style: 优化 Settings 页面样式

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
chore: 更新前端依赖和侧边栏导航
2026-03-09 15:43:24 +08:00 · 2026-03-09 15:43:07 +08:00 · 2026-03-09 15:42:55 +08:00 · 2026-03-09 15:42:42 +08:00 · 2026-03-09 15:42:35 +08:00
41 changed files with 5627 additions and 235 deletions
--- a/ai-core/config.example.yaml
+++ b/ai-core/config.example.yaml
@@ -0,0 +1,18 @@
+# AI-Core 配置文件示例
+# 复制此文件为 config.yaml 并填入实际配置
+
+# VLM 配置（可选）
+# 如果配置了 VLM，图片文件会自动使用 VLM 解析
+vlm:
+  enabled: false  # 是否启用 VLM
+  provider: "openai"  # openai / anthropic / qwen
+  model: "gpt-4o"  # 模型名称
+  api_key: ""  # API Key
+  base_url: ""  # 自定义 API 地址（可选）
+  prompt: ""  # 自定义提示词（可选）
+
+# 服务配置
+server:
+  port: 50051
+  max_workers: 10
+  log_level: INFO
--- a/ai-core/parser/config.py
+++ b/ai-core/parser/config.py
@@ -0,0 +1,84 @@
+"""
+配置管理模块
+"""
+import os
+import yaml
+import logging
+from typing import Optional, Dict, Any
+
+logger = logging.getLogger(__name__)
+
+# 默认配置
+DEFAULT_CONFIG = {
+    "vlm": {
+        "enabled": False,
+        "provider": "openai",
+        "model": "gpt-4o",
+        "api_key": "",
+        "base_url": "",
+        "prompt": ""
+    },
+    "server": {
+        "port": 50051,
+        "max_workers": 10,
+        "log_level": "INFO"
+    }
+}
+
+
+def load_config(config_path: Optional[str] = None) -> Dict[str, Any]:
+    """加载配置文件"""
+    if config_path is None:
+        # 默认查找 config.yaml
+        base_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+        config_path = os.path.join(base_dir, "config.yaml")
+
+    # 环境变量覆盖
+    vlm_api_key = os.environ.get("VLM_API_KEY", "")
+    if vlm_api_key:
+        DEFAULT_CONFIG["vlm"]["api_key"] = vlm_api_key
+        DEFAULT_CONFIG["vlm"]["enabled"] = True
+        logger.info("VLM enabled via environment variable")
+
+    vlm_provider = os.environ.get("VLM_PROVIDER", "")
+    if vlm_provider:
+        DEFAULT_CONFIG["vlm"]["provider"] = vlm_provider
+
+    vlm_model = os.environ.get("VLM_MODEL", "")
+    if vlm_model:
+        DEFAULT_CONFIG["vlm"]["model"] = vlm_model
+
+    # 尝试加载配置文件
+    if os.path.exists(config_path):
+        try:
+            with open(config_path, 'r', encoding='utf-8') as f:
+                file_config = yaml.safe_load(f)
+                if file_config:
+                    # 合并配置
+                    for key in file_config:
+                        if key in DEFAULT_CONFIG:
+                            DEFAULT_CONFIG[key].update(file_config[key])
+                    logger.info(f"Loaded config from {config_path}")
+        except Exception as e:
+            logger.warning(f"Failed to load config: {e}")
+
+    # 检查 VLM 是否有效
+    if DEFAULT_CONFIG["vlm"]["enabled"] and not DEFAULT_CONFIG["vlm"]["api_key"]:
+        logger.warning("VLM enabled but API key is empty, disabling VLM")
+        DEFAULT_CONFIG["vlm"]["enabled"] = False
+
+    return DEFAULT_CONFIG
+
+
+def get_vlm_config() -> Optional[Dict[str, Any]]:
+    """获取 VLM 配置"""
+    config = load_config()
+    if config.get("vlm", {}).get("enabled") and config["vlm"].get("api_key"):
+        return config["vlm"]
+    return None
+
+
+def get_server_config() -> Dict[str, Any]:
+    """获取服务器配置"""
+    config = load_config()
+    return config.get("server", DEFAULT_CONFIG["server"])
--- a/ai-core/parser/parser.py
+++ b/ai-core/parser/parser.py
@@ -1,39 +1,68 @@
 import logging
 import os
 import tempfile
-from typing import Optional
+from typing import Optional, Dict, Any
 from markitdown import MarkItDown

+from .vlm_client import VLMClient
+from .config import get_vlm_config
+
 logger = logging.getLogger(__name__)


 class Parser:
-    """基于 MarkItDown 的统一文档解析器
-    
+    """基于 MarkItDown + VLM 的统一文档解析器
+
    支持格式：PDF、DOCX、DOC、PPTX、PPT、XLSX、XLS、CSV、图片、网页、Markdown 等
+
+    VLM 解析：
+    - 方式一：启动时配置（config.yaml 或环境变量）
+    - 方式二：gRPC 请求时传入 VLM 配置（优先级更高）
    """

    def __init__(self):
        self.markitdown = MarkItDown()
-        logger.info("Parser initialized with MarkItDown")
+        self.vlm_client: Optional[VLMClient] = None

-    def parse(self, file_path: str, file_type: Optional[str] = None) -> dict:
+        # 尝试加载配置的 VLM
+        vlm_config = get_vlm_config()
+        if vlm_config:
+            self.vlm_client = VLMClient(vlm_config)
+            logger.info(f"VLM enabled: provider={vlm_config.get('provider')}, model={vlm_config.get('model')}")
+        else:
+            logger.info("VLM not configured, using MarkItDown only")
+
+    def set_vlm_config(self, config: Dict[str, Any]) -> None:
+        """手动设置 VLM 配置（优先级高于全局配置）"""
+        if config and config.get("enabled") and config.get("api_key"):
+            self.vlm_client = VLMClient(config)
+            logger.info(f"VLM enabled: provider={config.get('provider')}, model={config.get('model')}")
+        else:
+            self.vlm_client = None
+            logger.info("VLM disabled")
+
+    def parse(self, file_path: str, file_type: Optional[str] = None, vlm_config: Optional[Dict[str, Any]] = None) -> dict:
        """解析文档为 Markdown
-        
+
        Args:
            file_path: 文件路径或 URL
            file_type: 文件类型（可选，MarkItDown 会自动检测）
-            
+            vlm_config: VLM 配置（可选，优先级高于全局配置）
+
        Returns:
            dict: 包含 markdown 内容和元数据
        """
+        # 如果有 VLM 配置，覆盖全局配置
+        if vlm_config:
+            self.set_vlm_config(vlm_config)
+
        try:
            logger.info(f"Parsing file: {file_path}")
-            
+
            result = self.markitdown.convert(file_path)
-            
+
            logger.info(f"Parse successful: {len(result.text_content)} characters")
-            
+
            return {
                "success": True,
                "content": result.text_content,
@@ -49,29 +78,40 @@ class Parser:
                "error": str(e)
            }

-    def parse_bytes(self, content: bytes, file_name: str, file_type: Optional[str] = None) -> dict:
+    def parse_bytes(self, content: bytes, file_name: str, file_type: Optional[str] = None, vlm_config: Optional[Dict[str, Any]] = None) -> dict:
        """解析字节内容为 Markdown
-        
+
        Args:
            content: 文件字节内容
            file_name: 文件名
            file_type: 文件类型（可选）
-            
+            vlm_config: VLM 配置（可选，优先级高于全局配置）
+
        Returns:
            dict: 包含 markdown 内容和元数据
        """
+        # 如果有 VLM 配置，覆盖全局配置
+        if vlm_config:
+            self.set_vlm_config(vlm_config)
+
        try:
            logger.info(f"Parsing bytes: {file_name}, size: {len(content)} bytes")
-            
+
+            # 检查是否应该使用 VLM（根据文件名自动判断）
+            if self._should_use_vlm(file_name):
+                logger.info("Using VLM for parsing")
+                return self._parse_with_vlm(content, file_name)
+
+            # 否则使用 MarkItDown
            with tempfile.NamedTemporaryFile(delete=False, suffix=os.path.splitext(file_name)[1] or '') as temp_file:
                temp_file.write(content)
                temp_path = temp_file.name
-            
+
            try:
                result = self.markitdown.convert(temp_path)
-                
+
                logger.info(f"Parse successful: {len(result.text_content)} characters")
-                
+
                return {
                    "success": True,
                    "content": result.text_content,
@@ -89,10 +129,69 @@ class Parser:
                "error": str(e)
            }

+    def _should_use_vlm(self, file_name: str) -> bool:
+        """判断是否应该使用 VLM"""
+        if not self.vlm_client:
+            return False
+
+        # 图片文件使用 VLM
+        image_exts = ['.jpg', '.jpeg', '.png', '.gif', '.bmp', '.webp', '.tiff']
+        ext = os.path.splitext(file_name)[1].lower()
+        return ext in image_exts
+
+    def _parse_with_vlm(self, content: bytes, file_name: str) -> dict:
+        """使用 VLM 解析"""
+        if not self.vlm_client:
+            return {
+                "success": False,
+                "content": "",
+                "content_length": 0,
+                "error": "VLM not configured"
+            }
+
+        # 确定 MIME 类型
+        ext = os.path.splitext(file_name)[1].lower()
+        mime_types = {
+            '.jpg': 'image/jpeg',
+            '.jpeg': 'image/jpeg',
+            '.png': 'image/png',
+            '.gif': 'image/gif',
+            '.bmp': 'image/bmp',
+            '.webp': 'image/webp',
+            '.tiff': 'image/tiff',
+        }
+        mime_type = mime_types.get(ext, 'image/png')
+
+        try:
+            result = self.vlm_client.analyze_image(content, mime_type)
+
+            if result.get("success"):
+                return {
+                    "success": True,
+                    "content": result["content"],
+                    "content_length": len(result["content"]),
+                    "metadata": {"vlm_used": True}
+                }
+            else:
+                return {
+                    "success": False,
+                    "content": "",
+                    "content_length": 0,
+                    "error": result.get("error", "VLM parsing failed")
+                }
+        except Exception as e:
+            logger.error(f"VLM parsing error: {e}")
+            return {
+                "success": False,
+                "content": "",
+                "content_length": 0,
+                "error": str(e)
+            }
+

 if __name__ == "__main__":
    parser = Parser()
-    
+
    # 测试
    test_url = "https://example.com"
    result = parser.parse(test_url)
--- a/ai-core/parser/vlm_client.py
+++ b/ai-core/parser/vlm_client.py
@@ -0,0 +1,209 @@
+"""
+VLM 客户端 - 用于调用 VLM 模型进行文档理解
+"""
+import logging
+import base64
+import requests
+from typing import Optional, Dict, Any
+
+logger = logging.getLogger(__name__)
+
+
+class VLMClient:
+    """VLM 客户端，支持多种提供商"""
+
+    def __init__(self, config: Dict[str, Any]):
+        """
+        初始化 VLM 客户端
+
+        Args:
+            config: VLM 配置，包含 provider, model, api_key, base_url, prompt 等
+        """
+        self.config = config
+        self.provider = config.get("provider", "openai")
+        self.model = config.get("model", "gpt-4o")
+        self.api_key = config.get("api_key", "")
+        self.base_url = config.get("base_url", "")
+        self.prompt = config.get("prompt", "") or self._default_prompt()
+
+        logger.info(f"VLMClient initialized: provider={self.provider}, model={self.model}")
+
+    def _default_prompt(self) -> str:
+        """默认提示词"""
+        return """请分析这张图片中的文档内容，并将其转换为 Markdown 格式。
+要求：
+1. 保持原文的格式和结构
+2. 表格用 Markdown 表格格式
+3. 标题用 # ## ### 标记
+4. 代码块用 ``` 标记
+5. 尽量保留原文的所有信息"""
+
+    def analyze_image(self, image_data: bytes, mime_type: str = "image/png") -> Dict[str, Any]:
+        """
+        使用 VLM 分析图片
+
+        Args:
+            image_data: 图片二进制数据
+            mime_type: 图片 MIME 类型
+
+        Returns:
+            包含分析结果的字典
+        """
+        if self.provider == "openai":
+            return self._call_openai(image_data, mime_type)
+        elif self.provider == "anthropic":
+            return self._call_anthropic(image_data, mime_type)
+        elif self.provider == "qwen":
+            return self._call_qwen(image_data, mime_type)
+        else:
+            return {
+                "success": False,
+                "content": "",
+                "error": f"Unsupported provider: {self.provider}"
+            }
+
+    def _call_openai(self, image_data: bytes, mime_type: str) -> Dict[str, Any]:
+        """调用 OpenAI GPT-4o API"""
+        try:
+            url = (self.base_url or "https://api.openai.com/v1") + "/chat/completions"
+
+            # Base64 编码图片
+            image_base64 = base64.b64encode(image_data).decode("utf-8")
+            data_url = f"data:{mime_type};base64,{image_base64}"
+
+            headers = {
+                "Authorization": f"Bearer {self.api_key}",
+                "Content-Type": "application/json"
+            }
+
+            payload = {
+                "model": self.model,
+                "messages": [
+                    {
+                        "role": "user",
+                        "content": [
+                            {"type": "text", "text": self.prompt},
+                            {"type": "image_url", "image_url": {"url": data_url}}
+                        ]
+                    }
+                ],
+                "max_tokens": 4096
+            }
+
+            response = requests.post(url, headers=headers, json=payload, timeout=120)
+            response.raise_for_status()
+
+            result = response.json()
+            content = result["choices"][0]["message"]["content"]
+
+            return {
+                "success": True,
+                "content": content,
+                "usage": result.get("usage", {})
+            }
+        except Exception as e:
+            logger.error(f"OpenAI API error: {e}")
+            return {
+                "success": False,
+                "content": "",
+                "error": str(e)
+            }
+
+    def _call_anthropic(self, image_data: bytes, mime_type: str) -> Dict[str, Any]:
+        """调用 Anthropic Claude API"""
+        try:
+            url = (self.base_url or "https://api.anthropic.com/v1") + "/messages"
+
+            image_base64 = base64.b64encode(image_data).decode("utf-8")
+
+            headers = {
+                "x-api-key": self.api_key,
+                "anthropic-version": "2023-06-01",
+                "Content-Type": "application/json"
+            }
+
+            # Anthropic 支持 image 类型
+            payload = {
+                "model": self.model,
+                "max_tokens": 4096,
+                "messages": [
+                    {
+                        "role": "user",
+                        "content": [
+                            {"type": "text", "text": self.prompt},
+                            {
+                                "type": "image",
+                                "source": {
+                                    "type": "base64",
+                                    "media_type": mime_type,
+                                    "data": image_base64
+                                }
+                            }
+                        ]
+                    }
+                ]
+            }
+
+            response = requests.post(url, headers=headers, json=payload, timeout=120)
+            response.raise_for_status()
+
+            result = response.json()
+            content = result["content"][0]["text"]
+
+            return {
+                "success": True,
+                "content": content,
+                "usage": result.get("usage", {})
+            }
+        except Exception as e:
+            logger.error(f"Anthropic API error: {e}")
+            return {
+                "success": False,
+                "content": "",
+                "error": str(e)
+            }
+
+    def _call_qwen(self, image_data: bytes, mime_type: str) -> Dict[str, Any]:
+        """调用阿里 Qwen VL API"""
+        try:
+            url = (self.base_url or "https://dashscope.aliyuncs.com/compatible-mode/v1") + "/chat/completions"
+
+            image_base64 = base64.b64encode(image_data).decode("utf-8")
+
+            headers = {
+                "Authorization": f"Bearer {self.api_key}",
+                "Content-Type": "application/json"
+            }
+
+            # Qwen 格式
+            payload = {
+                "model": self.model,
+                "messages": [
+                    {
+                        "role": "user",
+                        "content": [
+                            {"type": "text", "text": self.prompt},
+                            {"type": "image_url", "image_url": {"url": f"data:{mime_type};base64,{image_base64}"}}
+                        ]
+                    }
+                ]
+            }
+
+            response = requests.post(url, headers=headers, json=payload, timeout=120)
+            response.raise_for_status()
+
+            result = response.json()
+            content = result["choices"][0]["message"]["content"]
+
+            return {
+                "success": True,
+                "content": content,
+                "usage": {}
+            }
+        except Exception as e:
+            logger.error(f"Qwen API error: {e}")
+            return {
+                "success": False,
+                "content": "",
+                "error": str(e)
+            }
--- a/ai-core/proto/document_parser.proto
+++ b/ai-core/proto/document_parser.proto
@@ -16,6 +16,18 @@ message ParseRequest {
  string file_type = 3;
  string parser_engine = 4;
  map<string, string> engine_overrides = 5;
+
+  // VLM 配置（可选）
+  VLMConfig vlm_config = 6;
+}
+
+message VLMConfig {
+  bool enabled = 1;              // 是否启用 VLM
+  string provider = 2;            // VLM 提供商: openai, anthropic, local 等
+  string model = 3;               // 模型名称
+  string api_key = 4;             // API Key
+  string base_url = 5;            // 自定义 API 地址
+  string prompt = 6;              // 自定义提示词
 }

 message ParseResponse {
@@ -44,4 +56,4 @@ message EngineInfo {
  repeated string supported_file_types = 3;
  bool available = 4;
  string unavailable_reason = 5;
-}
+}
--- a/ai-core/proto/document_parser_pb2.py
+++ b/ai-core/proto/document_parser_pb2.py
@@ -24,7 +24,7 @@ _sym_db = _symbol_database.Default()



-DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\x15\x64ocument_parser.proto\x12\tdocparser\"\xdd\x01\n\x0cParseRequest\x12\x10\n\x08\x66ile_url\x18\x01 \x01(\t\x12\x11\n\tfile_name\x18\x02 \x01(\t\x12\x11\n\tfile_type\x18\x03 \x01(\t\x12\x15\n\rparser_engine\x18\x04 \x01(\t\x12\x46\n\x10\x65ngine_overrides\x18\x05 \x03(\x0b\x32,.docparser.ParseRequest.EngineOverridesEntry\x1a\x36\n\x14\x45ngineOverridesEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\t:\x02\x38\x01\"\x84\x01\n\rParseResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07\x63ontent\x18\x02 \x01(\t\x12\x0f\n\x07message\x18\x03 \x01(\t\x12\x16\n\x0e\x63ontent_length\x18\x04 \x01(\x05\x12\x11\n\tfile_type\x18\x05 \x01(\t\x12\x15\n\rparser_engine\x18\x06 \x01(\t\"\x07\n\x05\x45mpty\"\xca\x01\n\x18SupportedFormatsResponse\x12\x12\n\nfile_types\x18\x01 \x03(\t\x12]\n\x16\x66ile_type_descriptions\x18\x02 \x03(\x0b\x32=.docparser.SupportedFormatsResponse.FileTypeDescriptionsEntry\x1a;\n\x19\x46ileTypeDescriptionsEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\t:\x02\x38\x01\"9\n\x0f\x45nginesResponse\x12&\n\x07\x65ngines\x18\x01 \x03(\x0b\x32\x15.docparser.EngineInfo\"|\n\nEngineInfo\x12\x0c\n\x04name\x18\x01 \x01(\t\x12\x13\n\x0b\x64\x65scription\x18\x02 \x01(\t\x12\x1c\n\x14supported_file_types\x18\x03 \x03(\t\x12\x11\n\tavailable\x18\x04 \x01(\x08\x12\x1a\n\x12unavailable_reason\x18\x05 \x01(\t2\xde\x01\n\x0e\x44ocumentParser\x12\x42\n\rParseDocument\x12\x17.docparser.ParseRequest\x1a\x18.docparser.ParseResponse\x12L\n\x13GetSupportedFormats\x12\x10.docparser.Empty\x1a#.docparser.SupportedFormatsResponse\x12:\n\nGetEngines\x12\x10.docparser.Empty\x1a\x1a.docparser.EnginesResponseB\x1aZ\x18x-agents/proto/docparserb\x06proto3')
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\x15\x64ocument_parser.proto\x12\tdocparser\"\x87\x02\n\x0cParseRequest\x12\x10\n\x08\x66ile_url\x18\x01 \x01(\t\x12\x11\n\tfile_name\x18\x02 \x01(\t\x12\x11\n\tfile_type\x18\x03 \x01(\t\x12\x15\n\rparser_engine\x18\x04 \x01(\t\x12\x46\n\x10\x65ngine_overrides\x18\x05 \x03(\x0b\x32,.docparser.ParseRequest.EngineOverridesEntry\x12(\n\nvlm_config\x18\x06 \x01(\x0b\x32\x14.docparser.VLMConfig\x1a\x36\n\x14\x45ngineOverridesEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\t:\x02\x38\x01\"p\n\tVLMConfig\x12\x0f\n\x07\x65nabled\x18\x01 \x01(\x08\x12\x10\n\x08provider\x18\x02 \x01(\t\x12\r\n\x05model\x18\x03 \x01(\t\x12\x0f\n\x07\x61pi_key\x18\x04 \x01(\t\x12\x10\n\x08\x62\x61se_url\x18\x05 \x01(\t\x12\x0e\n\x06prompt\x18\x06 \x01(\t\"\x84\x01\n\rParseResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07\x63ontent\x18\x02 \x01(\t\x12\x0f\n\x07message\x18\x03 \x01(\t\x12\x16\n\x0e\x63ontent_length\x18\x04 \x01(\x05\x12\x11\n\tfile_type\x18\x05 \x01(\t\x12\x15\n\rparser_engine\x18\x06 \x01(\t\"\x07\n\x05\x45mpty\"\xca\x01\n\x18SupportedFormatsResponse\x12\x12\n\nfile_types\x18\x01 \x03(\t\x12]\n\x16\x66ile_type_descriptions\x18\x02 \x03(\x0b\x32=.docparser.SupportedFormatsResponse.FileTypeDescriptionsEntry\x1a;\n\x19\x46ileTypeDescriptionsEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\t:\x02\x38\x01\"9\n\x0f\x45nginesResponse\x12&\n\x07\x65ngines\x18\x01 \x03(\x0b\x32\x15.docparser.EngineInfo\"|\n\nEngineInfo\x12\x0c\n\x04name\x18\x01 \x01(\t\x12\x13\n\x0b\x64\x65scription\x18\x02 \x01(\t\x12\x1c\n\x14supported_file_types\x18\x03 \x03(\t\x12\x11\n\tavailable\x18\x04 \x01(\x08\x12\x1a\n\x12unavailable_reason\x18\x05 \x01(\t2\xde\x01\n\x0e\x44ocumentParser\x12\x42\n\rParseDocument\x12\x17.docparser.ParseRequest\x1a\x18.docparser.ParseResponse\x12L\n\x13GetSupportedFormats\x12\x10.docparser.Empty\x1a#.docparser.SupportedFormatsResponse\x12:\n\nGetEngines\x12\x10.docparser.Empty\x1a\x1a.docparser.EnginesResponseB\x1aZ\x18x-agents/proto/docparserb\x06proto3')

 _globals = globals()
 _builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
@@ -37,21 +37,23 @@ if not _descriptor._USE_C_DESCRIPTORS:
  _globals['_SUPPORTEDFORMATSRESPONSE_FILETYPEDESCRIPTIONSENTRY']._loaded_options = None
  _globals['_SUPPORTEDFORMATSRESPONSE_FILETYPEDESCRIPTIONSENTRY']._serialized_options = b'8\001'
  _globals['_PARSEREQUEST']._serialized_start=37
-  _globals['_PARSEREQUEST']._serialized_end=258
-  _globals['_PARSEREQUEST_ENGINEOVERRIDESENTRY']._serialized_start=204
-  _globals['_PARSEREQUEST_ENGINEOVERRIDESENTRY']._serialized_end=258
-  _globals['_PARSERESPONSE']._serialized_start=261
-  _globals['_PARSERESPONSE']._serialized_end=393
-  _globals['_EMPTY']._serialized_start=395
-  _globals['_EMPTY']._serialized_end=402
-  _globals['_SUPPORTEDFORMATSRESPONSE']._serialized_start=405
-  _globals['_SUPPORTEDFORMATSRESPONSE']._serialized_end=607
-  _globals['_SUPPORTEDFORMATSRESPONSE_FILETYPEDESCRIPTIONSENTRY']._serialized_start=548
-  _globals['_SUPPORTEDFORMATSRESPONSE_FILETYPEDESCRIPTIONSENTRY']._serialized_end=607
-  _globals['_ENGINESRESPONSE']._serialized_start=609
-  _globals['_ENGINESRESPONSE']._serialized_end=666
-  _globals['_ENGINEINFO']._serialized_start=668
-  _globals['_ENGINEINFO']._serialized_end=792
-  _globals['_DOCUMENTPARSER']._serialized_start=795
-  _globals['_DOCUMENTPARSER']._serialized_end=1017
+  _globals['_PARSEREQUEST']._serialized_end=300
+  _globals['_PARSEREQUEST_ENGINEOVERRIDESENTRY']._serialized_start=246
+  _globals['_PARSEREQUEST_ENGINEOVERRIDESENTRY']._serialized_end=300
+  _globals['_VLMCONFIG']._serialized_start=302
+  _globals['_VLMCONFIG']._serialized_end=414
+  _globals['_PARSERESPONSE']._serialized_start=417
+  _globals['_PARSERESPONSE']._serialized_end=549
+  _globals['_EMPTY']._serialized_start=551
+  _globals['_EMPTY']._serialized_end=558
+  _globals['_SUPPORTEDFORMATSRESPONSE']._serialized_start=561
+  _globals['_SUPPORTEDFORMATSRESPONSE']._serialized_end=763
+  _globals['_SUPPORTEDFORMATSRESPONSE_FILETYPEDESCRIPTIONSENTRY']._serialized_start=704
+  _globals['_SUPPORTEDFORMATSRESPONSE_FILETYPEDESCRIPTIONSENTRY']._serialized_end=763
+  _globals['_ENGINESRESPONSE']._serialized_start=765
+  _globals['_ENGINESRESPONSE']._serialized_end=822
+  _globals['_ENGINEINFO']._serialized_start=824
+  _globals['_ENGINEINFO']._serialized_end=948
+  _globals['_DOCUMENTPARSER']._serialized_start=951
+  _globals['_DOCUMENTPARSER']._serialized_end=1173
 # @@protoc_insertion_point(module_scope)
--- a/ai-core/requirements.txt
+++ b/ai-core/requirements.txt
@@ -6,8 +6,9 @@ grpcio-tools>=1.60.0
 grpcio-reflection>=1.60.0
 protobuf>=4.25.0

-# HTTP 请求
+# 配置文件解析
+pyyaml>=6.0
 requests>=2.31.0

-# 文档解析
-markitdown>=0.0.1
+# 文档解析 - markitdown 及其所有依赖
+markitdown[pdf,docx,pptx,xlsx,all]>=0.0.1
--- a/ai-core/service/grpc_server.py
+++ b/ai-core/service/grpc_server.py
@@ -75,6 +75,21 @@ class DocumentParserServicer:
                    content_length=0,
                )

+            # 提取 VLM 配置
+            vlm_config = None
+            if hasattr(request, 'vlm_config') and request.vlm_config:
+                vlm_cfg = request.vlm_config
+                if vlm_cfg.enabled:
+                    vlm_config = {
+                        "enabled": vlm_cfg.enabled,
+                        "provider": vlm_cfg.provider,
+                        "model": vlm_cfg.model,
+                        "api_key": vlm_cfg.api_key,
+                        "base_url": vlm_cfg.base_url,
+                        "prompt": vlm_cfg.prompt,
+                    }
+                    logger.info(f"VLM config: provider={vlm_cfg.provider}, model={vlm_cfg.model}")
+
            logger.info("Downloading file from URL: %s", file_url)

            try:
@@ -95,9 +110,9 @@ class DocumentParserServicer:
                    content_length=0,
                )

-            logger.info("Parsing file with MarkItDown")
+            logger.info("Parsing file with MarkItDown + VLM")

-            result = self.parser.parse_bytes(content, file_name)
+            result = self.parser.parse_bytes(content, file_name, vlm_config=vlm_config)

            if not result.get("success", False):
                logger.warning("Parser returned failure: %s", result.get("error", "Unknown error"))
--- a/server/cmd/api/main.go
+++ b/server/cmd/api/main.go
@@ -87,7 +87,7 @@ func main() {
 	if err != nil {
 		log.Printf("Warning: Failed to initialize upload service: %v (files will not be available)", err)
 	}
-	knowledgeService := service.NewKnowledgeService(knowledgeRepo, modelRepo, uploadService, cfg.PythonServiceURL, cfg.AICoreServiceAddr)
+	knowledgeService := service.NewKnowledgeService(knowledgeRepo, modelRepo, uploadService, cfg.PythonServiceURL, cfg.AICoreServiceAddr, cfg.MarkdownLocalPath)

 	// 6. 初始化 Handler
 	dbHandler := handler.NewDatabaseHandler(dbService)
--- a/server/document_parser.pb.go
+++ b/server/document_parser.pb.go
@@ -0,0 +1,498 @@
+// Code generated by protoc-gen-go. DO NOT EDIT.
+// versions:
+// 	protoc-gen-go v1.36.11
+// 	protoc        v5.29.3
+// source: document_parser.proto
+
+package docparser
+
+import (
+	protoreflect "google.golang.org/protobuf/reflect/protoreflect"
+	protoimpl "google.golang.org/protobuf/runtime/protoimpl"
+	reflect "reflect"
+	sync "sync"
+	unsafe "unsafe"
+)
+
+const (
+	// Verify that this generated code is sufficiently up-to-date.
+	_ = protoimpl.EnforceVersion(20 - protoimpl.MinVersion)
+	// Verify that runtime/protoimpl is sufficiently up-to-date.
+	_ = protoimpl.EnforceVersion(protoimpl.MaxVersion - 20)
+)
+
+type ParseRequest struct {
+	state           protoimpl.MessageState `protogen:"open.v1"`
+	FileUrl         string                 `protobuf:"bytes,1,opt,name=file_url,json=fileUrl,proto3" json:"file_url,omitempty"`
+	FileName        string                 `protobuf:"bytes,2,opt,name=file_name,json=fileName,proto3" json:"file_name,omitempty"`
+	FileType        string                 `protobuf:"bytes,3,opt,name=file_type,json=fileType,proto3" json:"file_type,omitempty"`
+	ParserEngine    string                 `protobuf:"bytes,4,opt,name=parser_engine,json=parserEngine,proto3" json:"parser_engine,omitempty"`
+	EngineOverrides map[string]string      `protobuf:"bytes,5,rep,name=engine_overrides,json=engineOverrides,proto3" json:"engine_overrides,omitempty" protobuf_key:"bytes,1,opt,name=key" protobuf_val:"bytes,2,opt,name=value"`
+	unknownFields   protoimpl.UnknownFields
+	sizeCache       protoimpl.SizeCache
+}
+
+func (x *ParseRequest) Reset() {
+	*x = ParseRequest{}
+	mi := &file_document_parser_proto_msgTypes[0]
+	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+	ms.StoreMessageInfo(mi)
+}
+
+func (x *ParseRequest) String() string {
+	return protoimpl.X.MessageStringOf(x)
+}
+
+func (*ParseRequest) ProtoMessage() {}
+
+func (x *ParseRequest) ProtoReflect() protoreflect.Message {
+	mi := &file_document_parser_proto_msgTypes[0]
+	if x != nil {
+		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+		if ms.LoadMessageInfo() == nil {
+			ms.StoreMessageInfo(mi)
+		}
+		return ms
+	}
+	return mi.MessageOf(x)
+}
+
+// Deprecated: Use ParseRequest.ProtoReflect.Descriptor instead.
+func (*ParseRequest) Descriptor() ([]byte, []int) {
+	return file_document_parser_proto_rawDescGZIP(), []int{0}
+}
+
+func (x *ParseRequest) GetFileUrl() string {
+	if x != nil {
+		return x.FileUrl
+	}
+	return ""
+}
+
+func (x *ParseRequest) GetFileName() string {
+	if x != nil {
+		return x.FileName
+	}
+	return ""
+}
+
+func (x *ParseRequest) GetFileType() string {
+	if x != nil {
+		return x.FileType
+	}
+	return ""
+}
+
+func (x *ParseRequest) GetParserEngine() string {
+	if x != nil {
+		return x.ParserEngine
+	}
+	return ""
+}
+
+func (x *ParseRequest) GetEngineOverrides() map[string]string {
+	if x != nil {
+		return x.EngineOverrides
+	}
+	return nil
+}
+
+type ParseResponse struct {
+	state         protoimpl.MessageState `protogen:"open.v1"`
+	Success       bool                   `protobuf:"varint,1,opt,name=success,proto3" json:"success,omitempty"`
+	Content       string                 `protobuf:"bytes,2,opt,name=content,proto3" json:"content,omitempty"`
+	Message       string                 `protobuf:"bytes,3,opt,name=message,proto3" json:"message,omitempty"`
+	ContentLength int32                  `protobuf:"varint,4,opt,name=content_length,json=contentLength,proto3" json:"content_length,omitempty"`
+	FileType      string                 `protobuf:"bytes,5,opt,name=file_type,json=fileType,proto3" json:"file_type,omitempty"`
+	ParserEngine  string                 `protobuf:"bytes,6,opt,name=parser_engine,json=parserEngine,proto3" json:"parser_engine,omitempty"`
+	unknownFields protoimpl.UnknownFields
+	sizeCache     protoimpl.SizeCache
+}
+
+func (x *ParseResponse) Reset() {
+	*x = ParseResponse{}
+	mi := &file_document_parser_proto_msgTypes[1]
+	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+	ms.StoreMessageInfo(mi)
+}
+
+func (x *ParseResponse) String() string {
+	return protoimpl.X.MessageStringOf(x)
+}
+
+func (*ParseResponse) ProtoMessage() {}
+
+func (x *ParseResponse) ProtoReflect() protoreflect.Message {
+	mi := &file_document_parser_proto_msgTypes[1]
+	if x != nil {
+		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+		if ms.LoadMessageInfo() == nil {
+			ms.StoreMessageInfo(mi)
+		}
+		return ms
+	}
+	return mi.MessageOf(x)
+}
+
+// Deprecated: Use ParseResponse.ProtoReflect.Descriptor instead.
+func (*ParseResponse) Descriptor() ([]byte, []int) {
+	return file_document_parser_proto_rawDescGZIP(), []int{1}
+}
+
+func (x *ParseResponse) GetSuccess() bool {
+	if x != nil {
+		return x.Success
+	}
+	return false
+}
+
+func (x *ParseResponse) GetContent() string {
+	if x != nil {
+		return x.Content
+	}
+	return ""
+}
+
+func (x *ParseResponse) GetMessage() string {
+	if x != nil {
+		return x.Message
+	}
+	return ""
+}
+
+func (x *ParseResponse) GetContentLength() int32 {
+	if x != nil {
+		return x.ContentLength
+	}
+	return 0
+}
+
+func (x *ParseResponse) GetFileType() string {
+	if x != nil {
+		return x.FileType
+	}
+	return ""
+}
+
+func (x *ParseResponse) GetParserEngine() string {
+	if x != nil {
+		return x.ParserEngine
+	}
+	return ""
+}
+
+type Empty struct {
+	state         protoimpl.MessageState `protogen:"open.v1"`
+	unknownFields protoimpl.UnknownFields
+	sizeCache     protoimpl.SizeCache
+}
+
+func (x *Empty) Reset() {
+	*x = Empty{}
+	mi := &file_document_parser_proto_msgTypes[2]
+	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+	ms.StoreMessageInfo(mi)
+}
+
+func (x *Empty) String() string {
+	return protoimpl.X.MessageStringOf(x)
+}
+
+func (*Empty) ProtoMessage() {}
+
+func (x *Empty) ProtoReflect() protoreflect.Message {
+	mi := &file_document_parser_proto_msgTypes[2]
+	if x != nil {
+		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+		if ms.LoadMessageInfo() == nil {
+			ms.StoreMessageInfo(mi)
+		}
+		return ms
+	}
+	return mi.MessageOf(x)
+}
+
+// Deprecated: Use Empty.ProtoReflect.Descriptor instead.
+func (*Empty) Descriptor() ([]byte, []int) {
+	return file_document_parser_proto_rawDescGZIP(), []int{2}
+}
+
+type SupportedFormatsResponse struct {
+	state                protoimpl.MessageState `protogen:"open.v1"`
+	FileTypes            []string               `protobuf:"bytes,1,rep,name=file_types,json=fileTypes,proto3" json:"file_types,omitempty"`
+	FileTypeDescriptions map[string]string      `protobuf:"bytes,2,rep,name=file_type_descriptions,json=fileTypeDescriptions,proto3" json:"file_type_descriptions,omitempty" protobuf_key:"bytes,1,opt,name=key" protobuf_val:"bytes,2,opt,name=value"`
+	unknownFields        protoimpl.UnknownFields
+	sizeCache            protoimpl.SizeCache
+}
+
+func (x *SupportedFormatsResponse) Reset() {
+	*x = SupportedFormatsResponse{}
+	mi := &file_document_parser_proto_msgTypes[3]
+	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+	ms.StoreMessageInfo(mi)
+}
+
+func (x *SupportedFormatsResponse) String() string {
+	return protoimpl.X.MessageStringOf(x)
+}
+
+func (*SupportedFormatsResponse) ProtoMessage() {}
+
+func (x *SupportedFormatsResponse) ProtoReflect() protoreflect.Message {
+	mi := &file_document_parser_proto_msgTypes[3]
+	if x != nil {
+		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+		if ms.LoadMessageInfo() == nil {
+			ms.StoreMessageInfo(mi)
+		}
+		return ms
+	}
+	return mi.MessageOf(x)
+}
+
+// Deprecated: Use SupportedFormatsResponse.ProtoReflect.Descriptor instead.
+func (*SupportedFormatsResponse) Descriptor() ([]byte, []int) {
+	return file_document_parser_proto_rawDescGZIP(), []int{3}
+}
+
+func (x *SupportedFormatsResponse) GetFileTypes() []string {
+	if x != nil {
+		return x.FileTypes
+	}
+	return nil
+}
+
+func (x *SupportedFormatsResponse) GetFileTypeDescriptions() map[string]string {
+	if x != nil {
+		return x.FileTypeDescriptions
+	}
+	return nil
+}
+
+type EnginesResponse struct {
+	state         protoimpl.MessageState `protogen:"open.v1"`
+	Engines       []*EngineInfo          `protobuf:"bytes,1,rep,name=engines,proto3" json:"engines,omitempty"`
+	unknownFields protoimpl.UnknownFields
+	sizeCache     protoimpl.SizeCache
+}
+
+func (x *EnginesResponse) Reset() {
+	*x = EnginesResponse{}
+	mi := &file_document_parser_proto_msgTypes[4]
+	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+	ms.StoreMessageInfo(mi)
+}
+
+func (x *EnginesResponse) String() string {
+	return protoimpl.X.MessageStringOf(x)
+}
+
+func (*EnginesResponse) ProtoMessage() {}
+
+func (x *EnginesResponse) ProtoReflect() protoreflect.Message {
+	mi := &file_document_parser_proto_msgTypes[4]
+	if x != nil {
+		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+		if ms.LoadMessageInfo() == nil {
+			ms.StoreMessageInfo(mi)
+		}
+		return ms
+	}
+	return mi.MessageOf(x)
+}
+
+// Deprecated: Use EnginesResponse.ProtoReflect.Descriptor instead.
+func (*EnginesResponse) Descriptor() ([]byte, []int) {
+	return file_document_parser_proto_rawDescGZIP(), []int{4}
+}
+
+func (x *EnginesResponse) GetEngines() []*EngineInfo {
+	if x != nil {
+		return x.Engines
+	}
+	return nil
+}
+
+type EngineInfo struct {
+	state              protoimpl.MessageState `protogen:"open.v1"`
+	Name               string                 `protobuf:"bytes,1,opt,name=name,proto3" json:"name,omitempty"`
+	Description        string                 `protobuf:"bytes,2,opt,name=description,proto3" json:"description,omitempty"`
+	SupportedFileTypes []string               `protobuf:"bytes,3,rep,name=supported_file_types,json=supportedFileTypes,proto3" json:"supported_file_types,omitempty"`
+	Available          bool                   `protobuf:"varint,4,opt,name=available,proto3" json:"available,omitempty"`
+	UnavailableReason  string                 `protobuf:"bytes,5,opt,name=unavailable_reason,json=unavailableReason,proto3" json:"unavailable_reason,omitempty"`
+	unknownFields      protoimpl.UnknownFields
+	sizeCache          protoimpl.SizeCache
+}
+
+func (x *EngineInfo) Reset() {
+	*x = EngineInfo{}
+	mi := &file_document_parser_proto_msgTypes[5]
+	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+	ms.StoreMessageInfo(mi)
+}
+
+func (x *EngineInfo) String() string {
+	return protoimpl.X.MessageStringOf(x)
+}
+
+func (*EngineInfo) ProtoMessage() {}
+
+func (x *EngineInfo) ProtoReflect() protoreflect.Message {
+	mi := &file_document_parser_proto_msgTypes[5]
+	if x != nil {
+		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+		if ms.LoadMessageInfo() == nil {
+			ms.StoreMessageInfo(mi)
+		}
+		return ms
+	}
+	return mi.MessageOf(x)
+}
+
+// Deprecated: Use EngineInfo.ProtoReflect.Descriptor instead.
+func (*EngineInfo) Descriptor() ([]byte, []int) {
+	return file_document_parser_proto_rawDescGZIP(), []int{5}
+}
+
+func (x *EngineInfo) GetName() string {
+	if x != nil {
+		return x.Name
+	}
+	return ""
+}
+
+func (x *EngineInfo) GetDescription() string {
+	if x != nil {
+		return x.Description
+	}
+	return ""
+}
+
+func (x *EngineInfo) GetSupportedFileTypes() []string {
+	if x != nil {
+		return x.SupportedFileTypes
+	}
+	return nil
+}
+
+func (x *EngineInfo) GetAvailable() bool {
+	if x != nil {
+		return x.Available
+	}
+	return false
+}
+
+func (x *EngineInfo) GetUnavailableReason() string {
+	if x != nil {
+		return x.UnavailableReason
+	}
+	return ""
+}
+
+var File_document_parser_proto protoreflect.FileDescriptor
+
+const file_document_parser_proto_rawDesc = "" +
+	"\n" +
+	"\x15document_parser.proto\x12\tdocparser\"\xa5\x02\n" +
+	"\fParseRequest\x12\x19\n" +
+	"\bfile_url\x18\x01 \x01(\tR\afileUrl\x12\x1b\n" +
+	"\tfile_name\x18\x02 \x01(\tR\bfileName\x12\x1b\n" +
+	"\tfile_type\x18\x03 \x01(\tR\bfileType\x12#\n" +
+	"\rparser_engine\x18\x04 \x01(\tR\fparserEngine\x12W\n" +
+	"\x10engine_overrides\x18\x05 \x03(\v2,.docparser.ParseRequest.EngineOverridesEntryR\x0fengineOverrides\x1aB\n" +
+	"\x14EngineOverridesEntry\x12\x10\n" +
+	"\x03key\x18\x01 \x01(\tR\x03key\x12\x14\n" +
+	"\x05value\x18\x02 \x01(\tR\x05value:\x028\x01\"\xc6\x01\n" +
+	"\rParseResponse\x12\x18\n" +
+	"\asuccess\x18\x01 \x01(\bR\asuccess\x12\x18\n" +
+	"\acontent\x18\x02 \x01(\tR\acontent\x12\x18\n" +
+	"\amessage\x18\x03 \x01(\tR\amessage\x12%\n" +
+	"\x0econtent_length\x18\x04 \x01(\x05R\rcontentLength\x12\x1b\n" +
+	"\tfile_type\x18\x05 \x01(\tR\bfileType\x12#\n" +
+	"\rparser_engine\x18\x06 \x01(\tR\fparserEngine\"\a\n" +
+	"\x05Empty\"\xf7\x01\n" +
+	"\x18SupportedFormatsResponse\x12\x1d\n" +
+	"\n" +
+	"file_types\x18\x01 \x03(\tR\tfileTypes\x12s\n" +
+	"\x16file_type_descriptions\x18\x02 \x03(\v2=.docparser.SupportedFormatsResponse.FileTypeDescriptionsEntryR\x14fileTypeDescriptions\x1aG\n" +
+	"\x19FileTypeDescriptionsEntry\x12\x10\n" +
+	"\x03key\x18\x01 \x01(\tR\x03key\x12\x14\n" +
+	"\x05value\x18\x02 \x01(\tR\x05value:\x028\x01\"B\n" +
+	"\x0fEnginesResponse\x12/\n" +
+	"\aengines\x18\x01 \x03(\v2\x15.docparser.EngineInfoR\aengines\"\xc1\x01\n" +
+	"\n" +
+	"EngineInfo\x12\x12\n" +
+	"\x04name\x18\x01 \x01(\tR\x04name\x12 \n" +
+	"\vdescription\x18\x02 \x01(\tR\vdescription\x120\n" +
+	"\x14supported_file_types\x18\x03 \x03(\tR\x12supportedFileTypes\x12\x1c\n" +
+	"\tavailable\x18\x04 \x01(\bR\tavailable\x12-\n" +
+	"\x12unavailable_reason\x18\x05 \x01(\tR\x11unavailableReason2\xde\x01\n" +
+	"\x0eDocumentParser\x12B\n" +
+	"\rParseDocument\x12\x17.docparser.ParseRequest\x1a\x18.docparser.ParseResponse\x12L\n" +
+	"\x13GetSupportedFormats\x12\x10.docparser.Empty\x1a#.docparser.SupportedFormatsResponse\x12:\n" +
+	"\n" +
+	"GetEngines\x12\x10.docparser.Empty\x1a\x1a.docparser.EnginesResponseB\x1aZ\x18x-agents/proto/docparserb\x06proto3"
+
+var (
+	file_document_parser_proto_rawDescOnce sync.Once
+	file_document_parser_proto_rawDescData []byte
+)
+
+func file_document_parser_proto_rawDescGZIP() []byte {
+	file_document_parser_proto_rawDescOnce.Do(func() {
+		file_document_parser_proto_rawDescData = protoimpl.X.CompressGZIP(unsafe.Slice(unsafe.StringData(file_document_parser_proto_rawDesc), len(file_document_parser_proto_rawDesc)))
+	})
+	return file_document_parser_proto_rawDescData
+}
+
+var file_document_parser_proto_msgTypes = make([]protoimpl.MessageInfo, 8)
+var file_document_parser_proto_goTypes = []any{
+	(*ParseRequest)(nil),             // 0: docparser.ParseRequest
+	(*ParseResponse)(nil),            // 1: docparser.ParseResponse
+	(*Empty)(nil),                    // 2: docparser.Empty
+	(*SupportedFormatsResponse)(nil), // 3: docparser.SupportedFormatsResponse
+	(*EnginesResponse)(nil),          // 4: docparser.EnginesResponse
+	(*EngineInfo)(nil),               // 5: docparser.EngineInfo
+	nil,                              // 6: docparser.ParseRequest.EngineOverridesEntry
+	nil,                              // 7: docparser.SupportedFormatsResponse.FileTypeDescriptionsEntry
+}
+var file_document_parser_proto_depIdxs = []int32{
+	6, // 0: docparser.ParseRequest.engine_overrides:type_name -> docparser.ParseRequest.EngineOverridesEntry
+	7, // 1: docparser.SupportedFormatsResponse.file_type_descriptions:type_name -> docparser.SupportedFormatsResponse.FileTypeDescriptionsEntry
+	5, // 2: docparser.EnginesResponse.engines:type_name -> docparser.EngineInfo
+	0, // 3: docparser.DocumentParser.ParseDocument:input_type -> docparser.ParseRequest
+	2, // 4: docparser.DocumentParser.GetSupportedFormats:input_type -> docparser.Empty
+	2, // 5: docparser.DocumentParser.GetEngines:input_type -> docparser.Empty
+	1, // 6: docparser.DocumentParser.ParseDocument:output_type -> docparser.ParseResponse
+	3, // 7: docparser.DocumentParser.GetSupportedFormats:output_type -> docparser.SupportedFormatsResponse
+	4, // 8: docparser.DocumentParser.GetEngines:output_type -> docparser.EnginesResponse
+	6, // [6:9] is the sub-list for method output_type
+	3, // [3:6] is the sub-list for method input_type
+	3, // [3:3] is the sub-list for extension type_name
+	3, // [3:3] is the sub-list for extension extendee
+	0, // [0:3] is the sub-list for field type_name
+}
+
+func init() { file_document_parser_proto_init() }
+func file_document_parser_proto_init() {
+	if File_document_parser_proto != nil {
+		return
+	}
+	type x struct{}
+	out := protoimpl.TypeBuilder{
+		File: protoimpl.DescBuilder{
+			GoPackagePath: reflect.TypeOf(x{}).PkgPath(),
+			RawDescriptor: unsafe.Slice(unsafe.StringData(file_document_parser_proto_rawDesc), len(file_document_parser_proto_rawDesc)),
+			NumEnums:      0,
+			NumMessages:   8,
+			NumExtensions: 0,
+			NumServices:   1,
+		},
+		GoTypes:           file_document_parser_proto_goTypes,
+		DependencyIndexes: file_document_parser_proto_depIdxs,
+		MessageInfos:      file_document_parser_proto_msgTypes,
+	}.Build()
+	File_document_parser_proto = out.File
+	file_document_parser_proto_goTypes = nil
+	file_document_parser_proto_depIdxs = nil
+}
--- a/server/document_parser_grpc.pb.go
+++ b/server/document_parser_grpc.pb.go
@@ -0,0 +1,197 @@
+// Code generated by protoc-gen-go-grpc. DO NOT EDIT.
+// versions:
+// - protoc-gen-go-grpc v1.6.1
+// - protoc             v5.29.3
+// source: document_parser.proto
+
+package docparser
+
+import (
+	context "context"
+	grpc "google.golang.org/grpc"
+	codes "google.golang.org/grpc/codes"
+	status "google.golang.org/grpc/status"
+)
+
+// This is a compile-time assertion to ensure that this generated file
+// is compatible with the grpc package it is being compiled against.
+// Requires gRPC-Go v1.64.0 or later.
+const _ = grpc.SupportPackageIsVersion9
+
+const (
+	DocumentParser_ParseDocument_FullMethodName       = "/docparser.DocumentParser/ParseDocument"
+	DocumentParser_GetSupportedFormats_FullMethodName = "/docparser.DocumentParser/GetSupportedFormats"
+	DocumentParser_GetEngines_FullMethodName          = "/docparser.DocumentParser/GetEngines"
+)
+
+// DocumentParserClient is the client API for DocumentParser service.
+//
+// For semantics around ctx use and closing/ending streaming RPCs, please refer to https://pkg.go.dev/google.golang.org/grpc/?tab=doc#ClientConn.NewStream.
+type DocumentParserClient interface {
+	ParseDocument(ctx context.Context, in *ParseRequest, opts ...grpc.CallOption) (*ParseResponse, error)
+	GetSupportedFormats(ctx context.Context, in *Empty, opts ...grpc.CallOption) (*SupportedFormatsResponse, error)
+	GetEngines(ctx context.Context, in *Empty, opts ...grpc.CallOption) (*EnginesResponse, error)
+}
+
+type documentParserClient struct {
+	cc grpc.ClientConnInterface
+}
+
+func NewDocumentParserClient(cc grpc.ClientConnInterface) DocumentParserClient {
+	return &documentParserClient{cc}
+}
+
+func (c *documentParserClient) ParseDocument(ctx context.Context, in *ParseRequest, opts ...grpc.CallOption) (*ParseResponse, error) {
+	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
+	out := new(ParseResponse)
+	err := c.cc.Invoke(ctx, DocumentParser_ParseDocument_FullMethodName, in, out, cOpts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *documentParserClient) GetSupportedFormats(ctx context.Context, in *Empty, opts ...grpc.CallOption) (*SupportedFormatsResponse, error) {
+	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
+	out := new(SupportedFormatsResponse)
+	err := c.cc.Invoke(ctx, DocumentParser_GetSupportedFormats_FullMethodName, in, out, cOpts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+func (c *documentParserClient) GetEngines(ctx context.Context, in *Empty, opts ...grpc.CallOption) (*EnginesResponse, error) {
+	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
+	out := new(EnginesResponse)
+	err := c.cc.Invoke(ctx, DocumentParser_GetEngines_FullMethodName, in, out, cOpts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
+// DocumentParserServer is the server API for DocumentParser service.
+// All implementations must embed UnimplementedDocumentParserServer
+// for forward compatibility.
+type DocumentParserServer interface {
+	ParseDocument(context.Context, *ParseRequest) (*ParseResponse, error)
+	GetSupportedFormats(context.Context, *Empty) (*SupportedFormatsResponse, error)
+	GetEngines(context.Context, *Empty) (*EnginesResponse, error)
+	mustEmbedUnimplementedDocumentParserServer()
+}
+
+// UnimplementedDocumentParserServer must be embedded to have
+// forward compatible implementations.
+//
+// NOTE: this should be embedded by value instead of pointer to avoid a nil
+// pointer dereference when methods are called.
+type UnimplementedDocumentParserServer struct{}
+
+func (UnimplementedDocumentParserServer) ParseDocument(context.Context, *ParseRequest) (*ParseResponse, error) {
+	return nil, status.Error(codes.Unimplemented, "method ParseDocument not implemented")
+}
+func (UnimplementedDocumentParserServer) GetSupportedFormats(context.Context, *Empty) (*SupportedFormatsResponse, error) {
+	return nil, status.Error(codes.Unimplemented, "method GetSupportedFormats not implemented")
+}
+func (UnimplementedDocumentParserServer) GetEngines(context.Context, *Empty) (*EnginesResponse, error) {
+	return nil, status.Error(codes.Unimplemented, "method GetEngines not implemented")
+}
+func (UnimplementedDocumentParserServer) mustEmbedUnimplementedDocumentParserServer() {}
+func (UnimplementedDocumentParserServer) testEmbeddedByValue()                        {}
+
+// UnsafeDocumentParserServer may be embedded to opt out of forward compatibility for this service.
+// Use of this interface is not recommended, as added methods to DocumentParserServer will
+// result in compilation errors.
+type UnsafeDocumentParserServer interface {
+	mustEmbedUnimplementedDocumentParserServer()
+}
+
+func RegisterDocumentParserServer(s grpc.ServiceRegistrar, srv DocumentParserServer) {
+	// If the following call panics, it indicates UnimplementedDocumentParserServer was
+	// embedded by pointer and is nil.  This will cause panics if an
+	// unimplemented method is ever invoked, so we test this at initialization
+	// time to prevent it from happening at runtime later due to I/O.
+	if t, ok := srv.(interface{ testEmbeddedByValue() }); ok {
+		t.testEmbeddedByValue()
+	}
+	s.RegisterService(&DocumentParser_ServiceDesc, srv)
+}
+
+func _DocumentParser_ParseDocument_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(ParseRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(DocumentParserServer).ParseDocument(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: DocumentParser_ParseDocument_FullMethodName,
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(DocumentParserServer).ParseDocument(ctx, req.(*ParseRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _DocumentParser_GetSupportedFormats_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(Empty)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(DocumentParserServer).GetSupportedFormats(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: DocumentParser_GetSupportedFormats_FullMethodName,
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(DocumentParserServer).GetSupportedFormats(ctx, req.(*Empty))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+func _DocumentParser_GetEngines_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(Empty)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(DocumentParserServer).GetEngines(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: DocumentParser_GetEngines_FullMethodName,
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(DocumentParserServer).GetEngines(ctx, req.(*Empty))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
+// DocumentParser_ServiceDesc is the grpc.ServiceDesc for DocumentParser service.
+// It's only intended for direct use with grpc.RegisterService,
+// and not to be introspected or modified (even as a copy)
+var DocumentParser_ServiceDesc = grpc.ServiceDesc{
+	ServiceName: "docparser.DocumentParser",
+	HandlerType: (*DocumentParserServer)(nil),
+	Methods: []grpc.MethodDesc{
+		{
+			MethodName: "ParseDocument",
+			Handler:    _DocumentParser_ParseDocument_Handler,
+		},
+		{
+			MethodName: "GetSupportedFormats",
+			Handler:    _DocumentParser_GetSupportedFormats_Handler,
+		},
+		{
+			MethodName: "GetEngines",
+			Handler:    _DocumentParser_GetEngines_Handler,
+		},
+	},
+	Streams:  []grpc.StreamDesc{},
+	Metadata: "document_parser.proto",
+}
--- a/server/internal/config/config.go
+++ b/server/internal/config/config.go
@@ -23,9 +23,10 @@ type Config struct {
 	PythonServiceURL  string
 	AICoreServiceAddr string // AI-Core gRPC 服务地址，如 "localhost:50051"
 	// 文件上传配置
-	UploadMode      string // "local" 或 "minio"
-	UploadLocalPath string // 本地存储路径，如 "resource/files"
-	ServerBaseURL   string // 服务器基础 URL，用于生成本地文件 URL
+	UploadMode        string // "local" 或 "minio"
+	UploadLocalPath   string // 本地存储路径，如 "resource/files"
+	ServerBaseURL     string // 服务器基础 URL，用于生成本地文件 URL
+	MarkdownLocalPath string // Markdown 文件存储路径，如 "resource/markdown"
 	// MinIO 配置
 	MinIOEndpoint   string
 	MinIOAccessKey  string
@@ -56,6 +57,7 @@ func Load() *Config {
 	viper.SetDefault("upload_mode", "local")
 	viper.SetDefault("upload_local_path", "resource/files")
 	viper.SetDefault("server_base_url", "http://localhost:8080")
+	viper.SetDefault("markdown_local_path", "resource/markdown")
 	viper.SetDefault("minio_endpoint", "localhost:9000")
 	viper.SetDefault("minio_access_key", "")
 	viper.SetDefault("minio_secret_key", "")
@@ -87,9 +89,10 @@ func Load() *Config {
 		PythonServiceURL:  viper.GetString("python_service_url"),
 		AICoreServiceAddr:  viper.GetString("ai_core_service_addr"),
 		// 文件上传配置
-		UploadMode:      viper.GetString("upload_mode"),
-		UploadLocalPath: viper.GetString("upload_local_path"),
-		ServerBaseURL:   viper.GetString("server_base_url"),
+		UploadMode:        viper.GetString("upload_mode"),
+		UploadLocalPath:   viper.GetString("upload_local_path"),
+		ServerBaseURL:     viper.GetString("server_base_url"),
+		MarkdownLocalPath: viper.GetString("markdown_local_path"),
 		// MinIO 配置
 		MinIOEndpoint:  viper.GetString("minio_endpoint"),
 		MinIOAccessKey: viper.GetString("minio_access_key"),
--- a/server/internal/model/knowledge_info.go
+++ b/server/internal/model/knowledge_info.go
@@ -145,7 +145,8 @@ type UploadDocumentResponse struct {

 // DocumentPreviewResponse 文档预览响应
 type DocumentPreviewResponse struct {
-	TotalPages int    `json:"total_pages"`
-	CurrentPage int   `json:"current_page"`
-	Content    string `json:"content"`
+	TotalPages  int    `json:"total_pages"`
+	CurrentPage int    `json:"current_page"`
+	Content     string `json:"content"`
+	ContentType string `json:"content_type"` // url: 文件URL, html: HTML内容
 }
--- a/server/internal/service/document_parser_client.go
+++ b/server/internal/service/document_parser_client.go
@@ -7,6 +7,8 @@ import (

 	"google.golang.org/grpc"
 	"google.golang.org/grpc/credentials/insecure"
+
+	docparser "x-agents/server"
 )

 // AICoreClient AI-Core 文档解析服务客户端
@@ -53,7 +55,7 @@ func (c *AICoreClient) Close() {
 	}
 }

-// ParseDocument 解析文档
+// ParseDocument 解析文档 - 使用生成的 protobuf 代码
 func (c *AICoreClient) ParseDocument(fileURL, fileName, fileType string) (*ParseResult, error) {
 	if c.conn == nil {
 		if err := c.Connect(); err != nil {
@@ -61,17 +63,16 @@ func (c *AICoreClient) ParseDocument(fileURL, fileName, fileType string) (*Parse
 		}
 	}

-	// 使用 gRPC raw bytes 调用
-	// 由于没有生成 protobuf 代码，使用 raw bytes 方式调用
-	client := NewDocumentParserClient(c.conn)
+	// 使用生成的 protobuf 客户端
+	client := docparser.NewDocumentParserClient(c.conn)

-	req := &ParseRequest{
+	req := &docparser.ParseRequest{
 		FileUrl:  fileURL,
 		FileName: fileName,
 		FileType: fileType,
 	}

-	ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)
+	ctx, cancel := context.WithTimeout(context.Background(), 120*time.Second)
 	defer cancel()

 	resp, err := client.ParseDocument(ctx, req)
@@ -80,53 +81,11 @@ func (c *AICoreClient) ParseDocument(fileURL, fileName, fileType string) (*Parse
 	}

 	return &ParseResult{
-		Success:       resp.Success,
-		Content:       resp.Content,
-		Message:       resp.Message,
-		ContentLength: resp.ContentLength,
-		FileType:      resp.FileType,
-		ParserEngine:  resp.ParserEngine,
+		Success:       resp.GetSuccess(),
+		Content:       resp.GetContent(),
+		Message:       resp.GetMessage(),
+		ContentLength: resp.GetContentLength(),
+		FileType:      resp.GetFileType(),
+		ParserEngine:  resp.GetParserEngine(),
 	}, nil
 }
-
-// 以下是手动定义的 protobuf messages（与 proto 文件一致）
-// 不需要生成 .pb.go 文件，直接手动定义
-
-type ParseRequest struct {
-	FileUrl       string
-	FileName     string
-	FileType     string
-	ParserEngine string
-}
-
-type ParseResponse struct {
-	Success       bool
-	Content       string
-	Message       string
-	ContentLength int32
-	FileType      string
-	ParserEngine  string
-}
-
-// DocumentParserClient gRPC 客户端接口（手动实现）
-type DocumentParserClient interface {
-	ParseDocument(ctx context.Context, in *ParseRequest, opts ...grpc.CallOption) (*ParseResponse, error)
-}
-
-type documentParserClient struct {
-	cc grpc.ClientConnInterface
-}
-
-// NewDocumentParserClient 创建 DocumentParser 客户端
-func NewDocumentParserClient(cc grpc.ClientConnInterface) DocumentParserClient {
-	return &documentParserClient{cc: cc}
-}
-
-func (c *documentParserClient) ParseDocument(ctx context.Context, in *ParseRequest, opts ...grpc.CallOption) (*ParseResponse, error) {
-	out := new(ParseResponse)
-	err := c.cc.Invoke(ctx, "/docparser.DocumentParser/ParseDocument", in, out, opts...)
-	if err != nil {
-		return nil, err
-	}
-	return out, nil
-}
--- a/server/internal/service/knowledge_service.go
+++ b/server/internal/service/knowledge_service.go
@@ -29,9 +29,10 @@ type KnowledgeService struct {
 	uploadService    *UploadService
 	pythonServiceURL string
 	aiCoreClient     *AICoreClient
+	markdownLocalPath string // Markdown 本地存储路径
 }

-func NewKnowledgeService(repo *repository.KnowledgeRepository, modelRepo *repository.ModelRepository, uploadService *UploadService, pythonServiceURL, aiCoreServiceAddr string) *KnowledgeService {
+func NewKnowledgeService(repo *repository.KnowledgeRepository, modelRepo *repository.ModelRepository, uploadService *UploadService, pythonServiceURL, aiCoreServiceAddr, markdownLocalPath string) *KnowledgeService {
 	aiCoreClient, _ := NewAICoreClient(aiCoreServiceAddr)
 	return &KnowledgeService{
 		repo:             repo,
@@ -39,6 +40,7 @@ func NewKnowledgeService(repo *repository.KnowledgeRepository, modelRepo *reposi
 		uploadService:    uploadService,
 		pythonServiceURL: pythonServiceURL,
 		aiCoreClient:     aiCoreClient,
+		markdownLocalPath: markdownLocalPath,
 	}
 }

@@ -307,6 +309,13 @@ func (s *KnowledgeService) parseDocumentWithAICore(docID, fileURL, fileName stri

 	if result.Success && result.Content != "" {
 		knowledgeDebugLog.Printf("[AICore] 解析成功: docID=%s, contentLength=%d", docID, len(result.Content))
+
+		// 保存到本地文件
+		markdownPath := s.saveMarkdownToFile(docID, fileName, result.Content)
+		if markdownPath != "" {
+			knowledgeDebugLog.Printf("[AICore] Markdown 保存到本地: docID=%s, path=%s", docID, markdownPath)
+		}
+
 		// 更新文档的 Content 字段
 		s.repo.UpdateDocument(docID, map[string]interface{}{
 			"content": result.Content,
@@ -316,6 +325,31 @@ func (s *KnowledgeService) parseDocumentWithAICore(docID, fileURL, fileName stri
 	}
 }

+// saveMarkdownToFile 保存 Markdown 内容到本地文件
+func (s *KnowledgeService) saveMarkdownToFile(docID, fileName, content string) string {
+	if s.markdownLocalPath == "" {
+		s.markdownLocalPath = "resource/markdown"
+	}
+
+	// 创建目录
+	if err := os.MkdirAll(s.markdownLocalPath, 0755); err != nil {
+		knowledgeDebugLog.Printf("[AICore] 创建目录失败: path=%s, err=%v", s.markdownLocalPath, err)
+		return ""
+	}
+
+	// 生成文件名（用 docID + .md）
+	markdownFileName := docID + ".md"
+	markdownPath := s.markdownLocalPath + "/" + markdownFileName
+
+	// 写入文件
+	if err := os.WriteFile(markdownPath, []byte(content), 0644); err != nil {
+		knowledgeDebugLog.Printf("[AICore] 保存 Markdown 失败: path=%s, err=%v", markdownPath, err)
+		return ""
+	}
+
+	return markdownPath
+}
+
 // DeleteDocument 删除文档
 func (s *KnowledgeService) DeleteDocument(kbID, docID string) error {
 	// 验证文档存在
@@ -400,24 +434,66 @@ func (s *KnowledgeService) GetDocumentPreview(kbID, docID string, page int) (*mo
 		return nil, nil
 	}

-	// 如果已解析，返回解析内容；否则返回文件 URL
-	if doc.Status == "parsed" {
-		// TODO: 从存储中读取解析内容（可以存到数据库或文件）
-		// 暂时返回文件 URL
-		fileURL, _ := s.uploadService.GetFileURL(doc.FileKey)
+	// 获取文件URL
+	fileURL, _ := s.uploadService.GetFileURL(doc.FileKey)
+
+	// 根据文件类型决定预览方式
+	fileName := doc.Name
+	isPDF := strings.HasSuffix(strings.ToLower(fileName), ".pdf")
+	isOffice := false
+	officeExts := []string{".csv", ".xlsx", ".xls", ".docx", ".doc", ".pptx", ".ppt", ".txt", ".md"}
+	for _, ext := range officeExts {
+		if strings.HasSuffix(strings.ToLower(fileName), ext) {
+			isOffice = true
+			break
+		}
+	}
+
+	// PDF文件返回文件URL
+	if isPDF {
 		return &model.DocumentPreviewResponse{
 			TotalPages:  1,
 			CurrentPage: page,
 			Content:     fileURL,
+			ContentType: "url",
 		}, nil
 	}

-	// 未解析，返回文件 URL
-	fileURL, _ := s.uploadService.GetFileURL(doc.FileKey)
+	// Office文件调用解析服务转换为HTML
+	if isOffice && s.aiCoreClient != nil {
+		knowledgeDebugLog.Printf("[Preview] Parsing office file: %s, URL: %s", fileName, fileURL)
+		result, err := s.aiCoreClient.ParseDocument(fileURL, fileName, "")
+		if err != nil {
+			// 解析失败，返回文件URL
+			knowledgeDebugLog.Printf("[Preview] Parse document failed: %v", err)
+			return &model.DocumentPreviewResponse{
+				TotalPages:  1,
+				CurrentPage: page,
+				Content:     fileURL,
+				ContentType: "url",
+			}, nil
+		}
+
+		knowledgeDebugLog.Printf("[Preview] Parse result: success=%v, content_length=%d", result.Success, len(result.Content))
+
+		// 返回HTML内容
+		if result.Success && result.Content != "" {
+			knowledgeDebugLog.Printf("[Preview] Returning HTML content, length: %d", len(result.Content))
+			return &model.DocumentPreviewResponse{
+				TotalPages:  1,
+				CurrentPage: page,
+				Content:     result.Content,
+				ContentType: "html",
+			}, nil
+		}
+	}
+
+	// 其他情况返回文件URL
 	return &model.DocumentPreviewResponse{
 		TotalPages:  1,
 		CurrentPage: page,
 		Content:     fileURL,
+		ContentType: "url",
 	}, nil
 }

--- a/server/resource/markdown/0b12fa3a-e4ce-4eb8-b963-a11f78171bba.md
+++ b/server/resource/markdown/0b12fa3a-e4ce-4eb8-b963-a11f78171bba.md
@@ -0,0 +1,5 @@
+sk-5706307e3e3a4eb09452dbf0bb87fe31
+
+https://dashscope.aliyuncs.com/compatible-mode/v1
+
+qwen3.5-flash
--- a/server/resource/markdown/16dd7acd-ad0c-4e29-aa34-57ffd62c42a3.md
+++ b/server/resource/markdown/16dd7acd-ad0c-4e29-aa34-57ffd62c42a3.md
@@ -0,0 +1,5 @@
+sk-5706307e3e3a4eb09452dbf0bb87fe31
+
+https://dashscope.aliyuncs.com/compatible-mode/v1
+
+qwen3.5-flash
--- a/server/resource/markdown/1a54a087-4a07-4dbc-8631-e211275d20f7.md
+++ b/server/resource/markdown/1a54a087-4a07-4dbc-8631-e211275d20f7.md
@@ -0,0 +1,4 @@
+## students
+| 班级 | 姓名 | 年龄 | 性别 |
+| --- | --- | --- | --- |
+| 1 | 曹 | 123 | 男 |
--- a/server/resource/markdown/22d840d2-1752-450a-8788-a780903ca946.md
+++ b/server/resource/markdown/22d840d2-1752-450a-8788-a780903ca946.md
@@ -0,0 +1,13 @@
+# 1 . 你好么？
+
+### 表哥啊啊
+
+大叔大婶打扫的暗示打扫暗示
+
+> 太好了
+
+```python
+
+print("hello world")
+
+```
--- a/server/resource/markdown/2f2d2373-9f33-46c0-b129-5ababd3cedb8.md
+++ b/server/resource/markdown/2f2d2373-9f33-46c0-b129-5ababd3cedb8.md
@@ -0,0 +1,13 @@
+# 1 . 你好么？
+
+### 表哥啊啊
+
+大叔大婶打扫的暗示打扫暗示
+
+> 太好了
+
+```python
+
+print("hello world")
+
+```
--- a/server/resource/markdown/324ddd68-4c89-4f6b-9246-0447eddca73a.md
+++ b/server/resource/markdown/324ddd68-4c89-4f6b-9246-0447eddca73a.md
@@ -0,0 +1,13 @@
+# 1 . 你好么？
+
+### 表哥啊啊
+
+大叔大婶打扫的暗示打扫暗示
+
+> 太好了
+
+```python
+
+print("hello world")
+
+```
--- a/server/resource/markdown/617ddab2-dc12-4a58-91ff-45c614a012c1.md
+++ b/server/resource/markdown/617ddab2-dc12-4a58-91ff-45c614a012c1.md
@@ -0,0 +1,13 @@
+# 1 . 你好么？
+
+### 表哥啊啊
+
+大叔大婶打扫的暗示打扫暗示
+
+> 太好了
+
+```python
+
+print("hello world")
+
+```
--- a/server/resource/markdown/698e892b-cf0c-4fa8-998a-d46c5ec5dc5d.md
+++ b/server/resource/markdown/698e892b-cf0c-4fa8-998a-d46c5ec5dc5d.md
--- a/server/resource/markdown/782449c6-e4f1-47d5-b90e-341205540439.md
+++ b/server/resource/markdown/782449c6-e4f1-47d5-b90e-341205540439.md
@@ -0,0 +1,13 @@
+# 1 . 你好么？
+
+### 表哥啊啊
+
+大叔大婶打扫的暗示打扫暗示
+
+> 太好了
+
+```python
+
+print("hello world")
+
+```
--- a/server/resource/markdown/8acd5ffa-f5c1-4af9-9e4a-b7357324fd53.md
+++ b/server/resource/markdown/8acd5ffa-f5c1-4af9-9e4a-b7357324fd53.md
@@ -0,0 +1,5 @@
+sk-5706307e3e3a4eb09452dbf0bb87fe31
+
+https://dashscope.aliyuncs.com/compatible-mode/v1
+
+qwen3.5-flash
--- a/server/resource/markdown/979e94ae-6b4e-4952-8ebe-4de34d38d10d.md
+++ b/server/resource/markdown/979e94ae-6b4e-4952-8ebe-4de34d38d10d.md
@@ -0,0 +1,3 @@
+| 班级 | 姓名 | 年龄 | 性别 |
+| --- | --- | --- | --- |
+| 1 | 曹 | 123 | 男 |
--- a/server/resource/markdown/a64f5576-c3c1-483f-aadd-1cbbf0122651.md
+++ b/server/resource/markdown/a64f5576-c3c1-483f-aadd-1cbbf0122651.md
@@ -0,0 +1,5 @@
+sk-5706307e3e3a4eb09452dbf0bb87fe31
+
+https://dashscope.aliyuncs.com/compatible-mode/v1
+
+qwen3.5-flash
--- a/server/resource/markdown/b99f6ed4-0a54-4c4c-bb37-bb3e907bd5d1.md
+++ b/server/resource/markdown/b99f6ed4-0a54-4c4c-bb37-bb3e907bd5d1.md
@@ -0,0 +1,13 @@
+# 1 . 你好么？
+
+### 表哥啊啊
+
+大叔大婶打扫的暗示打扫暗示
+
+> 太好了
+
+```python
+
+print("hello world")
+
+```
--- a/server/resource/markdown/eabb7c06-a9ab-4c72-8361-9855d4d92731.md
+++ b/server/resource/markdown/eabb7c06-a9ab-4c72-8361-9855d4d92731.md
@@ -0,0 +1,3 @@
+| 겯섬 | 檎츰 | 쾨쥑 | 昑깎 |
+| --- | --- | --- | --- |
+| 1 | 꿀 | 123 | 켕 |
--- a/server/resource/markdown/ebbb8a77-5f54-400a-b398-1ce7ce72e6ef.md
+++ b/server/resource/markdown/ebbb8a77-5f54-400a-b398-1ce7ce72e6ef.md
@@ -0,0 +1,5 @@
+sk-5706307e3e3a4eb09452dbf0bb87fe31
+
+https://dashscope.aliyuncs.com/compatible-mode/v1
+
+qwen3.5-flash
--- a/server/resource/markdown/eea7fcdc-503d-47a8-af04-13e2adc0ca4a.md
+++ b/server/resource/markdown/eea7fcdc-503d-47a8-af04-13e2adc0ca4a.md
--- a/team-require/ai/ai-core-api.md
+++ b/team-require/ai/ai-core-api.md
@@ -6,6 +6,43 @@
 localhost:50051
 ```

+## VLM 配置（可选）
+
+VLM 用于提升图片文件的解析效果。如果不配置 VLM，则使用默认的 MarkItDown 解析。
+
+### 方式一：环境变量
+
+```bash
+# 设置环境变量
+export VLM_API_KEY="your-api-key"
+export VLM_PROVIDER="openai"  # openai / anthropic / qwen
+export VLM_MODEL="gpt-4o"
+```
+
+### 方式二：配置文件
+
+在 `ai-core/config.yaml` 中配置：
+
+```yaml
+vlm:
+  enabled: true
+  provider: "openai"      # openai / anthropic / qwen
+  model: "gpt-4o"        # 模型名称
+  api_key: "sk-xxx"      # API Key
+  base_url: ""           # 自定义 API 地址（可选）
+  prompt: ""             # 自定义提示词（可选）
+```
+
+### 支持的 VLM 提供商
+
+| 提供商 | 示例模型 |
+|--------|----------|
+| openai | gpt-4o, gpt-4o-mini |
+| anthropic | claude-3-opus, claude-3-sonnet |
+| qwen | qwen-vl-max, qwen2-vl-72b |
+
+---
+
 ## gRPC API 定义

 ### 1. ParseDocument - 解析文档
@@ -16,128 +53,85 @@ message ParseRequest {
  string file_url = 1;      // 文件 URL（必填）
  string file_name = 2;     // 文件名，带扩展名（必填）
  string file_type = 3;     // 文件类型（可选，自动检测）
-  map<string, string> engine_overrides = 4;  // 引擎配置
+  string parser_engine = 4; // 解析引擎（可选）
+  map<string, string> engine_overrides = 5;  // 引擎配置
+
+  // VLM 配置（可选，优先级高于全局配置）
+  VLMConfig vlm_config = 6;
+}
+
+message VLMConfig {
+  bool enabled = 1;
+  string provider = 2;
+  string model = 3;
+  string api_key = 4;
+  string base_url = 5;
+  string prompt = 6;
 }
 ```

 **响应 (ParseResponse)**
 ```protobuf
 message ParseResponse {
-  bool success = 1;           // 是否成功
+  bool success = 1;
  string content = 2;         // Markdown 内容
-  string message = 3;         // 状态消息
-  int32 content_length = 4;   // 内容长度
-  string file_type = 5;       // 文件类型
-  string parser_engine = 6;   // 解析引擎 (markitdown)
+  string message = 3;
+  int32 content_length = 4;
+  string file_type = 5;
+  string parser_engine = 6;
 }
 ```

-### 2. GetSupportedFormats - 获取支持的格式
-
-**请求**: 空消息
-
-**响应**
- `file_types`: string[] - 支持的扩展名列表
- `file_type_descriptions`: map<string, string> - 格式描述
-
 ---

 ## Golang 对接示例

-### 1. 安装依赖
-
-```bash
-go get google.golang.org/grpc
-go get google.golang.org/grpc/credentials/insecure
-```
-
-### 2. 生成 Go Proto 代码
-
-需要先将 `proto/document_parser.proto` 生成 Go 代码：
-
-```bash
-# 方法一：使用 grpc_tools
-go install google.golang.org/protobuf/cmd/protoc-gen-go@latest
-go install google.golang.org/grpc/cmd/protoc-gen-go-grpc@latest
-protoc --go_out=. --go_opt=paths=source_relative \
-       --go-grpc_out=. --go-grpc_opt=paths=source_relative \
-       proto/document_parser.proto
-```
-
-### 3. 完整调用代码
+### 基础调用（无 VLM 配置时使用 MarkItDown）

 ```go
-package main
-
-import (
-    "context"
-    "fmt"
-    "log"
-
-    "google.golang.org/grpc"
-    "google.golang.org/grpc/credentials/insecure"
-
-    pb "your-project/proto"  // 替换为你的 proto 包路径
-)
-
-func main() {
-    // 连接 gRPC 服务
-    conn, err := grpc.Dial(
-        "localhost:50051",
-        grpc.WithTransportCredentials(insecure.NewCredentials()),
-        grpc.WithBlock(),
-    )
-    if err != nil {
-        log.Fatalf("连接失败: %v", err)
-    }
-    defer conn.Close()
-
-    client := pb.NewDocumentParserClient(conn)
-    ctx := context.Background()
-
-    // 调用 ParseDocument
-    req := &pb.ParseRequest{
-        FileUrl:  "https://example.com/document.pdf",
-        FileName: "document.pdf",
-    }
-
-    resp, err := client.ParseDocument(ctx, req)
-    if err != nil {
-        log.Fatalf("解析失败: %v", err)
-    }
-
-    // 处理响应
-    if resp.Success {
-        fmt.Printf("解析成功！\n")
-        fmt.Printf("内容长度: %d 字符\n", resp.ContentLength)
-        fmt.Printf("Markdown 内容:\n%s\n", resp.Content)
-    } else {
-        fmt.Printf("解析失败: %s\n", resp.Message)
-    }
+req := &pb.ParseRequest{
+    FileUrl:  "https://example.com/document.pdf",
+    FileName: "document.pdf",
 }
+
+resp, client.ParseDocument(ctx, req)
 ```

-### 4. 获取支持的格式
+### 带 VLM 配置调用

 ```go
-// 获取支持的文件格式
-formatsReq := &pb.Empty{}
-formatsResp, err := client.GetSupportedFormats(ctx, formatsReq)
-if err != nil {
-    log.Fatal(err)
+req := &pb.ParseRequest{
+    FileUrl:  "https://example.com/image.png",
+    FileName: "image.png",
+    VlmConfig: &pb.VLMConfig{
+        Enabled: Provider: "open  true,
+       ai",
+        Model:    "gpt-4o",
+        ApiKey:   "sk-xxx",
+    },
 }

-fmt.Println("支持的格式:")
-for _, ft := range formatsResp.FileTypes {
-    desc := formatsResp.FileTypeDescriptions[ft]
-    fmt.Printf("  - %s: %s\n", ft, desc)
-}
+resp, err := client.ParseDocument(ctx, req)
 ```

 ---

+## 解析逻辑
+
+1. **图片文件** (jpg, png, webp 等)
+   - 如果配置了 VLM → 使用 VLM 解析
+   - 如果没有配置 VLM → 使用 MarkItDown 解析
+
+2. **PDF/DOCX/PPTX 等文档**
+   - 使用 MarkItDown 解析
+
+3. **VLM 优先级**
+   - gRPC 请求中的 vlm_config > 全局配置（config.yaml/环境变量）
+
+---
+
 ## 注意事项

-1. **文件 URL**: 必须是可直接访问的 URL，服务会下载文件到内存解析
-2. **文件名**: 必须带扩展名（如 `.pdf`, `.docx`），用于自动识别文件类型
-3. **返回内容**: 直接返回 Markdown 格式文本，可用于向量检索或 LLM 处理
+1. **文件 URL**: 必须是可直接访问的 URL
+2. **文件名**: 必须带扩展名（如 `.pdf`, `.png`）
+3. **返回内容**: Markdown 格式文本
--- a/team-require/ai/todo.md
+++ b/team-require/ai/todo.md
@@ -6,10 +6,15 @@

 - [ ] **AI-Core 文档解析服务对接**
  - 服务：ai-core (gRPC, 端口 50051)
-  - 功能：将文档（PDF/DOCX/PPTX 等）转换为 Markdown
+  - 功能：将文档（PDF/DOCX/PPTX/图片等）转换为 Markdown
  - 对接方式：gRPC 调用
  - 详细需求：[ai-core-api.md](./ai-core-api.md)

+- [ ] **VLM 调用支持**
+  - 支持 OpenAI GPT-4o、Anthropic Claude、阿里 Qwen VL
+  - 通过 vlm_config 配置启用
+  - 适用场景：图片文件（jpg, png 等）自动使用 VLM 解析
+
 ---

 > 需求完成后请完成者打 ✔
--- a/web/package-lock.json
+++ b/web/package-lock.json
@@ -8,10 +8,15 @@
      "name": "x-agent-dashboard",
      "version": "0.0.1",
      "dependencies": {
+        "@vue-office/docx": "^1.6.3",
+        "@vue-office/excel": "^1.7.14",
        "echarts": "^6.0.0",
        "element-plus": "^2.13.3",
+        "marked": "^17.0.4",
+        "papaparse": "^5.5.3",
        "pinia": "^2.1.7",
        "vue": "^3.4.21",
+        "vue-demi": "^0.14.10",
        "vue-router": "^4.3.0"
      },
      "devDependencies": {
@@ -1026,6 +1031,40 @@
        "vscode-uri": "^3.0.8"
      }
    },
+    "node_modules/@vue-office/docx": {
+      "version": "1.6.3",
+      "resolved": "https://registry.npmmirror.com/@vue-office/docx/-/docx-1.6.3.tgz",
+      "integrity": "sha512-Cs+3CAaRBOWOiW4XAhTwwxJ0dy8cPIf6DqfNvYcD3YACiLwO4kuawLF2IAXxyijhbuOeoFsfvoVbOc16A/4bZA==",
+      "hasInstallScript": true,
+      "license": "MIT",
+      "peerDependencies": {
+        "@vue/composition-api": "^1.7.1",
+        "vue": "^2.0.0 || >=3.0.0",
+        "vue-demi": "^0.14.6"
+      },
+      "peerDependenciesMeta": {
+        "@vue/composition-api": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@vue-office/excel": {
+      "version": "1.7.14",
+      "resolved": "https://registry.npmmirror.com/@vue-office/excel/-/excel-1.7.14.tgz",
+      "integrity": "sha512-pVUgt+emDQUnW7q22CfnQ+jl43mM/7IFwYzOg7lwOwPEbiVB4K4qEQf+y/bc4xGXz75w1/e3Kz3G6wAafmFBFg==",
+      "hasInstallScript": true,
+      "license": "MIT",
+      "peerDependencies": {
+        "@vue/composition-api": "^1.7.1",
+        "vue": "^2.0.0 || >=3.0.0",
+        "vue-demi": "^0.14.6"
+      },
+      "peerDependenciesMeta": {
+        "@vue/composition-api": {
+          "optional": true
+        }
+      }
+    },
    "node_modules/@vue/compiler-core": {
      "version": "3.5.29",
      "resolved": "https://registry.npmmirror.com/@vue/compiler-core/-/compiler-core-3.5.29.tgz",
@@ -1862,6 +1901,18 @@
        "@jridgewell/sourcemap-codec": "^1.5.5"
      }
    },
+    "node_modules/marked": {
+      "version": "17.0.4",
+      "resolved": "https://registry.npmmirror.com/marked/-/marked-17.0.4.tgz",
+      "integrity": "sha512-NOmVMM+KAokHMvjWmC5N/ZOvgmSWuqJB8FoYI019j4ogb/PeRMKoKIjReZ2w3376kkA8dSJIP8uD993Kxc0iRQ==",
+      "license": "MIT",
+      "bin": {
+        "marked": "bin/marked.js"
+      },
+      "engines": {
+        "node": ">= 20"
+      }
+    },
    "node_modules/memoize-one": {
      "version": "6.0.0",
      "resolved": "https://registry.npmmirror.com/memoize-one/-/memoize-one-6.0.0.tgz",
@@ -1988,6 +2039,12 @@
        "node": ">= 6"
      }
    },
+    "node_modules/papaparse": {
+      "version": "5.5.3",
+      "resolved": "https://registry.npmmirror.com/papaparse/-/papaparse-5.5.3.tgz",
+      "integrity": "sha512-5QvjGxYVjxO59MGU2lHVYpRWBBtKHnlIAcSe1uNFCkkptUh63NFRj0FJQm7nR67puEruUci/ZkjmEFrjCAyP4A==",
+      "license": "MIT"
+    },
    "node_modules/path-browserify": {
      "version": "1.0.1",
      "resolved": "https://registry.npmmirror.com/path-browserify/-/path-browserify-1.0.1.tgz",
@@ -2701,6 +2758,7 @@
      "integrity": "sha512-nMZBOwuzabUO0nLgIcc6rycZEebF6eeUfaiQx9+WSk8e29IbLvPU9feI6tqW4kTo3hvoYAJkMh8n8D0fuISphg==",
      "hasInstallScript": true,
      "license": "MIT",
+      "peer": true,
      "bin": {
        "vue-demi-fix": "bin/vue-demi-fix.js",
        "vue-demi-switch": "bin/vue-demi-switch.js"
--- a/web/package.json
+++ b/web/package.json
@@ -9,10 +9,15 @@
    "preview": "vite preview"
  },
  "dependencies": {
+    "@vue-office/docx": "^1.6.3",
+    "@vue-office/excel": "^1.7.14",
    "echarts": "^6.0.0",
    "element-plus": "^2.13.3",
+    "marked": "^17.0.4",
+    "papaparse": "^5.5.3",
    "pinia": "^2.1.7",
    "vue": "^3.4.21",
+    "vue-demi": "^0.14.10",
    "vue-router": "^4.3.0"
  },
  "devDependencies": {
--- a/web/src/components/FormDialog.vue
+++ b/web/src/components/FormDialog.vue
@@ -5,6 +5,7 @@ defineProps<{
  description?: string
  icon?: string
  iconClass?: string
+  class?: string
 }>()

 const emit = defineEmits<{
@@ -18,7 +19,7 @@ const close = () => {

 <template>
  <Teleport to="body">
-    <div v-if="modelValue" class="dialog-overlay" @click.self="close">
+    <div v-if="modelValue" class="dialog-overlay" :class="$props.class" @click.self="close">
      <div class="dialog-container">
        <!-- 头部 -->
        <div class="dialog-header">
@@ -61,7 +62,7 @@ const close = () => {
  display: flex;
  align-items: center;
  justify-content: center;
-  z-index: 9999;
+  z-index: 2000;
 }

 .dialog-container {
--- a/web/src/components/Sidebar.vue
+++ b/web/src/components/Sidebar.vue
@@ -21,12 +21,15 @@ const mainMenu: MenuItem[] = [
  { name: 'Dashboard', icon: 'fa-gauge', path: '/dashboard' },
  { name: 'Agents', icon: 'fa-robot', badge: 3, path: '/agents' },
  { name: 'Team', icon: 'fa-users', path: '/team' },
-  { name: 'Skills', icon: 'fa-wand-magic-sparkles', badge: 21, path: '/mcp' },
-  { name: 'Tools', icon: 'fa-tools', badge: 13, path: '/model-apis' },
  { name: 'Database', icon: 'fa-database', path: '/database' },
  { name: 'Knowledge', icon: 'fa-brain', path: '/knowledge' },
 ]

+const middleMenu: MenuItem[] = [
+  { name: 'Skills', icon: 'fa-wand-magic-sparkles', badge: 21, path: '/mcp' },
+  { name: 'Tools', icon: 'fa-tools', badge: 13, path: '/model-apis' },
+]
+
 const bottomMenu: MenuItem[] = [
  { name: 'Settings', icon: 'fa-gear', path: '/settings' },
 ]
@@ -117,6 +120,25 @@ const handleUserCommand = (command: string) => {
        <!-- 分隔线 -->
        <li class="my-4 border-t border-dark-500"></li>

+        <!-- Skills & Tools -->
+        <li v-for="item in middleMenu" :key="item.name">
+          <a
+            href="#"
+            class="flex items-center justify-between px-3 py-2.5 rounded-lg transition-colors text-sm"
+            :class="activeMenu === item.name ? 'bg-dark-600 text-white' : 'text-gray-400 hover:bg-dark-600 hover:text-white'"
+            @click="navigateTo(item)"
+          >
+            <div class="flex items-center gap-3">
+              <i :class="['fa-solid', item.icon, 'w-5', 'text-center']"></i>
+              <span>{{ item.name }}</span>
+            </div>
+            <span v-if="item.badge" class="bg-dark-500 text-xs px-2 py-0.5 rounded-full">{{ item.badge }}</span>
+          </a>
+        </li>
+
+        <!-- 分隔线 -->
+        <li class="my-4 border-t border-dark-500"></li>
+
        <!-- Settings -->
        <li v-for="item in bottomMenu" :key="item.name">
          <a
--- a/web/src/views/Knowledge.vue
+++ b/web/src/views/Knowledge.vue
@@ -3,6 +3,10 @@ import { ref, computed, onMounted } from 'vue'
 import { ElMessage } from 'element-plus'
 import { useModelSettings } from './settings/useModelSettings'
 import { fetchKnowledgeBases, createKnowledgeBase as apiCreateKnowledgeBase, deleteKnowledgeBase as apiDeleteKnowledgeBase, fetchKnowledgeDocuments } from './knowledge/useKnowledge'
+import VueOfficeDocx from '@vue-office/docx'
+import VueOfficeExcel from '@vue-office/excel'
+import Papa from 'papaparse'
+import { marked } from 'marked'
 import './knowledge/knowledge.css'

 // 获取已配置的模型列表
@@ -135,7 +139,12 @@ const loadingDocuments = ref(false)
 const fileInput = ref<HTMLInputElement | null>(null)
 const uploading = ref(false)
 const previewUrl = ref('') // 文档预览URL (blob URL)
+const previewHtml = ref('') // HTML内容预览
+const previewContentType = ref('') // content_type: url 或 html
 const previewDownloadUrl = ref('') // 原始下载链接
+const previewFileType = ref('') // 文件类型: pdf, docx, xlsx, csv
+const previewCsvData = ref<any[]>([]) // CSV数据
+const previewCsvHeaders = ref<string[]>([]) // CSV表头
 const loadingPreview = ref(false)
 const previewPage = ref(1) // 当前页码
 const previewTotalPages = ref(1) // 总页数
@@ -157,6 +166,20 @@ const loadPdfWithProxy = async (doc: any): Promise<string> => {
  }
 }

+// 将URL转换为blob URL (用于Excel组件)
+const loadFileAsBlob = async (url: string): Promise<Blob> => {
+  try {
+    const response = await fetch(url)
+    if (!response.ok) {
+      throw new Error(`Failed to fetch: ${response.status}`)
+    }
+    return await response.blob()
+  } catch (error) {
+    console.error('Failed to load file as blob:', error)
+    throw error
+  }
+}
+
 const newKbForm = ref({
  name: '',
  description: '',
@@ -337,6 +360,11 @@ const enterKnowledge = async (kb: any) => {
  selectedKnowledge.value = kb
  selectedFile.value = null
  previewUrl.value = ''
+  previewHtml.value = ''
+  previewContentType.value = ''
+  previewFileType.value = ''
+  previewCsvData.value = []
+  previewCsvHeaders.value = []
  previewDownloadUrl.value = ''

  // 获取文档列表
@@ -375,13 +403,126 @@ const selectDocument = async (doc: any) => {
  selectedFile.value = doc.id
  selectedDocument.value = doc
  previewUrl.value = ''
+  previewHtml.value = ''
+  previewContentType.value = ''
  previewDownloadUrl.value = ''
+  previewFileType.value = ''
+  previewCsvData.value = []
+  previewCsvHeaders.value = []
  previewPage.value = 1
  previewTotalPages.value = 1

-  // 优先使用代理接口加载PDF
+  // 检测文件类型
+  const fileName = doc.name || ''
+  const ext = fileName.split('.').pop()?.toLowerCase() || ''
+  console.log('File name:', fileName, 'Ext:', ext)
+  if (ext === 'pdf') {
+    previewFileType.value = 'pdf'
+  } else if (ext === 'docx' || ext === 'doc') {
+    previewFileType.value = 'docx'
+  } else if (ext === 'xlsx' || ext === 'xls') {
+    previewFileType.value = 'xlsx'
+  } else if (ext === 'csv') {
+    previewFileType.value = 'csv'
+  } else if (ext === 'txt' || ext === 'md') {
+    previewFileType.value = 'text'
+  }
+  console.log('Preview file type:', previewFileType.value)
+
+  // 纯文本文件 (txt, md) 直接读取内容显示
+  if (previewFileType.value === 'text') {
+    // 优先使用 file_url（上传时后端返回的）
+    let url = doc.file_url || doc.fileUrl || doc.url || doc.FileURL || ''
+    // 如果没有 file_url，尝试用 file_key 通过代理
+    if (!url && doc.file_key && selectedKnowledge.value) {
+      url = await loadPdfWithProxy(doc)
+    }
+
+    console.log('Text file URL:', url)
+    if (url) {
+      try {
+        const response = await fetch(url)
+        console.log('Response status:', response.status)
+        const arrayBuffer = await response.arrayBuffer()
+        const decoder = new TextDecoder('utf-8')
+        const content = decoder.decode(arrayBuffer)
+        console.log('Content length:', content.length, 'ext:', ext)
+
+        // .md 文件用 marked 渲染
+        if (ext === 'md') {
+          // marked.parse 返回 Promise，需要 await
+          const html = await marked.parse(content)
+          previewHtml.value = html as string
+        } else {
+          // .txt 文件直接显示
+          previewHtml.value = '<pre style="white-space: pre-wrap; word-wrap: break-word; color: #e8eaed;">' + content + '</pre>'
+        }
+        console.log('PreviewHtml set, length:', previewHtml.value.length)
+        return
+      } catch (error) {
+        console.error('Failed to load text file:', error)
+      }
+    } else {
+      console.log('No URL for text file, file_key:', doc.file_key, 'file_url:', doc.file_url)
+    }
+  }
+
+  // CSV需要特殊处理：用PapaParse解析
+  if (previewFileType.value === 'csv' && (doc.file_key || doc.file_url)) {
+    const url = doc.file_key && selectedKnowledge.value
+      ? await loadPdfWithProxy(doc)
+      : doc.file_url || doc.fileUrl || doc.url || ''
+    if (url) {
+      try {
+        const response = await fetch(url)
+        const arrayBuffer = await response.arrayBuffer()
+
+        // 尝试多种编码解码
+        let csvText = ''
+        const encodings = ['utf-8', 'gbk', 'gb2312', 'big5']
+        for (const encoding of encodings) {
+          try {
+            const decoder = new TextDecoder(encoding)
+            csvText = decoder.decode(arrayBuffer)
+            // 检查是否成功解码（如果没有乱码字符）
+            if (!csvText.includes('\uFFFD')) {
+              console.log(`CSV decoded with: ${encoding}`)
+              break
+            }
+          } catch (e) {
+            continue
+          }
+        }
+
+        const result = Papa.parse(csvText, { header: false, skipEmptyLines: true })
+        if (result.data && result.data.length > 0) {
+          previewCsvHeaders.value = result.data[0] as string[]
+          previewCsvData.value = result.data.slice(1) as any[][]
+        }
+        return
+      } catch (error) {
+        console.error('Failed to parse CSV:', error)
+      }
+    }
+  }
+
+  // 使用代理接口加载文件
  if (doc.file_key && selectedKnowledge.value) {
-    previewUrl.value = await loadPdfWithProxy(doc)
+    const proxyUrl = await loadPdfWithProxy(doc)
+
+    // Excel/Word需要blob URL
+    if ((previewFileType.value === 'xlsx' || previewFileType.value === 'docx') && proxyUrl) {
+      try {
+        const blob = await loadFileAsBlob(proxyUrl)
+        const blobUrl = URL.createObjectURL(blob)
+        previewUrl.value = blobUrl
+        return
+      } catch (error) {
+        console.error('Failed to convert to blob:', error)
+      }
+    }
+
+    previewUrl.value = proxyUrl
    if (previewUrl.value) {
      return
    }
@@ -397,11 +538,19 @@ const selectDocument = async (doc: any) => {
  // 如果没有file_url，调用预览API获取
  if (selectedKnowledge.value) {
    loadingPreview.value = true
+    previewUrl.value = ''
+    previewHtml.value = ''
+    previewContentType.value = ''
    try {
      const { getDocumentPreview } = await import('./knowledge/useKnowledge')
      const result = await getDocumentPreview(selectedKnowledge.value.id, doc.id)
      if (result.success && result.data?.content) {
-        previewUrl.value = result.data.content
+        previewContentType.value = result.data.content_type || 'url'
+        if (previewContentType.value === 'html') {
+          previewHtml.value = result.data.content
+        } else {
+          previewUrl.value = result.data.content
+        }
        previewTotalPages.value = result.data.total_pages || 1
        previewPage.value = result.data.current_page || 1
      }
@@ -416,6 +565,8 @@ const selectDocument = async (doc: any) => {
 // 翻页
 const changePreviewPage = async (page: number) => {
  if (!selectedKnowledge.value || !selectedFile.value || page < 1 || page > previewTotalPages.value) return
+  // HTML内容不支持翻页
+  if (previewContentType.value === 'html') return

  loadingPreview.value = true
  previewPage.value = page
@@ -423,7 +574,12 @@ const changePreviewPage = async (page: number) => {
    const { getDocumentPreview } = await import('./knowledge/useKnowledge')
    const result = await getDocumentPreview(selectedKnowledge.value.id, selectedFile.value, page)
    if (result.success && result.data?.content) {
-      previewUrl.value = result.data.content
+      previewContentType.value = result.data.content_type || 'url'
+      if (previewContentType.value === 'html') {
+        previewHtml.value = result.data.content
+      } else {
+        previewUrl.value = result.data.content
+      }
      previewTotalPages.value = result.data.total_pages || 1
      previewPage.value = result.data.current_page || page
    }
@@ -459,16 +615,8 @@ const handleFileSelect = async (event: Event) => {
      // 获取刚上传的文档
      const uploadedDoc = knowledgeDocuments.value.find(d => d.id === result.id)
      if (uploadedDoc) {
-        // 选中新上传的文档
-        selectedFile.value = result.id
-        selectedDocument.value = uploadedDoc
-
-        // 使用代理接口加载PDF
-        if (uploadedDoc.file_key) {
-          previewUrl.value = await loadPdfWithProxy(uploadedDoc)
-        } else if (uploadedDoc.file_url) {
-          previewUrl.value = uploadedDoc.file_url
-        }
+        // 选中新上传的文档，调用 selectDocument 处理预览
+        await selectDocument(uploadedDoc)
      }
    } else {
      ElMessage.error(result.message || 'Failed to upload file')
@@ -497,7 +645,12 @@ const deleteDocument = async (docId: string) => {
        selectedFile.value = null
        selectedDocument.value = null
        previewUrl.value = ''
-  previewDownloadUrl.value = ''
+        previewHtml.value = ''
+        previewContentType.value = ''
+        previewFileType.value = ''
+        previewCsvData.value = []
+        previewCsvHeaders.value = []
+        previewDownloadUrl.value = ''
      }
      // 刷新文档列表
      await changeFileFilter(fileFilter.value)
@@ -873,7 +1026,7 @@ const deleteDocument = async (docId: string) => {
            type="file"
            ref="fileInput"
            style="display: none"
-            accept=".pdf,.doc,.docx,.txt,.md"
+            accept=".pdf,.doc,.docx,.docx,.txt,.md,.csv,.xlsx,.xls,.pptx,.ppt"
            @change="handleFileSelect"
          />
          <button class="btn-primary" @click="triggerFileUpload">
@@ -987,7 +1140,40 @@ const deleteDocument = async (docId: string) => {
                  <i class="fa-solid fa-spinner fa-spin"></i>
                  <span>Loading preview...</span>
                </div>
-                <!-- 有blob预览URL时显示PDF (使用iframe) -->
+                <!-- Word文件预览 -->
+                <VueOfficeDocx
+                  v-else-if="previewUrl && previewFileType === 'docx'"
+                  :src="previewUrl"
+                  class="office-embed"
+                  @rendered="console.log('Docx rendered')"
+                  @error="console.error('Docx error', $event)"
+                />
+                <!-- Excel文件预览 -->
+                <VueOfficeExcel
+                  v-else-if="previewUrl && previewFileType === 'xlsx'"
+                  :src="previewUrl"
+                  class="office-embed"
+                  @rendered="console.log('Excel rendered')"
+                  @error="console.error('Excel error', $event)"
+                />
+                <!-- CSV文件预览 -->
+                <div v-else-if="previewFileType === 'csv' && previewCsvData.length > 0" class="csv-preview">
+                  <table>
+                    <thead>
+                      <tr>
+                        <th v-for="(header, index) in previewCsvHeaders" :key="index">{{ header }}</th>
+                      </tr>
+                    </thead>
+                    <tbody>
+                      <tr v-for="(row, rowIndex) in previewCsvData" :key="rowIndex">
+                        <td v-for="(cell, cellIndex) in row" :key="cellIndex">{{ cell }}</td>
+                      </tr>
+                    </tbody>
+                  </table>
+                </div>
+                <!-- 文本文件预览 (txt, md) -->
+                <div v-else-if="previewHtml" class="text-preview" v-html="previewHtml"></div>
+                <!-- PDF/其他文件预览 (使用iframe) -->
                <iframe
                  v-else-if="previewUrl"
                  :src="previewUrl"
@@ -996,9 +1182,9 @@ const deleteDocument = async (docId: string) => {
                <!-- 无预览但有下载链接时显示下载按钮 -->
                <div v-else-if="previewDownloadUrl" class="preview-no-file">
                  <i class="fa-solid fa-file-pdf"></i>
-                  <span>Cannot preview PDF directly</span>
+                  <span>Cannot preview file directly</span>
                  <a :href="previewDownloadUrl" target="_blank" class="download-link">
-                    <i class="fa-solid fa-download"></i> Download PDF
+                    <i class="fa-solid fa-download"></i> Download File
                  </a>
                </div>
                <!-- 无预览也无下载链接时显示提示 -->
--- a/web/src/views/Settings.vue
+++ b/web/src/views/Settings.vue
@@ -241,6 +241,7 @@ const showChangePassword = () => {
            description="Configure your model settings"
            icon="fa-solid fa-brain"
            icon-class="bg-gradient-to-br from-primary-orange to-red-500"
+            class="add-model-dialog"
          >
            <div class="space-y-4">
              <div>
--- a/web/src/views/knowledge/knowledge.css
+++ b/web/src/views/knowledge/knowledge.css
@@ -943,6 +943,166 @@
  border: none;
 }

+/* HTML内容预览 */
+.html-preview {
+  width: 100%;
+  height: 100%;
+  overflow: auto;
+  padding: 20px;
+  background-color: #1e1e24;
+  color: #e8eaed;
+  line-height: 1.6;
+}
+
+.html-preview table {
+  width: 100%;
+  border-collapse: collapse;
+  margin: 16px 0;
+}
+
+.html-preview table th,
+.html-preview table td {
+  border: 1px solid #3a3a4a;
+  padding: 8px 12px;
+  text-align: left;
+}
+
+.html-preview table th {
+  background-color: #2a2a3a;
+}
+
+.html-preview h1, .html-preview h2, .html-preview h3 {
+  margin-top: 16px;
+  margin-bottom: 8px;
+  color: #ffffff;
+}
+
+.html-preview pre {
+  background-color: #2a2a3a;
+  padding: 12px;
+  border-radius: 6px;
+  overflow-x: auto;
+}
+
+.html-preview code {
+  background-color: #2a2a3a;
+  padding: 2px 6px;
+  border-radius: 4px;
+  font-family: monospace;
+}
+
+/* Office 组件预览 */
+.office-embed {
+  width: 100%;
+  height: 100%;
+  overflow: auto;
+  background-color: #ffffff;
+}
+
+/* 修复 vue-office 表格黑色背景问题 */
+.office-embed :deep(table) {
+  background-color: #ffffff !important;
+  color: #333333 !important;
+}
+
+.office-embed :deep(td),
+.office-embed :deep(th) {
+  background-color: #ffffff !important;
+  color: #333333 !important;
+  border: 1px solid #dddddd !important;
+}
+
+/* CSV 表格预览 */
+.csv-preview {
+  width: 100%;
+  height: 100%;
+  overflow: auto;
+  background-color: #ffffff;
+  padding: 16px;
+}
+
+.csv-preview table {
+  width: 100%;
+  border-collapse: collapse;
+}
+
+.csv-preview th,
+.csv-preview td {
+  border: 1px solid #dddddd;
+  padding: 8px 12px;
+  text-align: left;
+  color: #333333;
+}
+
+.csv-preview th {
+  background-color: #f5f5f5;
+  font-weight: 600;
+}
+
+.csv-preview tr:nth-child(even) {
+  background-color: #fafafa;
+}
+
+/* 文本文件预览 */
+.text-preview {
+  width: 100%;
+  height: 100%;
+  overflow: auto;
+  padding: 16px;
+  background-color: #1e1e24;
+  color: #e8eaed;
+}
+
+.text-preview pre {
+  white-space: pre-wrap;
+  word-wrap: break-word;
+  color: #e8eaed;
+  font-family: monospace;
+  font-size: 14px;
+  line-height: 1.6;
+}
+
+/* Markdown 渲染样式 */
+.text-preview h1, .text-preview h2, .text-preview h3,
+.text-preview h4, .text-preview h5, .text-preview h6 {
+  color: #ffffff;
+  margin-top: 16px;
+  margin-bottom: 8px;
+}
+
+.text-preview code {
+  background-color: #2a2a3a;
+  padding: 2px 6px;
+  border-radius: 4px;
+  font-family: monospace;
+}
+
+.text-preview pre {
+  background-color: #2a2a3a;
+  padding: 12px;
+  border-radius: 6px;
+  overflow-x: auto;
+}
+
+.text-preview ul, .text-preview ol {
+  padding-left: 24px;
+}
+
+.text-preview li {
+  margin-bottom: 4px;
+}
+
+.text-preview a {
+  color: #f97316;
+}
+
+.text-preview blockquote {
+  border-left: 4px solid #f97316;
+  padding-left: 16px;
+  margin-left: 0;
+  color: #9ca3af;
+}
+
 .preview-info {
  display: grid;
  grid-template-columns: repeat(2, 1fr);
--- a/web/src/views/settings/modelSettings.css
+++ b/web/src/views/settings/modelSettings.css
@@ -256,10 +256,14 @@
  color: #6b7280;
 }

-.add-model-dialog :deep(.el-select) {
+.add-model-dialog .el-select {
  width: 100%;
 }

+.add-model-dialog .el-select .el-input {
+  pointer-events: auto;
+}
+
 .add-model-dialog :deep(.el-select .el-input__wrapper) {
  background-color: #171922;
  border: 1px solid #4b5563;
@@ -276,6 +280,7 @@
 .add-model-dialog :deep(.el-select-dropdown) {
  background-color: #1a1a24;
  border: 1px solid #2a2a3a;
+  z-index: 10000;
 }

 .add-model-dialog :deep(.el-select-dropdown__item) {
Author	SHA1	Message	Date
DESKTOP-72TV0V4\caoxiaozhu	e633462ace	style: 优化 Settings 页面样式 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-09 15:43:24 +08:00
DESKTOP-72TV0V4\caoxiaozhu	9ca267244d	chore: 更新前端依赖和侧边栏导航 - 添加必要的前端依赖 - 优化侧边栏交互 - 更新 AI-Core API 文档 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-09 15:43:07 +08:00
DESKTOP-72TV0V4\caoxiaozhu	49433f1681	feat: 增强 Knowledge 页面功能 - 优化文档预览功能 - 添加 CSV 文件解析支持 - 增强知识库详情展示 - 优化样式和交互 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-09 15:42:55 +08:00
DESKTOP-72TV0V4\caoxiaozhu	4a7199de93	feat: 完善后端知识库服务和配置 - 优化 AI-Core 客户端调用 - 添加更多知识库配置选项 - 完善文档解析逻辑 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-09 15:42:42 +08:00
DESKTOP-72TV0V4\caoxiaozhu	5012a25f99	feat: 增强 AI-Core 文档解析器 - 添加 VLM 客户端支持 - 优化解析器配置 - 添加配置示例文件 - 生成新的 gRPC protobuf 文件 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-03-09 15:42:35 +08:00