not db_knowledge:
+                if not (db_knowledge and db_knowledge.chunk_num > 0 and db_knowledge.status == 1):
                     raise RuntimeError("The knowledge base does not exist or access is denied.")
                 tasks.append(self.knowledge_retrieval(db, query, db_knowledge, kb_config))
             if tasks:

api/app/repositories/knowledge_repository.py

@@ -114,7 +114,7 @@ def get_knowledge_by_id(db: Session, knowledge_id: uuid.UUID) -> Knowledge | Non
 def get_knowledges_by_parent_id(db: Session, parent_id: uuid.UUID) -> list[Knowledge]:
     db_logger.debug(f"Query knowledge bases based on parent ID: parent_id={parent_id}")
     try:
-        knowledges = db.query(Knowledge).filter(Knowledge.parent_id == parent_id).all()
+        knowledges = db.query(Knowledge).filter(Knowledge.parent_id == parent_id, Knowledge.status == 1).all()
         if knowledges:
             db_logger.debug(f"Knowledge bases query successful: count={len(knowledges)} (parent_id: {parent_id})")
         else:

api/app/schemas/app_schema.py

@@ -155,6 +155,10 @@ class FileUploadConfig(BaseModel):
     document_allowed_extensions: List[str] = Field(
         default=["pdf", "docx", "doc", "xlsx", "xls", "txt", "csv", "json", "md"]
     )
+    document_image_recognition: bool = Field(
+        default=False,
+        description="是否识别文档中的图片（需配置视觉模型）"
+    )
     # 视频文件：MP4/MOV/AVI/WebM，最大 500MB
     video_enabled: bool = Field(default=False)
     video_max_size_mb: int = Field(default=50)

api/app/services/app_chat_service.py

@@ -16,7 +16,7 @@ from app.models import MultiAgentConfig, AgentConfig, ModelType
 from app.models import WorkflowConfig
 from app.repositories.tool_repository import ToolRepository
 from app.schemas import DraftRunRequest
-from app.schemas.app_schema import FileInput
+from app.schemas.app_schema import FileInput, FileType
 from app.schemas.model_schema import ModelInfo
 from app.schemas.prompt_schema import render_prompt_message, PromptMessageRole
 from app.services.conversation_service import ConversationService
@@ -165,8 +165,27 @@ class AppChatService:
         processed_files = None
         if files:
             multimodal_service = MultimodalService(self.db, model_info)
-            processed_files = await multimodal_service.process_files(files)
+            fu_config = features_config.get("file_upload", {})
+            if hasattr(fu_config, "model_dump"):
+                fu_config = fu_config.model_dump()
+            doc_img_recognition = isinstance(fu_config, dict) and fu_config.get("document_image_recognition", False)
+            processed_files = await multimodal_service.process_files(
+                files, document_image_recognition=doc_img_recognition
+            )
             logger.info(f"处理了 {len(processed_files)} 个文件")
+            if doc_img_recognition and "vision" in (api_key_obj.capability or []) and any(
+                f.type == FileType.DOCUMENT for f in files
+            ):
+                from langchain.agents import create_agent
+                agent.system_prompt += (
+                    "\n\n文档中包含图片，图片位置已在文本中以 [第N页 第M张图片]: URL 标记。"
+                    "请在回答中用 Markdown 格式 ![描述](URL) 展示相关图片，做到图文并茂。"
+                )
+                agent.agent = create_agent(
+                    model=agent.llm,
+                    tools=agent._wrap_tools_with_tracking(agent.tools) if agent.tools else None,
+                    system_prompt=agent.system_prompt
+                )
         # 为需要运行时上下文的工具注入上下文
         for t in tools:
             if hasattr(t, 'tool_instance') and hasattr(t.tool_instance, 'set_runtime_context'):
@@ -438,8 +457,27 @@ class AppChatService:
             processed_files = None
             if files:
                 multimodal_service = MultimodalService(self.db, model_info)
-                processed_files = await multimodal_service.process_files(files)
+                fu_config = features_config.get("file_upload", {})
+                if hasattr(fu_config, "model_dump"):
+                    fu_config = fu_config.model_dump()
+                doc_img_recognition = isinstance(fu_config, dict) and fu_config.get("document_image_recognition", False)
+                processed_files = await multimodal_service.process_files(
+                    files, document_image_recognition=doc_img_recognition
+                )
                 logger.info(f"处理了 {len(processed_files)} 个文件")
+                if doc_img_recognition and "vision" in (api_key_obj.capability or []) and any(
+                    f.type == FileType.DOCUMENT for f in files
+                ):
+                    from langchain.agents import create_agent
+                    agent.system_prompt += (
+                        "\n\n文档中包含图片，图片位置已在文本中以 [第N页 第M张图片]: URL 标记。"
+                        "请在回答中用 Markdown 格式 ![描述](URL) 展示相关图片，做到图文并茂。"
+                    )
+                    agent.agent = create_agent(
+                        model=agent.llm,
+                        tools=agent._wrap_tools_with_tracking(agent.tools) if agent.tools else None,
+                        system_prompt=agent.system_prompt
+                    )
 
             # 为需要运行时上下文的工具注入上下文
             for t in tools:

api/app/services/auth_service.py

@@ -1,3 +1,5 @@
+import uuid
+
 from sqlalchemy.orm import Session
 from typing import Optional, Tuple, Union
 import jwt
@@ -130,7 +132,7 @@ def register_user_with_invite(
     email: str,
     password: str,
     invite_token: str,
-    workspace_id: str,
+    workspace_id: uuid.UUID,
     username: Optional[str] = None,
 ) -> User:
     """
@@ -147,6 +149,7 @@ def register_user_with_invite(
     from app.schemas.user_schema import UserCreate
     from app.schemas.workspace_schema import InviteAcceptRequest
     from app.services import user_service, workspace_service
+    from app.repositories import workspace_repository as ws_repo
     from app.core.logging_config import get_business_logger
     
     logger = get_business_logger()
@@ -159,7 +162,8 @@ def register_user_with_invite(
             password=password,
             username=email.split('@')[0] if not username else username
         )
-        user = user_service.create_user(db=db, user=user_create)
+        workspace = ws_repo.get_workspace_by_id(db=db, workspace_id=workspace_id)
+        user = user_service.create_user(db=db, user=user_create, workspace=workspace)
         logger.info(f"用户创建成功: {user.email} (ID: {user.id})")
         
         # 接受工作空间邀请（此时用户已成为工作空间成员，并且会 commit）

api/app/services/draft_run_service.py

@@ -10,6 +10,7 @@ import time
 import uuid
 from typing import Any, AsyncGenerator, Dict, List, Optional
 
+from langchain.agents import create_agent
 from langchain.tools import tool
 from pydantic import BaseModel, Field
 from sqlalchemy import select
@@ -27,7 +28,7 @@ from app.core.rag.nlp.search import knowledge_retrieval
 from app.db import get_db_context
 from app.models import AgentConfig, ModelConfig
 from app.repositories.tool_repository import ToolRepository
-from app.schemas.app_schema import FileInput, Citation
+from app.schemas.app_schema import FileInput, Citation, FileType
 from app.schemas.model_schema import ModelInfo
 from app.schemas.prompt_schema import PromptMessageRole, render_prompt_message
 from app.services.conversation_service import ConversationService
@@ -637,12 +638,35 @@ class AgentRunService:
 
             # 6. 处理多模态文件
             processed_files = None
+            has_doc_with_images = False
             if files:
-                # 获取 provider 信息
                 provider = api_key_config.get("provider", "openai")
                 multimodal_service = MultimodalService(self.db, model_info)
-                processed_files = await multimodal_service.process_files(files)
+                fu_config = features_config.get("file_upload", {})
+                if hasattr(fu_config, "model_dump"):
+                    fu_config = fu_config.model_dump()
+                doc_img_recognition = isinstance(fu_config, dict) and fu_config.get("document_image_recognition", False)
+                processed_files = await multimodal_service.process_files(
+                    files, document_image_recognition=doc_img_recognition
+                )
                 logger.info(f"处理了 {len(processed_files)} 个文件，provider={provider}")
+                capability = api_key_config.get("capability", [])
+                has_doc_with_images = (
+                    doc_img_recognition
+                    and "vision" in capability
+                    and any(f.type == FileType.DOCUMENT for f in files)
+                )
+            if has_doc_with_images:
+                agent.system_prompt += (
+                    "\n\n文档中包含图片，图片位置已在文本中以 [第N页 第M张图片]: URL 标记。"
+                    "请在回答中用 Markdown 格式 ![描述](URL) 展示相关图片，做到图文并茂。"
+                )
+                # 重建 agent graph 以使新 system_prompt 生效
+                agent.agent = create_agent(
+                    model=agent.llm,
+                    tools=agent._wrap_tools_with_tracking(agent.tools) if agent.tools else None,
+                    system_prompt=agent.system_prompt
+                )
             # 为需要运行时上下文的工具注入上下文
             for t in tools:
                 if hasattr(t, 'tool_instance') and hasattr(t.tool_instance, 'set_runtime_context'):
@@ -895,12 +919,34 @@ class AgentRunService:
 
             # 6. 处理多模态文件
             processed_files = None
+            has_doc_with_images = False
             if files:
-                # 获取 provider 信息
                 provider = api_key_config.get("provider", "openai")
                 multimodal_service = MultimodalService(self.db, model_info)
-                processed_files = await multimodal_service.process_files(files)
+                fu_config = features_config.get("file_upload", {})
+                if hasattr(fu_config, "model_dump"):
+                    fu_config = fu_config.model_dump()
+                doc_img_recognition = isinstance(fu_config, dict) and fu_config.get("document_image_recognition", False)
+                processed_files = await multimodal_service.process_files(
+                    files, document_image_recognition=doc_img_recognition
+                )
                 logger.info(f"处理了 {len(processed_files)} 个文件，provider={provider}")
+                capability = api_key_config.get("capability", [])
+                has_doc_with_images = (
+                    doc_img_recognition
+                    and "vision" in capability
+                    and any(f.type == FileType.DOCUMENT for f in files)
+                )
+            if has_doc_with_images:
+                agent.system_prompt += (
+                    "\n\n文档中包含图片，图片位置已在文本中以 [第N页 第M张图片]: URL 标记。"
+                    "请在回答中用 Markdown 格式 ![描述](URL) 展示相关图片，做到图文并茂。"
+                )
+                agent.agent = create_agent(
+                    model=agent.llm,
+                    tools=agent._wrap_tools_with_tracking(agent.tools) if agent.tools else None,
+                    system_prompt=agent.system_prompt
+                )
             # 为需要运行时上下文的工具注入上下文
             for t in tools:
                 if hasattr(t, 'tool_instance') and hasattr(t.tool_instance, 'set_runtime_context'):

api/app/services/memory_dashboard_service.py

@@ -821,7 +821,7 @@ def get_rag_content(
         for document in documents:
             try:
                 kb = knowledge_repository.get_knowledge_by_id(db, document.kb_id)
-                if not kb:
+                if not (kb and kb.status == 1):
                     business_logger.warning(f"知识库不存在: kb_id={document.kb_id}")
                     continue
                 

api/app/services/multimodal_service.py

@@ -344,6 +344,7 @@ class MultimodalService:
     async def process_files(
             self,
             files: Optional[List[FileInput]],
+            document_image_recognition: bool = False,
     ) -> List[Dict[str, Any]]:
         """
         处理文件列表，返回 LLM 可用的格式
@@ -379,6 +380,31 @@ class MultimodalService:
                 elif file.type == FileType.DOCUMENT:
                     is_support, content = await self._process_document(file, strategy)
                     result.append(content)
+                    # 仅当开关开启且模型支持视觉时，才提取文档内嵌图片
+                    if document_image_recognition and "vision" in self.capability:
+                        img_infos = await self.extract_document_images(file)
+                        for img_info in img_infos:
+                            page = img_info["page"]
+                            index = img_info["index"]
+                            ext = img_info.get("ext", "png")
+                            try:
+                                _, img_url = await self._save_doc_image_to_storage(img_info["bytes"], ext)
+                                placeholder = f"第{page}页 第{index + 1}张图片" if page > 0 else f"第{index + 1}张图片"
+                                # 在文本内容中追加图片位置标记
+                                if result and result[-1].get("type") in ("text", "document"):
+                                    key = "text" if "text" in result[-1] else list(result[-1].keys())[-1]
+                                    result[-1][key] = result[-1].get(key, "") + f"\n[{placeholder}]: {img_url}"
+                                # 将图片以视觉格式追加到消息内容中
+                                img_file = FileInput(
+                                    type=FileType.IMAGE,
+                                    transfer_method=TransferMethod.REMOTE_URL,
+                                    url=img_url,
+                                    file_type="image/png",
+                                )
+                                _, img_content = await self._process_image(img_file, strategy_class(img_file))
+                                result.append(img_content)
+                            except Exception as img_err:
+                                logger.warning(f"文档图片处理失败: {img_err}")
                 elif file.type == FileType.AUDIO and "audio" in self.capability:
                     is_support, content = await self._process_audio(file, strategy)
                     result.append(content)
@@ -431,12 +457,8 @@ class MultimodalService:
         """
         处理文档文件（PDF、Word 等）
         
-        Args:
-            file: 文档文件输入
-            strategy: 格式化策略
-            
         Returns:
-            Dict: 根据 provider 返回不同格式的文档内容
+            仅返回文本内容（图片通过 process_files 中的额外步骤追加）
         """
         if file.transfer_method == TransferMethod.REMOTE_URL:
             return True, {
@@ -444,19 +466,63 @@ class MultimodalService:
                 "text": f"<document url=\"{file.url}\">\n{await self.extract_document_text(file)}\n</document>"
             }
         else:
-            # 本地文件，提取文本内容
             server_url = settings.FILE_LOCAL_SERVER_URL
             file.url = f"{server_url}/storage/permanent/{file.upload_file_id}"
             text = await self.extract_document_text(file)
             file_metadata = self.db.query(FileMetadata).filter(
                 FileMetadata.id == file.upload_file_id
             ).first()
-
             file_name = file_metadata.file_name if file_metadata else "unknown"
-
-            # 使用策略格式化文档
             return await strategy.format_document(file_name, text)
 
+    async def _save_doc_image_to_storage(
+            self,
+            img_bytes: bytes,
+            ext: str,
+    ) -> tuple[str, str]:
+        """
+        将文档内嵌图片保存到存储后端，写入 FileMetadata。
+        tenant_id / workspace_id 从 api_config 所在的 FileMetadata 上下文获取，
+        无法获取时使用占位 UUID（图片仍可通过 permanent URL 访问）。
+
+        Returns:
+            (file_id_str, permanent_url)
+        """
+        import uuid as _uuid
+        from app.services.file_storage_service import FileStorageService, generate_file_key
+        from app.db import get_db_context
+
+        file_id = _uuid.uuid4()
+        file_ext = f".{ext}" if not ext.startswith(".") else ext
+        content_type = f"image/{ext}"
+
+        # tenant_id / workspace_id 尽量从已有 FileMetadata 推断，否则用占位值
+        placeholder = _uuid.UUID(int=0)
+        tenant_id = placeholder
+        workspace_id = placeholder
+
+        file_key = generate_file_key(tenant_id, workspace_id, file_id, file_ext)
+        storage_svc = FileStorageService()
+        await storage_svc.storage.upload(file_key, img_bytes, content_type)
+
+        with get_db_context() as db:
+            meta = FileMetadata(
+                id=file_id,
+                tenant_id=tenant_id,
+                workspace_id=workspace_id,
+                file_key=file_key,
+                file_name=f"doc_image_{file_id}{file_ext}",
+                file_ext=file_ext,
+                file_size=len(img_bytes),
+                content_type=content_type,
+                status="completed",
+            )
+            db.add(meta)
+            db.commit()
+
+        url = f"{settings.FILE_LOCAL_SERVER_URL}/storage/permanent/{file_id}"
+        return str(file_id), url
+
     async def _process_audio(self, file: FileInput, strategy) -> tuple[bool, Dict[str, Any]]:
         """
         处理音频文件
@@ -582,6 +648,84 @@ class MultimodalService:
             logger.error(f"Failed to load file. - {e}")
             return "[Failed to load file.]"
 
+    async def extract_document_images(self, file: FileInput) -> list[dict]:
+        """
+        提取文档中的内嵌图片（支持 PDF 和 DOCX），附带位置信息。
+
+        Returns:
+            list[dict]: 每项包含:
+                - bytes: 图片二进制
+                - page: 所在页码（PDF 从 1 开始，DOCX 为 0）
+                - index: 该页/文档内的图片序号（从 0 开始）
+                - ext: 图片扩展名（如 png、jpeg）
+        """
+        try:
+            file_content = file.get_content()
+            if not file_content:
+                async with httpx.AsyncClient(timeout=30.0) as client:
+                    response = await client.get(file.url, follow_redirects=True)
+                    response.raise_for_status()
+                    file_content = response.content
+                    file.set_content(file_content)
+
+            file_mime_type = magic.from_buffer(file_content, mime=True)
+            if file_mime_type in PDF_MIME:
+                return self._extract_pdf_images(file_content)
+            elif self._is_word_file(file_content, file_mime_type):
+                return self._extract_docx_images(file_content)
+            return []
+        except Exception as e:
+            logger.error(f"提取文档图片失败: {e}")
+            return []
+
+    @staticmethod
+    def _extract_pdf_images(file_content: bytes) -> list[dict]:
+        """从 PDF 提取内嵌图片，附带页码和序号"""
+        images = []
+        try:
+            import fitz  # PyMuPDF
+            doc = fitz.open(stream=file_content, filetype="pdf")
+            for page_num, page in enumerate(doc, start=1):
+                for idx, img in enumerate(page.get_images(full=True)):
+                    xref = img[0]
+                    base_image = doc.extract_image(xref)
+                    images.append({
+                        "bytes": base_image["image"],
+                        "ext": base_image.get("ext", "png"),
+                        "page": page_num,
+                        "index": idx,
+                    })
+            doc.close()
+        except ImportError:
+            logger.warning("PyMuPDF 未安装，无法提取 PDF 图片，请执行: uv add pymupdf")
+        except Exception as e:
+            logger.error(f"提取 PDF 图片失败: {e}")
+        return images
+
+    @staticmethod
+    def _extract_docx_images(file_content: bytes) -> list[dict]:
+        """从 DOCX 提取内嵌图片，附带序号（DOCX 无页码概念，page 固定为 0）"""
+        images = []
+        try:
+            if file_content[:2] != b'PK':
+                return []
+            with zipfile.ZipFile(io.BytesIO(file_content)) as zf:
+                media_files = sorted(
+                    name for name in zf.namelist()
+                    if name.startswith("word/media/") and not name.endswith("/")
+                )
+                for idx, name in enumerate(media_files):
+                    ext = name.rsplit(".", 1)[-1].lower() if "." in name else "png"
+                    images.append({
+                        "bytes": zf.read(name),
+                        "ext": ext,
+                        "page": 0,
+                        "index": idx,
+                    })
+        except Exception as e:
+            logger.error(f"提取 DOCX 图片失败: {e}")
+        return images
+
     @staticmethod
     async def _extract_pdf_text(file_content: bytes) -> str:
         """提取 PDF 文本"""

api/app/services/user_service.py

@@ -8,6 +8,7 @@ from sqlalchemy.orm import Session
 import uuid
 
 from app.aioRedis import aio_redis_set, aio_redis_get, aio_redis_delete
+from app.models import Workspace
 from app.models.user_model import User
 from app.repositories import user_repository
 from app.schemas.user_schema import UserCreate
@@ -74,7 +75,7 @@ def create_initial_superuser(db: Session):
         )
 
 
-def create_user(db: Session, user: UserCreate) -> User:
+def create_user(db: Session, user: UserCreate, workspace: Workspace) -> User:
     business_logger.info(f"创建用户: {user.username}, email: {user.email}")
     
     try:
@@ -93,24 +94,9 @@ def create_user(db: Session, user: UserCreate) -> User:
         business_logger.debug(f"开始创建用户: {user.username}")
         hashed_password = get_password_hash(user.password)
         
-        # 获取默认租户（第一个活跃租户）
-        from app.repositories.tenant_repository import TenantRepository
-        tenant_repo = TenantRepository(db)
-        tenants = tenant_repo.get_tenants(skip=0, limit=1, is_active=True)
-        
-        if not tenants:
-            business_logger.error("系统中没有可用的租户")
-            raise BusinessException(
-                "系统配置错误：没有可用的租户", 
-                code=BizCode.TENANT_NOT_FOUND,
-                context={"username": user.username, "email": user.email}
-            )
-        
-        default_tenant = tenants[0]
-        
         new_user = user_repository.create_user(
             db=db, user=user, hashed_password=hashed_password, 
-            tenant_id=default_tenant.id, is_superuser=False
+            tenant_id=workspace.tenant_id, is_superuser=False
         )
 
         db.commit()

api/app/services/workflow_service.py

@@ -694,7 +694,8 @@ class WorkflowService:
             "nodes": config.nodes,
             "edges": config.edges,
             "variables": config.variables,
-            "execution_config": config.execution_config
+            "execution_config": config.execution_config,
+            "features": feature_configs
         }
 
         try:
@@ -894,7 +895,8 @@ class WorkflowService:
             "nodes": config.nodes,
             "edges": config.edges,
             "variables": config.variables,
-            "execution_config": config.execution_config
+            "execution_config": config.execution_config,
+            "features": feature_configs
         }
 
         try:

api/app/version_info.json

@@ -1,4 +1,36 @@
 {
+  "v0.3.1": {
+    "introduction": {
+      "codeName": "无境",
+      "releaseDate": "2026-4-22",
+      "upgradePosition": "🐻 聚焦应用体验优化、记忆 API 开放与工作流可靠性提升，打破边界，自由流动",
+      "coreUpgrades": [
+        "1. 应用与模型增强<br>* 模型 Key 全删后自动关闭：避免无 Key 运行时错误<br>* 模型 JSON 格式化输出开关：支持旧工作流迁移的稳定 JSON 输出<br>* 配置导入覆盖：支持完整替换当前配置<br>* 导入时缺失资源清理：自动清空不存在的工具和知识库引用",
+        "2. 记忆 API 与智能 📚<br>* 记忆读写 API 与 End-User Key 供给：支持第三方直接交互记忆层<br>* 记忆库 API 与配置更新：程序化控制记忆设置（提供顺序接口）<br>* End-User 元数据存储：丰富用户上下文持久化",
+        "3. 工作流与体验优化 ⚙️<br>* 会话历史文件元数据：增加文件大小、名称和类型<br>* 迭代节点并行输入修复：恢复并发执行行为<br>* API Key 后四位展示：便于密钥识别<br>* 条件分支多文件子变量：更精细的条件逻辑<br>* Agent 模型配置重置接口：完善前后端契约<br>* 三级变量键盘导航：提升变量选择体验<br>* 应用标签页动态标题：动态显示应用名称<br>* 变量聚合三级勾选修复：修复勾选行为<br>* 工作流检查清单校验增强：工具必填和视觉变量必填<br>* 变量聚合器到参数提取器输出：修复输出变量获取",
+        "4. 知识库与性能 ⚡<br>* 文档解析与 Graph 异步执行：提升文档摄入吞吐量",
+        "5. 稳健性与缺陷修复 🔧<br>* 工具节点原始参数类型：修复类型不匹配问题<br>* 前端部署后资源过期导入错误：解决缓存资源导入失败<br>* 工作流工具节点必填校验：防止不完整配置发布",
+        "<br>",
+        "v0.3.1 是平台哲学演进中的关键时刻——边界的打破。记忆 API 开放和应用体验优化为社区用户提供更强大的集成能力。展望未来，我们将持续提升记忆智能管线的萃取精度与自适应遗忘策略，深化工作流引擎能力。破界而行，臻于无境。",
+        "MemoryBear — 无境 🐻✨"
+      ]
+    },
+    "introduction_en": {
+      "codeName": "WuJing",
+      "releaseDate": "2026-4-22",
+      "upgradePosition": "🐻 Focused application improvements, memory API openness, and workflow reliability — dissolving boundaries, flowing freely",
+      "coreUpgrades": [
+        "1. Application & Model Enhancements<br>* Model Auto-Disable on Key Deletion: Prevents keyless runtime errors<br>* Model JSON Formatted Output Toggle: Stable JSON output for legacy workflow migration<br>* Configuration Import with Override: Full configuration replacement support<br>* Import Cleanup for Missing Resources: Auto-clears missing tool and knowledge base references",
+        "2. Memory API & Intelligence 📚<br>* Memory Read/Write API with End-User Key Provisioning: Third-party memory layer interaction<br>* Memory Store API & Configuration Update: Programmatic memory settings control with sequential interface<br>* End-User Metadata Storage: Richer user context persistence",
+        "3. Workflow & UX Improvements ⚙️<br>* Conversation History File Metadata: File size, name, and type labels<br>* Iteration Node Parallel Input Fix: Restored concurrent execution<br>* API Key Last Four Digits Display: Key identification without exposure<br>* Condition Branch Multi-File Sub-Variables: Granular conditional logic<br>* Agent Model Config Reset Endpoint: Completed frontend-backend contract<br>* Three-Level Variable Keyboard Navigation: Improved selection experience<br>* Dynamic Tab Title for Applications: Dynamic app name in browser tab<br>* Variable Aggregator Three-Level Checkbox Fix: Corrected checkbox behavior<br>* Workflow Checklist Validation Enhancements: Tool required and vision variable validation<br>* Variable Aggregator to Parameter Extractor Output: Fixed output variable access",
+        "4. Knowledge Base & Performance ⚡<br>* Async Document Parsing & Graph Execution: Improved document ingestion throughput",
+        "5. Robustness & Bug Fixes 🔧<br>* Tool Node Raw Parameter Types: Fixed type mismatch issues<br>* Stale Frontend Resource Import Error: Resolved cached resource import failure<br>* Workflow Tool Node Required Validation: Prevents incomplete configuration publishing",
+        "<br>",
+        "v0.3.1 marks a pivotal moment in the platform's evolution — the dissolution of boundaries. Memory API openness and application experience improvements provide community users with stronger integration capabilities. Looking ahead, we will continue improving extraction accuracy, adaptive forgetting strategies, and deepening workflow engine capabilities. Beyond boundaries — the boundless awaits.",
+        "MemoryBear — The Boundless 🐻✨"
+      ]
+    }
+  },
   "v0.3.0": {
     "introduction": {
       "codeName": "破晓",

api/pyproject.toml

@@ -147,7 +147,8 @@ dependencies = [
     "modelscope>=1.34.0",
     "python-magic>=0.4.14; sys_platform == 'linux' or sys_platform == 'darwin'",
     "python-magic-bin>=0.4.14; sys_platform=='win32'",
-    "volcengine-python-sdk[ark]==5.0.19"
+    "volcengine-python-sdk[ark]==5.0.19",
+    "pymupdf>=1.27.2.2",
 ]
 
 [tool.pytest.ini_options]