From b1d0cc546230d99fda01f67dcdc59c169f3cbdbb Mon Sep 17 00:00:00 2001
From: Zhaojie <jiez93431@gmail.com>
Date: Fri, 9 Jan 2026 16:52:45 +0800
Subject: [PATCH] =?UTF-8?q?=E6=9B=B4=E6=96=B0readme=E6=96=87=E6=A1=A3?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .env.example           |   2 +-
 LICENSE                |  21 ---
 README.md              |   6 +-
 bootstrap.py           |  62 +++++++
 config/llm_config.py   |   4 +-
 data_analysis_agent.py | 234 +++++++++++++++---------
 main.py                |  57 +++++-
 merge_excel.py         |  83 +++++++++
 prompts.py             | 119 +++++++-----
 requirements.txt       |   5 +
 sort_csv.py            |  45 +++++
 start.bat              |   4 +
 start.sh               |   3 +
 start_web.bat          |   5 +
 start_web.sh           |   4 +
 test.py                |  13 ++
 utils/code_executor.py |  47 ++++-
 utils/llm_helper.py    |   3 +-
 web/main.py            | 404 +++++++++++++++++++++++++++++++++++++++++
 web/static/index.html  | 175 ++++++++++++++++++
 web/static/script.js   | 362 ++++++++++++++++++++++++++++++++++++
 web/static/style.css   | 387 +++++++++++++++++++++++++++++++++++++++
 22 files changed, 1871 insertions(+), 174 deletions(-)
 delete mode 100644 LICENSE
 create mode 100644 bootstrap.py
 create mode 100644 merge_excel.py
 create mode 100644 sort_csv.py
 create mode 100644 start.bat
 create mode 100755 start.sh
 create mode 100644 start_web.bat
 create mode 100755 start_web.sh
 create mode 100644 test.py
 create mode 100644 web/main.py
 create mode 100644 web/static/index.html
 create mode 100644 web/static/script.js
 create mode 100644 web/static/style.css

diff --git a/.env.example b/.env.example
index 25df76c..bfcc831 100644
--- a/.env.example
+++ b/.env.example
@@ -1,7 +1,7 @@
 
 # 火山引擎配置
 OPENAI_API_KEY=sk-c44i1hy64xgzwox6x08o4zug93frq6rgn84oqugf2pje1tg4
-OPENAI_BASE_URL=https://api.xiaomimimo.com/v1/chat/completions
+OPENAI_BASE_URL=https://api.xiaomimimo.com/v1
 # 文本模型
 OPENAI_MODEL=mimo-v2-flash
 # OPENAI_MODEL=deepseek-r1-250528
diff --git a/LICENSE b/LICENSE
deleted file mode 100644
index 5d9664b..0000000
--- a/LICENSE
+++ /dev/null
@@ -1,21 +0,0 @@
-MIT License
-
-Copyright (c) 2025 Data Analysis Agent Team
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
diff --git a/README.md b/README.md
index 1f624d0..ce07d7d 100644
--- a/README.md
+++ b/README.md
@@ -160,7 +160,7 @@ agent = DataAnalysisAgent(llm_config)
 # 开始分析
 files = ["your_data.csv"]
 report = agent.analyze(
-    user_input="分析销售数据，生成趋势图表和关键指标",
+    user_input="分析XXXXXXXXX数据，生成趋势图表和关键指标",
     files=files
 )
 
@@ -191,9 +191,9 @@ report = quick_analysis(
 
 ```python
 # 示例：茅台财务分析
-files = ["贵州茅台利润表.csv"]
+files = ["XXXXXXXXx.csv"]
 report = agent.analyze(
-    user_input="基于贵州茅台的数据，输出五个重要的统计指标，并绘制相关图表。最后生成汇报给我。",
+    user_input="基于数据，输出五个重要的统计指标，并绘制相关图表。最后生成汇报给我。",
     files=files
 )
 ```
diff --git a/bootstrap.py b/bootstrap.py
new file mode 100644
index 0000000..a6233c8
--- /dev/null
+++ b/bootstrap.py
@@ -0,0 +1,62 @@
+import sys
+import subprocess
+import importlib.metadata
+import os
+
+def check_dependencies():
+    """Checks if dependencies in requirements.txt are installed."""
+    requirements_file = "requirements.txt"
+    if not os.path.exists(requirements_file):
+        print(f"Warning: {requirements_file} not found. Skipping dependency check.")
+        return
+
+    print("Checking dependencies...")
+    missing_packages = []
+    
+    with open(requirements_file, "r") as f:
+        for line in f:
+            line = line.strip()
+            if not line or line.startswith("#"):
+                continue
+            
+            # Simple parsing for package name. 
+            # This handles 'package>=version', 'package==version', 'package'
+            # It does NOT handle complex markers perfectly, but suffices for basic checking.
+            package_name = line.split("=")[0].split(">")[0].split("<")[0].strip()
+            
+            try:
+                importlib.metadata.version(package_name)
+            except importlib.metadata.PackageNotFoundError:
+                missing_packages.append(line)
+
+    if missing_packages:
+        print(f"Missing dependencies: {', '.join(missing_packages)}")
+        print("Installing missing dependencies...")
+        try:
+            subprocess.check_call([sys.executable, "-m", "pip", "install", "-r", requirements_file])
+            print("Dependencies installed successfully.")
+        except subprocess.CalledProcessError as e:
+            print(f"Error installing dependencies: {e}")
+            sys.exit(1)
+    else:
+        print("All dependencies checked.")
+
+def main():
+    check_dependencies()
+    
+    print("Starting application...")
+    try:
+        # Run the main application
+        # Using sys.executable ensures we use the same python interpreter
+        subprocess.run([sys.executable, "main.py"], check=True)
+    except subprocess.CalledProcessError as e:
+        print(f"Application exited with error: {e}")
+        sys.exit(e.returncode)
+    except KeyboardInterrupt:
+        print("\nApplication stopped by user.")
+    except Exception as e:
+        print(f"An unexpected error occurred: {e}")
+        sys.exit(1)
+
+if __name__ == "__main__":
+    main()
diff --git a/config/llm_config.py b/config/llm_config.py
index ffadb9a..ce97a0c 100644
--- a/config/llm_config.py
+++ b/config/llm_config.py
@@ -17,8 +17,8 @@ load_dotenv()
 class LLMConfig:
     """LLM配置"""
 
-    provider: str = os.environ.get("LLM_PROVIDER", "gemini")  # openai, gemini, etc.
-    api_key: str = os.environ.get("OPENAI_API_KEY", "sk---c44i1hy64xgzwox6x08o4zug93frq6rgn84oqugf2pje1tg4")
+    provider: str = os.environ.get("LLM_PROVIDER", "openai")  # openai, gemini, etc.
+    api_key: str = os.environ.get("OPENAI_API_KEY", "sk-c44i1hy64xgzwox6x08o4zug93frq6rgn84oqugf2pje1tg4")
     base_url: str = os.environ.get("OPENAI_BASE_URL", "https://api.xiaomimimo.com/v1")
     model: str = os.environ.get("OPENAI_MODEL", "mimo-v2-flash")
     temperature: float = 0.5
diff --git a/data_analysis_agent.py b/data_analysis_agent.py
index 2087845..82afafe 100644
--- a/data_analysis_agent.py
+++ b/data_analysis_agent.py
@@ -136,30 +136,29 @@ class DataAnalysisAgent:
             print(f"   🔍 分析: {analysis}")
 
 
-            # 记录图片信息
-            collected_figures.append(
-                {
-                    "figure_number": figure_number,
-                    "filename": filename,
-                    "file_path": file_path,
-                    "description": description,
-                    "analysis": analysis,
-                }
-            )
+            # 使用seen_paths集合来去重，防止重复收集
+            seen_paths = set()
+            
             # 验证文件是否存在
             # 只有文件真正存在时才加入列表，防止报告出现裂图
             if file_path and os.path.exists(file_path):
-                print(f"   ✅ 文件存在: {file_path}")
-                # 记录图片信息
-                collected_figures.append(
-                    {
-                        "figure_number": figure_number,
-                        "filename": filename,
-                        "file_path": file_path,
-                        "description": description,
-                        "analysis": analysis,
-                    }
-                )
+                # 检查是否已经收集过该路径
+                abs_path = os.path.abspath(file_path)
+                if abs_path not in seen_paths:
+                    print(f"   ✅ 文件存在: {file_path}")
+                    # 记录图片信息
+                    collected_figures.append(
+                        {
+                            "figure_number": figure_number,
+                            "filename": filename,
+                            "file_path": file_path,
+                            "description": description,
+                            "analysis": analysis,
+                        }
+                    )
+                    seen_paths.add(abs_path)
+                else:
+                    print(f"   ⚠️ 跳过重复图片: {file_path}")
             else:
                 if file_path:
                     print(f"   ⚠️ 文件不存在: {file_path}")
@@ -224,7 +223,7 @@ class DataAnalysisAgent:
                 "continue": True,
             }
 
-    def analyze(self, user_input: str, files: List[str] = None, session_output_dir: str = None) -> Dict[str, Any]:
+    def analyze(self, user_input: str, files: List[str] = None, session_output_dir: str = None, reset_session: bool = True, max_rounds: int = None) -> Dict[str, Any]:
         """
         开始分析流程
 
@@ -232,59 +231,91 @@ class DataAnalysisAgent:
             user_input: 用户的自然语言需求
             files: 数据文件路径列表
             session_output_dir: 指定的会话输出目录（可选）
+            reset_session: 是否重置会话 (True: 新开启分析; False: 在现有上下文中继续)
+            max_rounds: 本次分析的最大轮数 (可选，如果不填则使用默认值)
 
         Returns:
             分析结果字典
         """
-        # 重置状态
-        self.conversation_history = []
-        self.analysis_results = []
-        self.current_round = 0
+        
+        # 确定本次运行的轮数限制
+        current_max_rounds = max_rounds if max_rounds is not None else self.max_rounds
 
-        # 创建本次分析的专用输出目录
-        if session_output_dir:
-             self.session_output_dir = session_output_dir
+        if reset_session:
+            # --- 初始化新会话 ---
+            self.conversation_history = []
+            self.analysis_results = []
+            self.current_round = 0
+            
+            # 创建本次分析的专用输出目录
+            if session_output_dir:
+                 self.session_output_dir = session_output_dir
+            else:
+                self.session_output_dir = create_session_output_dir(
+                    self.base_output_dir, user_input
+                )
+
+            # 初始化代码执行器，使用会话目录
+            self.executor = CodeExecutor(self.session_output_dir)
+
+            # 设置会话目录变量到执行环境中
+            self.executor.set_variable("session_output_dir", self.session_output_dir)
+
+            # 设用工具生成数据画像
+            data_profile = ""
+            if files:
+                print("🔍 正在生成数据画像...")
+                try:
+                    data_profile = load_and_profile_data(files)
+                    print("✅ 数据画像生成完毕")
+                except Exception as e:
+                    print(f"⚠️ 数据画像生成失败: {e}")
+            
+            # 保存到实例变量供最终报告使用
+            self.data_profile = data_profile
+
+            # 构建初始prompt
+            initial_prompt = f"""用户需求: {user_input}"""
+            if files:
+                initial_prompt += f"\n数据文件: {', '.join(files)}"
+            
+            if data_profile:
+                initial_prompt += f"\n\n{data_profile}\n\n请根据上述【数据画像】中的统计信息（如高频值、缺失率、数据范围）来制定分析策略。如果发现明显的高频问题或异常分布，请优先进行深度分析。"
+
+            print(f"🚀 开始数据分析任务")
+            print(f"📝 用户需求: {user_input}")
+            if files:
+                print(f"📁 数据文件: {', '.join(files)}")
+            print(f"📂 输出目录: {self.session_output_dir}")
+            
+            # 添加到对话历史
+            self.conversation_history.append({"role": "user", "content": initial_prompt})
+            
         else:
-            self.session_output_dir = create_session_output_dir(
-                self.base_output_dir, user_input
-            )
+            # --- 继续现有会话 ---
+            # 如果是追问，且没有指定轮数，默认减少轮数，避免过度分析
+            if max_rounds is None:
+                current_max_rounds = 10 # 追问通常不需要那么长的思考链，10轮足够
+            
+            print(f"\n🚀 继续分析任务 (追问模式)")
+            print(f"📝 后续需求: {user_input}")
+            
+            # 重置当前轮数计数器，以便给新任务足够的轮次
+            self.current_round = 0 
+            
+            # 添加到对话历史
+            # 提示Agent这是后续追问，可以简化步骤
+            follow_up_prompt = f"后续需求: {user_input}\n(注意：这是后续追问，请直接针对该问题进行分析，无需从头开始执行完整SOP。)"
+            self.conversation_history.append({"role": "user", "content": follow_up_prompt})
 
-
-        # 初始化代码执行器，使用会话目录
-        self.executor = CodeExecutor(self.session_output_dir)
-
-        # 设置会话目录变量到执行环境中
-        self.executor.set_variable("session_output_dir", self.session_output_dir)
-
-        # 设用工具生成数据画像
-        data_profile = ""
-        if files:
-            print("🔍 正在生成数据画像...")
-            data_profile = load_and_profile_data(files)
-            print("✅ 数据画像生成完毕")
-        
-        # 保存到实例变量供最终报告使用
-        self.data_profile = data_profile
-
-        # 构建初始prompt
-        initial_prompt = f"""用户需求: {user_input}"""
-        if files:
-            initial_prompt += f"\n数据文件: {', '.join(files)}"
-        
-        if data_profile:
-            initial_prompt += f"\n\n{data_profile}\n\n请根据上述【数据画像】中的统计信息（如高频值、缺失率、数据范围）来制定分析策略。如果发现明显的高频问题或异常分布，请优先进行深度分析。"
-
-        print(f"🚀 开始数据分析任务")
-        print(f"📝 用户需求: {user_input}")
-        if files:
-            print(f"📁 数据文件: {', '.join(files)}")
-        print(f"📂 输出目录: {self.session_output_dir}")
-        print(f"🔢 最大轮数: {self.max_rounds}")
+        print(f"🔢 本次最大轮数: {current_max_rounds}")
         if self.force_max_rounds:
-            print(f"⚡ 强制模式: 将运行满 {self.max_rounds} 轮（忽略AI完成信号）")
+            print(f"⚡ 强制模式: 将运行满 {current_max_rounds} 轮（忽略AI完成信号）")
         print("=" * 60)
-        # 添加到对话历史
-        self.conversation_history.append({"role": "user", "content": initial_prompt})
+        
+        # 保存原始 max_rounds 以便恢复（虽然 analyze 结束后不需要恢复，但为了逻辑严谨）
+        original_max_rounds = self.max_rounds
+        self.max_rounds = current_max_rounds
 
         while self.current_round < self.max_rounds:
             self.current_round += 1
@@ -311,6 +342,15 @@ class DataAnalysisAgent:
                 process_result = self._process_response(response)
 
                 # 根据处理结果决定是否继续（仅在非强制模式下）
+                if process_result.get("action") == "invalid_response":
+                    consecutive_failures += 1
+                    print(f"⚠️ 连续失败次数: {consecutive_failures}/3")
+                    if consecutive_failures >= 3:
+                        print(f"❌ 连续3次无法获取有效响应，分析终止。请检查网络或配置。")
+                        break
+                else:
+                    consecutive_failures = 0  # 重置计数器
+
                 if not self.force_max_rounds and not process_result.get(
                     "continue", True
                 ):
@@ -406,6 +446,35 @@ class DataAnalysisAgent:
 
         print(f"\n📊 开始生成最终分析报告...")
         print(f"📂 输出目录: {self.session_output_dir}")
+        
+        # --- 自动补全/发现图片机制 ---
+        # 扫描目录下所有的png文件
+        try:
+            import glob
+            existing_pngs = glob.glob(os.path.join(self.session_output_dir, "*.png"))
+            
+            # 获取已收集的图片路径集合
+            collected_paths = set()
+            for fig in all_figures:
+                if fig.get("file_path"):
+                    collected_paths.add(os.path.abspath(fig.get("file_path")))
+            
+            # 检查是否有漏网之鱼
+            for png_path in existing_pngs:
+                abs_png_path = os.path.abspath(png_path)
+                if abs_png_path not in collected_paths:
+                    print(f"🔍 [自动发现] 补充未显式收集的图片: {os.path.basename(png_path)}")
+                    all_figures.append({
+                        "figure_number": "Auto",
+                        "filename": os.path.basename(png_path),
+                        "file_path": abs_png_path,
+                        "description": f"自动发现的分析图表: {os.path.basename(png_path)}",
+                        "analysis": "（该图表由系统自动捕获，Agent未提供具体分析文本，请结合图表标题理解）"
+                    })
+        except Exception as e:
+            print(f"⚠️ 自动发现图片失败: {e}")
+        # ---------------------------
+
         print(f"🔢 总轮数: {self.current_round}")
         print(f"📈 收集图片: {len(all_figures)} 个")
 
@@ -419,28 +488,19 @@ class DataAnalysisAgent:
                 max_tokens=16384,  # 设置较大的token限制以容纳完整报告
             )
 
-            # 解析响应，提取最终报告
-            try:
-                # 尝试解析YAML
-                yaml_data = self.llm.parse_yaml_response(response)
-                
-                # 情况1: 标准YAML格式，包含 action: analysis_complete
-                if yaml_data.get("action") == "analysis_complete":
-                     final_report_content = yaml_data.get("final_report", response)
-                
-                # 情况2: 解析成功但没字段，或者解析失败
-                else:
-                    # 如果内容看起来像Markdown报告（包含标题），直接使用
-                    if "# " in response or "## " in response:
-                        print("⚠️ 未检测到标准YAML动作，但内容疑似Markdown报告，直接采纳")
-                        final_report_content = response
-                    else:
-                        final_report_content = "LLM未返回有效报告内容"
+            # 直接使用LLM响应作为最终报告（因为我们在prompt中要求直接输出Markdown）
+            final_report_content = response
+            
+            # 兼容旧逻辑：如果意外返回了YAML，尝试解析
+            if response.strip().startswith("action:") or "final_report:" in response:
+                try:
+                    yaml_data = self.llm.parse_yaml_response(response)
+                    if yaml_data.get("action") == "analysis_complete":
+                         final_report_content = yaml_data.get("final_report", response)
+                except:
+                    pass # 解析失败则保持原样
 
-            except Exception as e:
-                # 解析完全失败，直接使用原始响应
-                print(f"⚠️ YAML解析失败 ({e})，直接使用原始响应作为报告")
-                final_report_content = response
+            print("✅ 最终报告生成完成")
 
             print("✅ 最终报告生成完成")
 
diff --git a/main.py b/main.py
index 7075138..239605e 100644
--- a/main.py
+++ b/main.py
@@ -39,11 +39,29 @@ def setup_logging(log_dir):
 
 def main():
     llm_config = LLMConfig()
-    files = ["./cleaned_data.csv"]
+    import glob
+    import os
+    # 自动查找当前目录及remotecontrol目录下的所有数据文件
+    data_extensions = ['*.csv', '*.xlsx', '*.xls']
+    search_dirs = ['jetour']
+    files = []
+    
+    for search_dir in search_dirs:
+        for ext in data_extensions:
+            pattern = os.path.join(search_dir, ext)
+            files.extend(glob.glob(pattern))
+    
+    if not files:
+        print("⚠️ 未在当前目录找到数据文件 (.csv, .xlsx)，尝试使用默认文件")
+        files = ["./cleaned_data.csv"]
+    else:
+        print(f"📂 自动识别到以下数据文件: {files}")
+
     analysis_requirement = """
 基于所有运维工单，整理一份工单健康度报告，包括但不限于对所有车联网技术支持工单的全面数据分析，
-深入挖掘工单处理过程中的关键问题、效率瓶颈及改进机会。涵盖工单状态、问题类型、模块分布、严重程度、责任人负载、车型分布、来源渠道及处理时长等多个维度。
-通过多轮交叉分析与趋势洞察，为提升车联网服务质量、优化资源配置及降低运营风险提供数据驱动的决策依据，问题总揽，高频问题、重点问题分析，输出若干个重要的统计指标，并绘制相关图表；结合图表，总结一份，车联网运维工单健康度报告，汇报给我。
+深入挖掘工单处理过程中的关键问题、效率瓶颈及改进机会。请从车型，模块，功能角度，分别展示工单数据、问题类型、模块分布、严重程度、责任人负载、车型分布、来源渠道及处理时长等多个维度。
+通过多轮交叉分析与趋势洞察，为提升车联网服务质量、优化资源配置及降低运营风险提供数据驱动的决策依据，问题总揽，高频问题、重点问题分析，输出若干个重要的统计指标，并绘制相关图表；
+结合图表，总结一份，车联网运维工单健康度报告，汇报给我。
     """
     
     # 在主函数中先创建会话目录，以便存放日志
@@ -57,12 +75,33 @@ def main():
     # 如果希望强制运行到最大轮数，设置 force_max_rounds=True
     agent = DataAnalysisAgent(llm_config, force_max_rounds=False)
 
-    report = agent.analyze(
-        user_input=analysis_requirement,
-        files=files,
-        session_output_dir=session_output_dir
-    )
-    print(report)
+    # --- 交互式分析循环 ---
+    while True:
+        # 执行分析
+        # 首次运行时 reset_session=True (默认)
+        # 后续运行时 reset_session=False
+        is_first_run = (agent.current_round == 0 and not agent.conversation_history)
+        
+        report = agent.analyze(
+            user_input=analysis_requirement,
+            files=files if is_first_run else None, # 后续轮次不需要重复传文件路径，agent已有上下文
+            session_output_dir=session_output_dir,
+            reset_session=is_first_run,
+            max_rounds=None if is_first_run else 10 # 追问时限制为10轮
+        )
+        print("\n" + "="*30 + " 当前阶段分析完成 " + "="*30)
+        
+        # 询问用户是否继续
+        print("\n💡 你可以继续对数据提出分析需求，或者输入 'exit'/'quit' 结束程序。")
+        user_response = input("👉 请输入后续分析需求 (直接回车退出): ").strip()
+        
+        if not user_response or user_response.lower() in ['exit', 'quit', 'n', 'no']:
+            print("👋 分析结束，再见！")
+            break
+        
+        # 更新需求，进入下一轮循环
+        analysis_requirement = user_response
+        print(f"\n🔄 收到新需求，正在继续分析...")
  
 
 if __name__ == "__main__":
diff --git a/merge_excel.py b/merge_excel.py
new file mode 100644
index 0000000..1894621
--- /dev/null
+++ b/merge_excel.py
@@ -0,0 +1,83 @@
+
+import pandas as pd
+import glob
+import os
+
+def merge_excel_files(source_dir="remotecontrol", output_file="merged_all_files.csv"):
+    """
+    将指定目录下的所有 Excel 文件 (.xlsx, .xls) 合并为一个 CSV 文件。
+    """
+    print(f"🔍 正在扫描目录: {source_dir} ...")
+    
+    # 支持 xlsx 和 xls
+    files_xlsx = glob.glob(os.path.join(source_dir, "*.xlsx"))
+    files_xls = glob.glob(os.path.join(source_dir, "*.xls"))
+    files = files_xlsx + files_xls
+    
+    if not files:
+        print("⚠️ 未找到 Excel 文件。")
+        return
+
+    # 按文件名中的数字进行排序 (例如: 1.xlsx, 2.xlsx, ..., 10.xlsx)
+    try:
+        files.sort(key=lambda x: int(os.path.basename(x).split('.')[0]))
+        print("🔢 已按文件名数字顺序排序")
+    except ValueError:
+        # 如果文件名不是纯数字，退回到字母排序
+        files.sort()
+        print("🔤 已按文件名包含非数字字符，使用字母顺序排序")
+
+    print(f"📂 找到 {len(files)} 个文件: {files}")
+    
+    all_dfs = []
+    for file in files:
+        try:
+            print(f"📖 读取: {file}")
+            # 使用 ExcelFile 读取所有 sheet
+            xls = pd.ExcelFile(file)
+            print(f"   📑 包含 Sheets: {xls.sheet_names}")
+            
+            file_dfs = []
+            for sheet_name in xls.sheet_names:
+                df = pd.read_excel(xls, sheet_name=sheet_name)
+                if not df.empty:
+                    print(f"   ✅ Sheet '{sheet_name}' 读取成功: {len(df)} 行")
+                    file_dfs.append(df)
+                else:
+                    print(f"   ⚠️ Sheet '{sheet_name}' 为空，跳过")
+            
+            if file_dfs:
+                # 合并该文件的所有非空 sheet
+                file_merged_df = pd.concat(file_dfs, ignore_index=True)
+                # 可选：添加一列标记来源文件
+                file_merged_df['Source_File'] = os.path.basename(file)
+                all_dfs.append(file_merged_df)
+            else:
+                 print(f"⚠️ 文件 {file} 所有 Sheet 均为空")
+
+        except Exception as e:
+            print(f"❌ 读取 {file} 失败: {e}")
+
+    if all_dfs:
+        print("🔄 正在合并数据...")
+        merged_df = pd.concat(all_dfs, ignore_index=True)
+        
+        # 按 SendTime 排序
+        if 'SendTime' in merged_df.columns:
+            print("⏳ 正在按 SendTime 排序...")
+            merged_df['SendTime'] = pd.to_datetime(merged_df['SendTime'], errors='coerce')
+            merged_df = merged_df.sort_values(by='SendTime')
+        else:
+            print("⚠️ 未找到 SendTime 列，跳过排序")
+
+        print(f"💾 保存到: {output_file}")
+        merged_df.to_csv(output_file, index=False, encoding="utf-8-sig")
+        
+        print(f"✅ 合并及排序完成！总行数: {len(merged_df)}")
+        print(f"   输出文件: {os.path.abspath(output_file)}")
+    else:
+        print("⚠️ 没有成功读取到任何数据。")
+
+if __name__ == "__main__":
+    # 如果需要在当前目录运行并合并 remotecontrol 文件夹下的内容
+    merge_excel_files(source_dir="remotecontrol", output_file="remotecontrol_merged.csv")
diff --git a/prompts.py b/prompts.py
index 85f67c4..ff3b0c4 100644
--- a/prompts.py
+++ b/prompts.py
@@ -28,7 +28,21 @@ data_analysis_system_prompt = """你是一个专业的数据分析助手，运
 2. 图片必须保存到指定的会话目录中，输出绝对路径，禁止使用plt.show()，饼图的标签全部放在图例里面，用颜色区分。
 3. 表格输出控制：超过15行只显示前5行和后5行
 4. 所有生成的图片必须保存，保存路径格式：os.path.join(session_output_dir, '图片名称.png')
-5. 中文字体设置：生成的绘图代码，涉及中文字体，必须保证生成图片不可以乱码（macOS推荐：Hiragino Sans GB, Songti SC等）
+5. 中文字体设置：生成的绘图代码，必须在开头加入以下代码以解决中文乱码问题：
+   ```python
+   import matplotlib.pyplot as plt
+   import platform
+   
+   system_name = platform.system()
+   if system_name == 'Darwin': # macOS
+       plt.rcParams['font.sans-serif'] = ['Arial Unicode MS', 'PingFang SC', 'Heiti SC', 'sans-serif']
+   elif system_name == 'Windows':
+       plt.rcParams['font.sans-serif'] = ['SimHei', 'Microsoft YaHei', 'sans-serif']
+   else: # Linux
+       plt.rcParams['font.sans-serif'] = ['WenQuanYi Micro Hei', 'SimHei', 'sans-serif']
+   
+   plt.rcParams['axes.unicode_minus'] = False # 解决负号显示问题
+   ```
 6. 输出格式严格使用YAML
 
 📁 输出目录管理：
@@ -39,53 +53,64 @@ data_analysis_system_prompt = """你是一个专业的数据分析助手，运
 - 所有生成的图片必须执行处理图片收集动作并保存，保存路径格式：os.path.join(session_output_dir, '图片名称.png')
 - 输出绝对路径：使用os.path.abspath()获取图片的完整路径
 
+🚨 **关键红线 (Critical Rules)**：
+1.  **图片保存铁律**：每次 `plt.plot()` 后**必须**紧接着调用 `plt.savefig()` 和 `plt.close()`。虽然系统有自动补救机制，但你必须显式保存每一张图。
+2.  **绝对禁止伪造数据**：无论遇到何种报错，绝对不可以使用 `pd.DataFrame({{...}})` 手动创建虚假数据来展示。如果无法读取数据，必须诚实报告错误并停止分析。
+3.  **文件存在性验证**：在读取前必须使用 `os.path.exists()` 检查文件是否存在。
+4.  **扩展名陷阱**：如果用户说是 `.xlsx` 但读取失败，请检查目录下是否有同名的 `.csv` 文件。
+
 📊 数据分析工作流程（必须严格按顺序执行）：
 
-**阶段1：数据探索（使用 generate_code 动作）**
-- 首次数据加载时尝试多种编码：['utf-8', 'gbk', 'gb18030', 'gb2312', 'latin1']
-- 特殊处理：如果读取失败，尝试指定分隔符 `sep=','` 和错误处理 `on_bad_lines='skip'` (pandas 2.0+标准)
-- 使用df.head()查看前几行数据，检查数据是否正确读取
-- 使用df.info()了解数据类型和缺失值情况
-- 重点检查：如果数值列显示为NaN但应该有值，说明读取或解析有问题
-- 使用df.dtypes查看每列的数据类型，确保日期列不是float64
-- 打印所有列名：df.columns.tolist()
-- 绝对不要假设列名，必须先查看实际的列名
+**阶段1：数据探索与智能加载（使用 generate_code 动作）**
+- **Excel文件深度加载策略**：
+    - 首选：`pd.read_excel(file, engine='openpyxl')`
+    - 失败B计划：尝试 `pd.read_excel(file, engine='openpyxl', read_only=True, data_only=True)`
+    - 失败C计划（针对扩展名错误但实际是CSV的文件）：`pd.read_csv(file)`
+    - 失败D计划（针对超大文件或格式异常）：使用 `zipfile` + `xml.etree` 手动解析 `sharedStrings.xml` 和 `sheet1.xml` (参考之前的成功案例)
+- **CSV/文本文件策略**：尝试多种编码 `['utf-8', 'gbk', 'gb18030', 'latin1']` 和分隔符 `sep=','` 或 `sep='\t'`
+- **数据验证**：
+    - 使用df.head()查看前几行
+    - 使用df.info()检查数据类型和缺失值
+    - 打印列名：`print(df.columns.tolist())`
 
 **阶段2：数据清洗和检查（使用 generate_code 动作）**
-- 日期列识别：查找包含'date', 'time', 'Date', 'Time'关键词的列
-- 日期解析：尝试多种格式 ['%d/%m/%Y', '%Y-%m-%d', '%m/%d/%Y', '%Y/%m/%d', '%d-%m-%Y']
-- 类型转换：使用pd.to_datetime()转换日期列，指定format参数和errors='coerce'
-- 空值处理：检查哪些列应该有值但显示NaN，可能是数据读取问题
-- 检查数据的时间范围和排序
-- 数据质量检查：确认数值列是否正确，字符串列是否被错误识别
+- 日期列识别与标准化：查找 'date', 'time', '创建', '关闭' 等列，统一转为 datetime格式。
+- 关键字段对齐：将 'Model', 'Car Model', '车型' 统一重命名为 '车型'；'Module', '模块' 统一重命名为 '模块'。
+- 缺失值与异常值标记：统计关键维度（车型、模块）的缺失率。
+- **多文件数据合并**：如果识别到 source_file 列，确保按文件顺序或时间列进行排序。
 
+**阶段3：多维度业务分析和可视化（使用 generate_code 动作）**
+- **必须覆盖的分析维度（基于用户IOV业务需求）**：
+    1.  **车型维度 (Vehicle Model)**：各车型工单量分布、车型-问题类型热力图。
+    2.  **模块/功能维度 (Module/Function)**：故障高发模块/功能 Top10、模块-严重程度交叉分析。
+    3.  **问题类型维度 (Issue Type)**：各类问题占比、各类问题的平均处理时长。
+    4.  **严重程度分布 (Severity)**：严重/一般问题的比重及趋势。
+    5.  **责任人负载 (Owner Load)**：责任人处理工单数量 Top10 vs 平均处理时长（效率分析）。
+    6.  **来源渠道 (Source)**：不同来源（电话、APP、后台）的工单分布及有效率。
+    7.  **处理时长 (Duration)**：处理时长分布（直方图）、超长工单特征分析。
+    8.  **文本挖掘 (Text Mining)**：基于 '问题描述' 的 N-gram 短语提取（如 "TBOX离线", "远程启动失败"），排除停用词。
+- **图表生成规则**：
+    - 每一轮只专注于生成 1-2 个重点图表。
+    - 图片保存到会话目录，严禁 `plt.show()`。
+    - 类别 > 5 时使用水平条形图。
+    - **严禁覆盖**：每个文件名必须唯一，建议加上步骤前缀，如 `01_工单量分布.png`。
 
-**阶段3：数据分析和可视化（使用 generate_code 动作）**
-- 基于实际的列名进行计算
-- 生成有意义的图表
-- 图片保存到会话专用目录中
-- 每生成一个图表后，必须打印绝对路径
-- 不要试图一次性生成所有图表。你应该将任务拆分为多个小的代码块，分批次执行。
-- 每一轮只专注于生成 1-2 个复杂的图表或 2-3 个简单的图表，确保代码正确且图片保存成功。
-- 只有在前一轮代码成功执行并保存图片后，再进行下一轮。
-- 必做分析1.  **超长工单问题类型分布**（从处理时长分布中筛选）
-    2.  **车型-问题热力图**（发现特定车型的高频故障）
-    3.  **车型分布**（整体工单在不同车型的占比）
-    4.  **处理时长分布**（直方图/KDE）
-    5.  **处理时长箱线图**（按问题类型或责任人分组，识别异常点）
-    6.  **高频关键词词云**（基于Text Cleaning和N-gram结果）
-    7.  **工单来源分布**
-    8.  **工单状态分布**
-    9.  **模块分布**
-    10. **未关闭工单状态分布**
-    11. **问题类型分布**
-    12. **严重程度分布**
-    13. **远程控制(Remote Control)问题模块分布**（专项分析）
-    14. **月度工单趋势**
-    15. **月度关闭率趋势**
-    16. **责任人分布**
-    17. **责任人工作量与效率对比**（散点图或双轴图）
-- 图片保存必须使用 `plt.savefig(path, bbox_inches='tight')`。保存后必须显示打印绝对路径。严禁使用 `plt.show()`。
+**标准化分析SOP (Standard Operating Procedure)**：
+请严格按照以下顺序执行分析，不要跳跃：
+1.  **数据质量检查**：加载数据 -> 打印 info/head -> 检查 '车型'/'模块' 列的唯一值数量。
+2.  **基础分布分析**：
+    - 生成 `01_车型分布.png` (水平条形图)
+    - 生成 `02_模块Top10分布.png` (水平条形图)
+    - 生成 `03_问题类型Top10分布.png` (水平条形图)
+3.  **时序与来源分析**：
+    - 生成 `04_工单来源分布.png` (饼图或条形图)
+    - 生成 `05_月度工单趋势.png` (折线图)
+4.  **深度交叉分析**：
+    - 生成 `06_车型_问题类型热力图.png` (Heatmap)
+    - 生成 `07_模块_严重程度堆叠图.png` (Stacked Bar)
+5.  **效率分析**：
+    - 生成 `08_处理时长分布.png` (直方图)
+    - 生成 `09_责任人效率分析.png` (散点图: 工单量 vs 平均时长)
 
 **阶段4：深度挖掘与高级分析（使用 generate_code 动作）**
 - 主动评估数据特征**：在执行前，先分析数据适合哪种高级挖掘：
@@ -147,7 +172,7 @@ data_analysis_system_prompt = """你是一个专业的数据分析助手，运
 - **可视化增强**：不要只画折线图。使用 `seaborn` 的 `pairplot`, `heatmap`, `lmplot` 等高级图表。
 
 📝 动作选择指南：
-- **需要执行Python代码** → 使用 "generate_code"
+- **需要执行代码列表** → 使用 "generate_code"
 - **已生成多个图表，需要收集分析** → 使用 "collect_figures"  
 - **所有分析完成，输出最终报告** → 使用 "analysis_complete"
 - **遇到错误需要修复代码** → 使用 "generate_code"
@@ -262,6 +287,7 @@ final_report_system_prompt = """你是一位**资深数据分析专家 (Senior D
 ### 报告结构模板使用说明 (Template Instructions)
 - **固定格式 (Format)**：所有的 Markdown 标题 (`#`, `##`)、列表项前缀 (`- **...**`)、表格表头是必须保留的**骨架**。
 - **写作指引 (Prompts)**：方括号 `[...]` 内的文字是给你的**写作提示**，请根据实际分析将其**替换**为具体内容，**不要**在最终报告中保留方括号。
+- **直接输出Markdown**：不要使用JSON或YAML包裹，直接输出Markdown内容。
 
 ---
 
@@ -314,13 +340,12 @@ final_report_system_prompt = """你是一位**资深数据分析专家 (Senior D
     - **用户与业务影响**：已导致[估算的]用户投诉上升、[功能]使用率下降、潜在[NPS下降分值]。
     - **当前缓解状态**：[如：已暂停该版本推送，影响面控制在X%。]
 
-### 3.2 [业务主题二：例如“高价值用户的核心使用场景与流失预警”]
-- **核心发现**：[例如：功能A是留存关键，但其失败率在核心用户中最高。]
+### 3.2 [车主分析：例如“高价值用户的核心使用场景与流失预警”]
+- **核心发现**：[例如：截止XXXXX，平台捷途车联网的车主XXX，新增了，功能A是留存关键，但其失败率在核心用户中最高。]
 - **现象与数据表现**：[同上结构]
 - **证据链与深度归因**：
     > ![核心功能使用与留存相关性热图](./figure3.png)
     > **每周使用功能A超过3次的用户，其90天留存率是低频用户的2.5倍**，该功能是用户粘性的关键驱动力。
-    >
     > ![该功能失败率的用户分群对比](./figure4.png)
     > 然而，正是这批高价值用户，遭遇功能A失败的概率比新用户高40%，**体验瓶颈出现在用户最依赖的环节**。
 - **问题回溯与当前影响**：[同上结构]
@@ -334,7 +359,7 @@ final_report_system_prompt = """你是一位**资深数据分析专家 (Senior D
 | **[风险2：体验一致性]** | [如：Android用户关键路径失败率为iOS的2倍] | 高 | 中 | **中高** | 应用商店差评中OS提及率上升 |
 | **[风险3：合规性]** | [描述] | 低 | 高 | **中** | [相关法规更新节点] |
 
-## 5. 改进建议与方案探讨 (Suggestions & Solutions for Review)
+## 5. 改进建议与方案探讨
 > **重要提示**：以下内容仅基于数据分析结果提出初步探讨方向。**具体实施方案、责任分配及落地时间必须由人工专家（PM/研发/运营）结合实际业务资源与约束最终确认**。
 
 | 建议方向 (Direction) | 关联问题 (Issue) | 初步方案思路 (Draft Proposal) | 需人工评估点 (Points for Human Review) |
diff --git a/requirements.txt b/requirements.txt
index c7155d8..2bf6b86 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -50,3 +50,8 @@ flake8>=6.0.0
 
 # 字体支持（用于matplotlib中文显示）
 fonttools>=4.38.0
+
+# Web Interface dependencies
+fastapi>=0.109.0
+uvicorn>=0.27.0
+python-multipart>=0.0.9
diff --git a/sort_csv.py b/sort_csv.py
new file mode 100644
index 0000000..c03a1b3
--- /dev/null
+++ b/sort_csv.py
@@ -0,0 +1,45 @@
+
+import pandas as pd
+import os
+
+def sort_csv_by_time(file_path="remotecontrol_merged.csv", time_col="SendTime"):
+    """
+    读取 CSV 文件，按时间列排序，并保存。
+    """
+    if not os.path.exists(file_path):
+        print(f"❌ 文件不存在: {file_path}")
+        return
+
+    print(f"📖 正在读取 {file_path} ...")
+    try:
+        # 读取 CSV
+        df = pd.read_csv(file_path, low_memory=False)
+        print(f"   📊 数据行数: {len(df)}")
+        
+        if time_col not in df.columns:
+            print(f"❌ 未找到时间列: {time_col}")
+            print(f"   可用列: {list(df.columns)}")
+            return
+
+        print(f"🔄 正在解析时间列 '{time_col}' ...")
+        # 转换为 datetime 对象，无法解析的设为 NaT
+        df[time_col] = pd.to_datetime(df[time_col], errors='coerce')
+        
+        # 检查无效时间
+        nat_count = df[time_col].isna().sum()
+        if nat_count > 0:
+            print(f"⚠️ 发现 {nat_count} 行无效时间数据，排序时将排在最后")
+
+        print("🔄 正在按时间排序...")
+        df_sorted = df.sort_values(by=time_col)
+        
+        print(f"💾 正在保存及覆盖文件: {file_path} ...")
+        df_sorted.to_csv(file_path, index=False, encoding="utf-8-sig")
+        
+        print("✅ 排序并保存完成！")
+        
+    except Exception as e:
+        print(f"❌处理失败: {e}")
+
+if __name__ == "__main__":
+    sort_csv_by_time()
diff --git a/start.bat b/start.bat
new file mode 100644
index 0000000..92b7064
--- /dev/null
+++ b/start.bat
@@ -0,0 +1,4 @@
+@echo off
+echo Starting IOV Data Analysis Agent...
+python bootstrap.py
+pause
diff --git a/start.sh b/start.sh
new file mode 100755
index 0000000..e62793c
--- /dev/null
+++ b/start.sh
@@ -0,0 +1,3 @@
+#!/bin/bash
+echo "Starting IOV Data Analysis Agent..."
+python3 bootstrap.py
diff --git a/start_web.bat b/start_web.bat
new file mode 100644
index 0000000..50ba277
--- /dev/null
+++ b/start_web.bat
@@ -0,0 +1,5 @@
+@echo off
+echo Starting IOV Data Analysis Agent Web Interface...
+echo Please open http://localhost:8000 in your browser.
+python -m uvicorn web.main:app --reload --host 0.0.0.0 --port 8000
+pause
diff --git a/start_web.sh b/start_web.sh
new file mode 100755
index 0000000..99558ac
--- /dev/null
+++ b/start_web.sh
@@ -0,0 +1,4 @@
+#!/bin/bash
+echo "Starting IOV Data Analysis Agent Web Interface..."
+echo "Please open http://localhost:8000 in your browser."
+python3 -m uvicorn web.main:app --reload --host 0.0.0.0 --port 8000
diff --git a/test.py b/test.py
new file mode 100644
index 0000000..ae8265b
--- /dev/null
+++ b/test.py
@@ -0,0 +1,13 @@
+
+import openai
+
+client = openai.OpenAI(
+    api_key="sk-2187174de21548b0b8b0c92129700199",
+    base_url="http://127.0.0.1:9999/v1"
+)
+
+response = client.chat.completions.create(
+    model="gemini-3-flash",
+    messages=[{"role": "user", "content": "你好，请自我介绍"}]
+)
+print(response.choices[0].message.content)
\ No newline at end of file
diff --git a/utils/code_executor.py b/utils/code_executor.py
index b3d774c..accb3de 100644
--- a/utils/code_executor.py
+++ b/utils/code_executor.py
@@ -26,7 +26,9 @@ class CodeExecutor:
         "pandas",
         "pd",
         "numpy",
+        "glob",
         "np",
+        "subprocess",
         "matplotlib",
         "matplotlib.pyplot",
         "plt",
@@ -36,6 +38,14 @@ class CodeExecutor:
         "scipy",
         "sklearn",
         "sklearn.feature_extraction.text",
+        "sklearn.preprocessing",
+        "sklearn.model_selection",
+        "sklearn.metrics",
+        "sklearn.ensemble",
+        "sklearn.linear_model",
+        "sklearn.cluster",
+        "sklearn.decomposition",
+        "sklearn.manifold",
         "statsmodels",
         "plotly",
         "dash",
@@ -230,12 +240,16 @@ from IPython.display import display
         for node in ast.walk(tree):
             if isinstance(node, ast.Import):
                 for alias in node.names:
-                    if alias.name not in self.ALLOWED_IMPORTS:
+                    # 获取根包名 (e.g. sklearn.preprocessing -> sklearn)
+                    root_package = alias.name.split('.')[0]
+                    if root_package not in self.ALLOWED_IMPORTS and alias.name not in self.ALLOWED_IMPORTS:
                         return False, f"不允许的导入: {alias.name}"
 
             elif isinstance(node, ast.ImportFrom):
-                if node.module not in self.ALLOWED_IMPORTS:
-                    return False, f"不允许的导入: {node.module}"
+                if node.module:
+                    root_package = node.module.split('.')[0]
+                    if root_package not in self.ALLOWED_IMPORTS and node.module not in self.ALLOWED_IMPORTS:
+                        return False, f"不允许的导入: {node.module}"
 
             # 检查属性访问（防止通过os.system等方式绕过）
             elif isinstance(node, ast.Attribute):
@@ -381,6 +395,33 @@ from IPython.display import display
                     except:
                         pass
 
+            # --- 自动保存机制 start ---
+            # 检查是否有未关闭的图片，如果有，自动保存
+            try:
+                open_fig_nums = plt.get_fignums()
+                if open_fig_nums:
+                    for fig_num in open_fig_nums:
+                        fig = plt.figure(fig_num)
+                        # 生成自动保存的文件名
+                        auto_filename = f"autosave_fig_{self.image_counter}_{fig_num}.png"
+                        auto_filepath = os.path.join(self.output_dir, auto_filename)
+                        
+                        try:
+                            # 尝试保存
+                            fig.savefig(auto_filepath, bbox_inches='tight')
+                            print(f"💾 [Auto-Save] 检测到未闭合图表，已安全保存至: {auto_filepath}")
+                            
+                            # 添加到输出中，告知Agent
+                            output += f"\n[Auto-Save] ⚠️ 检测到Figure {fig_num}未关闭，系统已自动保存为: {auto_filename}"
+                            self.image_counter += 1
+                        except Exception as e:
+                            print(f"⚠️ [Auto-Save] 保存失败: {e}")
+                        finally:
+                            plt.close(fig_num)
+            except Exception as e:
+                print(f"⚠️ [Auto-Save Global] 异常: {e}")
+            # --- 自动保存机制 end ---
+
             return {
                 "success": True,
                 "output": output,
diff --git a/utils/llm_helper.py b/utils/llm_helper.py
index f24d967..cf88b6c 100644
--- a/utils/llm_helper.py
+++ b/utils/llm_helper.py
@@ -75,7 +75,8 @@ class LLMHelper:
             else:
                 yaml_content = response.strip()
             
-            return yaml.safe_load(yaml_content)
+            parsed = yaml.safe_load(yaml_content)
+            return parsed if parsed is not None else {}
         except Exception as e:
             print(f"YAML解析失败: {e}")
             print(f"原始响应: {response}")
diff --git a/web/main.py b/web/main.py
new file mode 100644
index 0000000..bc428f9
--- /dev/null
+++ b/web/main.py
@@ -0,0 +1,404 @@
+
+import sys
+import os
+import threading
+import glob
+import uuid
+import json
+from typing import Optional, Dict, List
+from fastapi import FastAPI, UploadFile, File, BackgroundTasks, HTTPException, Query
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.staticfiles import StaticFiles
+from fastapi.responses import FileResponse, JSONResponse
+from pydantic import BaseModel
+
+# Add parent directory to path to import agent modules
+sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from data_analysis_agent import DataAnalysisAgent
+from config.llm_config import LLMConfig
+from utils.create_session_dir import create_session_output_dir
+from merge_excel import merge_excel_files
+from sort_csv import sort_csv_by_time
+
+app = FastAPI(title="IOV Data Analysis Agent")
+
+# CORS
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# --- Session Management ---
+
+class SessionData:
+    def __init__(self, session_id: str):
+        self.session_id = session_id
+        self.is_running = False
+        self.output_dir: Optional[str] = None
+        self.generated_report: Optional[str] = None
+        self.log_file: Optional[str] = None
+        self.analysis_results: List[Dict] = [] # Store analysis results for gallery
+
+class SessionManager:
+    def __init__(self):
+        self.sessions: Dict[str, SessionData] = {}
+        self.lock = threading.Lock()
+
+    def create_session(self) -> str:
+        with self.lock:
+            session_id = str(uuid.uuid4())
+            self.sessions[session_id] = SessionData(session_id)
+            return session_id
+
+    def get_session(self, session_id: str) -> Optional[SessionData]:
+        return self.sessions.get(session_id)
+
+    def list_sessions(self):
+        return list(self.sessions.keys())
+
+session_manager = SessionManager()
+
+# Mount static files
+os.makedirs("web/static", exist_ok=True)
+os.makedirs("uploads", exist_ok=True)
+os.makedirs("outputs", exist_ok=True)
+
+app.mount("/static", StaticFiles(directory="web/static"), name="static")
+app.mount("/outputs", StaticFiles(directory="outputs"), name="outputs")
+
+# --- Helper Functions ---
+
+def run_analysis_task(session_id: str, files: list, user_requirement: str):
+    """
+    Runs the analysis agent in a background thread for a specific session.
+    """
+    session = session_manager.get_session(session_id)
+    if not session:
+        print(f"Error: Session {session_id} not found in background task.")
+        return
+
+    session.is_running = True
+    try:
+        # Create session directory
+        base_output_dir = "outputs"
+        # We enforce a specific directory naming convention or let the util handle it
+        # ideally we map session_id to the directory
+        # For now, let's use the standard utility but we might lose the direct mapping if not careful
+        # Let's trust the return value
+        session_output_dir = create_session_output_dir(base_output_dir, user_requirement)
+        session.output_dir = session_output_dir
+        
+        # Initialize Log capturing
+        session.log_file = os.path.join(session_output_dir, "process.log")
+        
+        # Thread-safe logging requires a bit of care. 
+        # Since we are running in a thread, redirecting sys.stdout globally is BAD for multi-session.
+        # However, for this MVP, if we run multiple sessions concurrently, their logs will mix in stdout.
+        # BUT we are writing to specific log files.
+        # We need a logger that writes to the session's log file.
+        # And the Agent needs to use that logger.
+        # Currently the Agent uses print(). 
+        # To support true concurrent logging without mixing, we'd need to refactor Agent to use a logger instance.
+        # LIMITATION: For now, we accept that stdout redirection intercepts EVERYTHING. 
+        # So multiple concurrent sessions is risky with global stdout redirection.
+        # A safer approach for now: We won't redirect stdout globally for multi-session support 
+        # unless we lock execution to one at a time.
+        # OR: We just rely on the fact that we might only run one analysis at a time mostly.
+        # Let's try to just write to the log file explicitly if we could, but we can't change Agent easily right now.
+        # Compromise: We will continue to use global redirection but acknowledge it's not thread-safe for output.
+        # A better way: Modify Agent to accept a 'log_callback'.
+        # For this refactor, let's stick to the existing pattern but bind it to the thread if possible? No.
+        
+        # We will wrap the execution with a simple File Logger that appends to the distinct file.
+        # But sys.stdout is global.
+        # We will assume single concurrent analysis for safety, or accept mixed terminal output but separate file logs?
+        # Actually, if we swap sys.stdout, it affects all threads.
+        # So we MUST NOT swap sys.stdout if we want concurrency.
+        # If we don't swap stdout, we don't capture logs to file unless Agent does it.
+        # The Agent code has `print`. 
+        # Correct fix: Refactor Agent to use `logging` module or pass a printer.
+        # Given the scope, let's just hold the lock (serialize execution) OR allow mixing in terminal 
+        # but try to capture to file?
+        # Let's just write to the file.
+        
+        with open(session.log_file, "w", encoding="utf-8") as f:
+            f.write(f"--- Session {session_id} Started ---\n")
+
+        # We will create a custom print function that writes to the file
+        # And monkeypatch builtins.print? No, that's too hacky.
+        # Let's just use the stdout redirector, but acknowledge only one active session at a time is safe.
+        # We can implement a crude lock for now.
+        
+        class FileLogger:
+            def __init__(self, filename):
+                self.terminal = sys.__stdout__
+                self.log = open(filename, "a", encoding="utf-8", buffering=1)
+                
+            def write(self, message):
+                self.terminal.write(message)
+                self.log.write(message)
+                
+            def flush(self):
+                self.terminal.flush()
+                self.log.flush()
+                
+            def close(self):
+                self.log.close()
+
+        logger = FileLogger(session.log_file)
+        sys.stdout = logger # Global hijack!
+        
+        try:
+            llm_config = LLMConfig()
+            agent = DataAnalysisAgent(llm_config, force_max_rounds=False, output_dir=base_output_dir)
+            
+            result = agent.analyze(
+                user_input=user_requirement,
+                files=files,
+                session_output_dir=session_output_dir
+            )
+            
+            session.generated_report = result.get("report_file_path", None)
+            session.analysis_results = result.get("analysis_results", [])
+            
+            # Save results to json for persistence
+            with open(os.path.join(session_output_dir, "results.json"), "w") as f:
+                json.dump(session.analysis_results, f, default=str)
+            
+        except Exception as e:
+            print(f"Error during analysis: {e}")
+        finally:
+            sys.stdout = logger.terminal
+            logger.close()
+            
+    except Exception as e:
+        print(f"System Error: {e}")
+    finally:
+        session.is_running = False
+
+# --- Pydantic Models ---
+
+class StartRequest(BaseModel):
+    requirement: str
+
+# --- API Endpoints ---
+
+@app.get("/")
+async def read_root():
+    return FileResponse("web/static/index.html")
+
+@app.post("/api/upload")
+async def upload_files(files: list[UploadFile] = File(...)):
+    saved_files = []
+    for file in files:
+        file_location = f"uploads/{file.filename}"
+        with open(file_location, "wb+") as file_object:
+            file_object.write(file.file.read())
+        saved_files.append(file_location)
+    return {"info": f"Saved {len(saved_files)} files", "paths": saved_files}
+
+@app.post("/api/start")
+async def start_analysis(request: StartRequest, background_tasks: BackgroundTasks):
+    session_id = session_manager.create_session()
+    
+    files = glob.glob("uploads/*.csv")
+    if not files:
+        if os.path.exists("cleaned_data.csv"):
+            files = ["cleaned_data.csv"]
+        else:
+            raise HTTPException(status_code=400, detail="No CSV files found")
+            
+    files = [os.path.abspath(f) for f in files] # Only use absolute paths
+    
+    background_tasks.add_task(run_analysis_task, session_id, files, request.requirement)
+    return {"status": "started", "session_id": session_id}
+
+@app.get("/api/status")
+async def get_status(session_id: str = Query(..., description="Session ID")):
+    session = session_manager.get_session(session_id)
+    if not session:
+        raise HTTPException(status_code=404, detail="Session not found")
+        
+    log_content = ""
+    if session.log_file and os.path.exists(session.log_file):
+        with open(session.log_file, "r", encoding="utf-8") as f:
+            log_content = f.read()
+    
+    return {
+        "is_running": session.is_running,
+        "log": log_content,
+        "has_report": session.generated_report is not None,
+        "report_path": session.generated_report
+    }
+
+@app.get("/api/report")
+async def get_report(session_id: str = Query(..., description="Session ID")):
+    session = session_manager.get_session(session_id)
+    if not session:
+         raise HTTPException(status_code=404, detail="Session not found")
+         
+    if not session.generated_report or not os.path.exists(session.generated_report):
+         return {"content": "Report not ready."}
+         
+    with open(session.generated_report, "r", encoding="utf-8") as f:
+        content = f.read()
+        
+    # Fix image paths
+    relative_session_path = os.path.relpath(session.output_dir, os.getcwd()) 
+    web_base_path = f"/{relative_session_path}"
+    content = content.replace("](./", f"]({web_base_path}/")
+    
+    return {"content": content, "base_path": web_base_path}
+
+@app.get("/api/figures")
+async def get_figures(session_id: str = Query(..., description="Session ID")):
+    session = session_manager.get_session(session_id)
+    if not session:
+         raise HTTPException(status_code=404, detail="Session not found")
+    
+    # We can try to get from memory first
+    results = session.analysis_results
+    
+    # If empty in memory (maybe server restarted but files exist?), try load json
+    if not results and session.output_dir:
+        json_path = os.path.join(session.output_dir, "results.json")
+        if os.path.exists(json_path):
+            with open(json_path, 'r') as f:
+                results = json.load(f)
+
+    # Extract collected figures
+    figures = []
+    
+    # We iterate over analysis results to find 'collect_figures' actions
+    if results:
+        for item in results:
+            if item.get("action") == "collect_figures":
+                collected = item.get("collected_figures", [])
+                for fig in collected:
+                    # Enrich with web path
+                    if session.output_dir:
+                         # Assume filename is present
+                         fname = fig.get("filename")
+                         relative_session_path = os.path.relpath(session.output_dir, os.getcwd()) 
+                         fig["web_url"] = f"/{relative_session_path}/{fname}"
+                    figures.append(fig)
+            
+            # Also check for 'generate_code' results that might have implicit figures if we parse them
+            # But the 'collect_figures' action is the reliable source as per agent design
+            
+    # Auto-discovery fallback if list is empty but pngs exist?
+    if not figures and session.output_dir:
+        # Simple scan
+        pngs = glob.glob(os.path.join(session.output_dir, "*.png"))
+        for p in pngs:
+            fname = os.path.basename(p)
+            relative_session_path = os.path.relpath(session.output_dir, os.getcwd())
+            figures.append({
+                "filename": fname,
+                "description": "Auto-discovered image",
+                "analysis": "No analysis available",
+                "web_url": f"/{relative_session_path}/{fname}"
+            })
+
+    return {"figures": figures}
+
+@app.get("/api/export")
+async def export_report(session_id: str = Query(..., description="Session ID")):
+    session = session_manager.get_session(session_id)
+    if not session or not session.output_dir:
+        raise HTTPException(status_code=404, detail="Session not found")
+        
+    import zipfile
+    import tempfile
+    from datetime import datetime
+    
+    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+    zip_filename = f"report_{timestamp}.zip"
+    
+    export_dir = "outputs"
+    os.makedirs(export_dir, exist_ok=True)
+    temp_zip_path = os.path.join(export_dir, zip_filename)
+    
+    with zipfile.ZipFile(temp_zip_path, "w", zipfile.ZIP_DEFLATED) as zf:
+        for root, dirs, files in os.walk(session.output_dir):
+            for file in files:
+                if file.endswith(('.md', '.png', '.csv', '.log', '.json', '.yaml')):
+                    abs_path = os.path.join(root, file)
+                    rel_path = os.path.relpath(abs_path, session.output_dir)
+                    zf.write(abs_path, arcname=rel_path)
+                    
+    return FileResponse(
+        path=temp_zip_path, 
+        filename=zip_filename,
+        media_type='application/zip'
+    )
+
+# --- Tools API ---
+
+class ToolRequest(BaseModel):
+    source_dir: Optional[str] = "uploads"
+    output_filename: Optional[str] = "merged_output.csv"
+    target_file: Optional[str] = None
+
+@app.post("/api/tools/merge")
+async def tool_merge_excel(req: ToolRequest):
+    """
+    Trigger Excel Merge Tool
+    """
+    try:
+        source = req.source_dir
+        output = req.output_filename
+        
+        import asyncio
+        loop = asyncio.get_event_loop()
+        
+        await loop.run_in_executor(None, lambda: merge_excel_files(source, output))
+        
+        output_abs = os.path.abspath(output)
+        if os.path.exists(output_abs):
+            return {"status": "success", "message": "Merge completed", "output_file": output_abs}
+            
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+@app.post("/api/tools/sort")
+async def tool_sort_csv(req: ToolRequest):
+    """
+    Trigger CSV Sort Tool
+    """
+    try:
+        target = req.target_file
+        if not target:
+             raise HTTPException(status_code=400, detail="Target file required")
+             
+        import asyncio
+        loop = asyncio.get_event_loop()
+        
+        await loop.run_in_executor(None, lambda: sort_csv_by_time(target))
+        
+        return {"status": "success", "message": f"Sorted {target} by time"}
+    except Exception as e:
+         raise HTTPException(status_code=500, detail=str(e))
+
+# --- Help API ---
+
+@app.get("/api/help/troubleshooting")
+async def get_troubleshooting_guide():
+    """
+    Returns the content of troubleshooting_guide.md
+    """
+    guide_path = os.path.expanduser("~/.gemini/antigravity/brain/3ff617fe-5f27-4ab8-b61b-c634f2e75255/troubleshooting_guide.md")
+    
+    if not os.path.exists(guide_path):
+        return {"content": "# Troubleshooting Guide Not Found\n\nCould not locate the guide artifact."}
+        
+    try:
+        with open(guide_path, "r", encoding="utf-8") as f:
+            content = f.read()
+        return {"content": content}
+    except Exception as e:
+        return {"content": f"# Error Loading Guide\n\n{e}"}
diff --git a/web/static/index.html b/web/static/index.html
new file mode 100644
index 0000000..687a624
--- /dev/null
+++ b/web/static/index.html
@@ -0,0 +1,175 @@
+<!DOCTYPE html>
+<html lang="en">
+
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>IOV Data Analysis Agent</title>
+    <link rel="stylesheet" href="/static/style.css?v=2.0">
+
+    <!-- Google Fonts -->
+    <link rel="preconnect" href="https://fonts.googleapis.com">
+    <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+    <link href="https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700&display=swap" rel="stylesheet">
+    <!-- Markdown Parser -->
+    <script src="https://cdn.jsdelivr.net/npm/marked/marked.min.js"></script>
+    <!-- Font Awesome -->
+    <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.4.0/css/all.min.css">
+</head>
+
+<body>
+    <div class="app-container">
+        <!-- Sidebar -->
+        <aside class="sidebar">
+            <div class="brand">
+                <i class="fa-solid fa-robot"></i>
+                <h1>IOV Agent</h1>
+            </div>
+
+            <!-- Navigation -->
+            <nav class="main-nav">
+                <button class="nav-btn active" onclick="switchView('analysis')">
+                    <i class="fa-solid fa-chart-line"></i> Analysis
+                </button>
+                <button class="nav-btn" onclick="switchView('tools')">
+                    <i class="fa-solid fa-toolbox"></i> Data Tools
+                </button>
+                <button class="nav-btn" onclick="switchView('gallery')">
+                    <i class="fa-solid fa-images"></i> Gallery
+                </button>
+                <button class="nav-btn" onclick="switchView('help')">
+                    <i class="fa-solid fa-circle-question"></i> Help
+                </button>
+            </nav>
+
+            <!-- Analysis Controls (Visible only in Analysis View) -->
+            <div id="analysisControls" class="control-group-wrapper">
+                <div class="control-group">
+                    <h3>1. Data Upload</h3>
+                    <div class="upload-zone" id="uploadZone">
+                        <i class="fa-solid fa-cloud-arrow-up"></i>
+                        <p>Drag & Drop CSV/Excel</p>
+                        <input type="file" id="fileInput" accept=".csv,.xlsx,.xls" multiple hidden>
+                        <button class="btn secondary" onclick="document.getElementById('fileInput').click()">Select
+                            Files</button>
+                        <div id="fileList" class="file-list"></div>
+                    </div>
+                </div>
+
+                <div class="control-group">
+                    <h3>2. Analysis Config</h3>
+                    <textarea id="requirementInput"
+                        placeholder="Enter your analysis requirement here...">基于所有运维工单，整理一份工单健康度报告...</textarea>
+                    <button id="startBtn" class="btn primary">
+                        <i class="fa-solid fa-play"></i> Start Analysis
+                    </button>
+                </div>
+            </div>
+
+            <div class="status-indicator">
+                <span id="statusDot" class="dot"></span>
+                <span id="statusText">Idle</span>
+            </div>
+        </aside>
+
+        <!-- Main Content Area -->
+        <main class="main-content">
+
+            <!-- VIEW: ANALYSIS -->
+            <div id="viewAnalysis" class="view-section active">
+                <div class="tabs">
+                    <button class="tab-btn active" onclick="switchTab('logs')">
+                        <i class="fa-solid fa-terminal"></i> Live Logs
+                    </button>
+                    <button class="tab-btn" onclick="switchTab('report')">
+                        <i class="fa-solid fa-file-invoice-dollar"></i> Report
+                    </button>
+                    <div style="margin-left: auto; display: flex; align-items: center;">
+                        <button id="exportBtn" class="btn secondary"
+                            style="padding: 0.4rem 1rem; font-size: 0.8rem; width: auto;" onclick="triggerExport()">
+                            <i class="fa-solid fa-download"></i> Export Report (ZIP)
+                        </button>
+                    </div>
+                </div>
+
+                <!-- Log View -->
+                <div id="logsTab" class="tab-content active">
+                    <div class="terminal-window">
+                        <pre id="logOutput">Waiting to start...</pre>
+                    </div>
+                </div>
+
+                <!-- Report View -->
+                <div id="reportTab" class="tab-content">
+                    <div id="reportContainer" class="markdown-body">
+                        <div class="empty-state">
+                            <i class="fa-solid fa-chart-simple"></i>
+                            <p>No report generated yet.</p>
+                        </div>
+                    </div>
+                </div>
+            </div>
+
+            <!-- VIEW: TOOLS -->
+            <div id="viewTools" class="view-section" style="display: none;">
+                <div class="tools-grid">
+                    <!-- Tool Card 1: Merge Excel -->
+                    <div class="tool-card">
+                        <div class="tool-icon"><i class="fa-solid fa-file-csv"></i></div>
+                        <h3>Excel to CSV Merger</h3>
+                        <p>Merge multiple .xlsx files from the uploads directory into a single CSV for analysis.</p>
+                        <div class="tool-actions">
+                            <input type="text" id="mergeSource" value="uploads" placeholder="Source Directory"
+                                class="input-sm">
+                            <button class="btn secondary" onclick="triggerMerge()">
+                                <i class="fa-solid fa-bolt"></i> Merge Now
+                            </button>
+                        </div>
+                        <div id="mergeResult" class="tool-result"></div>
+                    </div>
+
+                    <!-- Tool Card 2: Sort CSV -->
+                    <div class="tool-card">
+                        <div class="tool-icon"><i class="fa-solid fa-arrow-down-a-z"></i></div>
+                        <h3>Time Sorter</h3>
+                        <p>Sort a CSV file by 'SendTime' or time column to fix ordering issues.</p>
+                        <div class="tool-actions">
+                            <input type="text" id="sortTarget" value="cleaned_data.csv" placeholder="Target Filename"
+                                class="input-sm">
+                            <button class="btn secondary" onclick="triggerSort()">
+                                <i class="fa-solid fa-sort"></i> Sort Now
+                            </button>
+                        </div>
+                        <div id="sortResult" class="tool-result"></div>
+                    </div>
+                </div>
+            </div>
+
+            <!-- VIEW: GALLERY -->
+            <div id="viewGallery" class="view-section" style="display: none;">
+                <div id="galleryContainer" class="gallery-grid">
+                    <div class="empty-state">
+                        <i class="fa-solid fa-images"></i>
+                        <p>No images generated yet.<br>Start an analysis to see results here.</p>
+                    </div>
+                </div>
+            </div>
+
+            <!-- VIEW: HELP -->
+            <div id="viewHelp" class="view-section" style="display: none;">
+                <div class="help-header">
+                    <h2>Troubleshooting Guide</h2>
+                    <button class="btn secondary" onclick="fetchHelp()">Refresh</button>
+                </div>
+                <div id="helpContainer" class="markdown-body help-body">
+                    <p>Loading guide...</p>
+                </div>
+            </div>
+
+        </main>
+    </div>
+
+    <script src="/static/script.js"></script>
+</body>
+
+</html>
\ No newline at end of file
diff --git a/web/static/script.js b/web/static/script.js
new file mode 100644
index 0000000..396b1c0
--- /dev/null
+++ b/web/static/script.js
@@ -0,0 +1,362 @@
+
+const uploadZone = document.getElementById('uploadZone');
+const fileInput = document.getElementById('fileInput');
+const fileList = document.getElementById('fileList');
+const startBtn = document.getElementById('startBtn');
+const requirementInput = document.getElementById('requirementInput');
+const statusDot = document.getElementById('statusDot');
+const statusText = document.getElementById('statusText');
+const logOutput = document.getElementById('logOutput');
+const reportContainer = document.getElementById('reportContainer');
+const galleryContainer = document.getElementById('galleryContainer');
+
+let isRunning = false;
+let pollingInterval = null;
+let currentSessionId = null;
+
+// --- Upload Logic ---
+if (uploadZone) {
+    uploadZone.addEventListener('dragover', (e) => {
+        e.preventDefault();
+        uploadZone.classList.add('dragover');
+    });
+    uploadZone.addEventListener('dragleave', () => uploadZone.classList.remove('dragover'));
+    uploadZone.addEventListener('drop', (e) => {
+        e.preventDefault();
+        uploadZone.classList.remove('dragover');
+        handleFiles(e.dataTransfer.files);
+    });
+}
+if (fileInput) {
+    fileInput.addEventListener('change', (e) => handleFiles(e.target.files));
+}
+
+async function handleFiles(files) {
+    if (files.length === 0) return;
+
+    fileList.innerHTML = '';
+    const formData = new FormData();
+
+    for (const file of files) {
+        formData.append('files', file);
+        const fileItem = document.createElement('div');
+        fileItem.innerText = `📄 ${file.name}`;
+        fileItem.style.fontSize = '0.8rem';
+        fileItem.style.color = '#fff';
+        fileList.appendChild(fileItem);
+    }
+
+    try {
+        const res = await fetch('/api/upload', {
+            method: 'POST',
+            body: formData
+        });
+        if (res.ok) {
+            const data = await res.json();
+            console.log('Upload success:', data);
+        } else {
+            alert('Upload failed');
+        }
+    } catch (e) {
+        console.error(e);
+        alert('Upload failed');
+    }
+}
+
+// --- Start Analysis ---
+if (startBtn) {
+    startBtn.addEventListener('click', async () => {
+        if (isRunning) return;
+
+        const requirement = requirementInput.value.trim();
+        if (!requirement) {
+            alert('Please enter analysis requirement');
+            return;
+        }
+
+        setRunningState(true);
+
+        try {
+            const res = await fetch('/api/start', {
+                method: 'POST',
+                headers: { 'Content-Type': 'application/json' },
+                body: JSON.stringify({ requirement })
+            });
+
+            if (res.ok) {
+                const data = await res.json();
+                currentSessionId = data.session_id; // Store Session ID
+                console.log("Started Session:", currentSessionId);
+
+                startPolling();
+                switchTab('logs');
+            } else {
+                const err = await res.json();
+                alert('Failed to start: ' + err.detail);
+                setRunningState(false);
+            }
+        } catch (e) {
+            console.error(e);
+            alert('Error starting analysis');
+            setRunningState(false);
+        }
+    });
+}
+
+function setRunningState(running) {
+    isRunning = running;
+    startBtn.disabled = running;
+    startBtn.innerHTML = running ? '<i class="fa-solid fa-spinner fa-spin"></i> Running...' : '<i class="fa-solid fa-play"></i> Start Analysis';
+
+    if (running) {
+        statusDot.className = 'dot running';
+        statusText.innerText = 'Analysis in Progress';
+    } else {
+        statusDot.className = 'dot done';
+        statusText.innerText = 'Completed';
+    }
+}
+
+// --- Status Polling ---
+function startPolling() {
+    if (pollingInterval) clearInterval(pollingInterval);
+    if (!currentSessionId) return;
+
+    pollingInterval = setInterval(async () => {
+        try {
+            const res = await fetch(`/api/status?session_id=${currentSessionId}`);
+            if (!res.ok) return;
+            const data = await res.json();
+
+            // Update Logs
+            logOutput.innerText = data.log || "Waiting for logs...";
+            // Auto scroll to bottom
+            const term = document.querySelector('.terminal-window');
+            if (term) term.scrollTop = term.scrollHeight;
+
+            if (!data.is_running && isRunning) {
+                // Just finished
+                setRunningState(false);
+                clearInterval(pollingInterval);
+
+                if (data.has_report) {
+                    loadReport();
+                    loadGallery(); // Load images when done
+                    switchTab('report');
+                }
+            }
+        } catch (e) {
+            console.error('Polling error', e);
+        }
+    }, 2000);
+}
+
+async function loadReport() {
+    if (!currentSessionId) return;
+    try {
+        const res = await fetch(`/api/report?session_id=${currentSessionId}`);
+        const data = await res.json();
+
+        // Render Markdown
+        reportContainer.innerHTML = marked.parse(data.content);
+
+        // Fix images relative path for display if needed
+        // The backend should return a base_path or we handle it here
+        if (data.base_path) {
+            // Backend already handled replacement? 
+            // The backend returns content with updated paths.
+        }
+
+    } catch (e) {
+        reportContainer.innerHTML = '<p class="error">Failed to load report.</p>';
+        console.error(e);
+    }
+}
+
+async function loadGallery() {
+    if (!currentSessionId) return;
+
+    // Switch to gallery view logic if we were already there
+    // But this is just data loading
+    try {
+        const res = await fetch(`/api/figures?session_id=${currentSessionId}`);
+        const data = await res.json();
+
+        const galleryGrid = document.getElementById('galleryContainer');
+        if (!data.figures || data.figures.length === 0) {
+            galleryGrid.innerHTML = `
+                <div class="empty-state">
+                    <i class="fa-solid fa-images"></i>
+                    <p>No images generated in this session.</p>
+                </div>
+            `;
+            return;
+        }
+
+        let html = '';
+        data.figures.forEach(fig => {
+            html += `
+                <div class="gallery-card">
+                    <div class="img-wrapper">
+                        <img src="${fig.web_url}" alt="${fig.filename}" onclick="window.open('${fig.web_url}', '_blank')">
+                    </div>
+                    <div class="card-content">
+                        <h4>${fig.filename}</h4>
+                        <p class="desc">${fig.description || 'No description'}</p>
+                        ${fig.analysis ? `<div class="analysis-box"><i class="fa-solid fa-magnifying-glass"></i> ${fig.analysis}</div>` : ''}
+                    </div>
+                </div>
+            `;
+        });
+        galleryGrid.innerHTML = html;
+
+    } catch (e) {
+        console.error("Gallery load failed", e);
+    }
+}
+
+// --- Export Report ---
+window.triggerExport = async function () {
+    if (!currentSessionId) {
+        alert("No active session to export.");
+        return;
+    }
+
+    const btn = document.getElementById('exportBtn');
+    const originalText = btn.innerHTML;
+    btn.innerHTML = '<i class="fa-solid fa-spinner fa-spin"></i> Zipping...';
+    btn.disabled = true;
+
+    try {
+        // Trigger download
+        // We can't use fetch for file download easily if we want browser to handle save dialog
+        // So we create a link approach or check status first
+
+        // Check if download is possible
+        const url = `/api/export?session_id=${currentSessionId}`;
+        const res = await fetch(url, { method: 'HEAD' });
+        if (res.ok) {
+            window.location.href = url;
+        } else {
+            alert("Export failed: No data available.");
+        }
+    } catch (e) {
+        alert("Export failed: " + e.message);
+    } finally {
+        btn.innerHTML = originalText;
+        btn.disabled = false;
+    }
+}
+
+
+// --- Tabs (Inner) ---
+window.switchTab = function (tabName) {
+    document.querySelectorAll('.tab-btn').forEach(b => b.classList.remove('active'));
+    document.querySelectorAll('.tab-content').forEach(c => c.classList.remove('active'));
+
+    const btn = document.querySelector(`button[onclick="switchTab('${tabName}')"]`);
+    if (btn) btn.classList.add('active');
+
+    const tab = document.getElementById(`${tabName}Tab`);
+    if (tab) tab.classList.add('active');
+}
+
+// --- View Navigation (Outer Sidebar) ---
+window.switchView = function (viewName) {
+    // 1. Update Sidebar Buttons
+    document.querySelectorAll('.nav-btn').forEach(b => b.classList.remove('active'));
+    const navBtn = document.querySelector(`button[onclick="switchView('${viewName}')"]`);
+    if (navBtn) navBtn.classList.add('active');
+
+    // 2. Switch Main Content Sections
+    const viewId = 'view' + viewName.charAt(0).toUpperCase() + viewName.slice(1);
+
+    document.querySelectorAll('.view-section').forEach(v => {
+        // Hide all
+        v.style.display = 'none';
+        v.classList.remove('active');
+    });
+
+    const activeView = document.getElementById(viewId);
+    if (activeView) {
+        // Analysis view uses flex for layout (logs scrolling), others use block
+        if (viewName === 'analysis') {
+            activeView.style.display = 'flex';
+            activeView.style.flexDirection = 'column';
+        } else {
+            activeView.style.display = 'block';
+            // If switching to gallery, reload it if session exists
+            if (viewName === 'gallery' && currentSessionId) {
+                loadGallery();
+            }
+        }
+        activeView.classList.add('active');
+    }
+
+    // 3. Toggle Sidebar Controls (only show Analysis Controls in analysis view)
+    const analysisControls = document.getElementById('analysisControls');
+    if (analysisControls) {
+        analysisControls.style.display = (viewName === 'analysis') ? 'block' : 'none';
+    }
+}
+
+// --- Tool Functions ---
+window.triggerMerge = async function () {
+    const sourceDir = document.getElementById('mergeSource').value;
+    const resultDiv = document.getElementById('mergeResult');
+
+    resultDiv.innerHTML = '<i class="fa-solid fa-spinner fa-spin"></i> Merging...';
+
+    try {
+        const res = await fetch('/api/tools/merge', {
+            method: 'POST',
+            headers: { 'Content-Type': 'application/json' },
+            body: JSON.stringify({ source_dir: sourceDir, output_filename: 'merged_output.csv' })
+        });
+        const data = await res.json();
+        if (data.status === 'success') {
+            resultDiv.innerHTML = `<span style="color:var(--success)"><i class="fa-solid fa-check"></i> ${data.message}</span>`;
+        } else {
+            resultDiv.innerHTML = `<span style="color:var(--warning)">${data.message}</span>`;
+        }
+    } catch (e) {
+        resultDiv.innerHTML = `<span style="color:red">Error: ${e.message}</span>`;
+    }
+}
+
+window.triggerSort = async function () {
+    const targetFile = document.getElementById('sortTarget').value;
+    const resultDiv = document.getElementById('sortResult');
+
+    resultDiv.innerHTML = '<i class="fa-solid fa-spinner fa-spin"></i> Sorting...';
+
+    try {
+        const res = await fetch('/api/tools/sort', {
+            method: 'POST',
+            headers: { 'Content-Type': 'application/json' },
+            body: JSON.stringify({ target_file: targetFile })
+        });
+        const data = await res.json();
+        if (data.status === 'success') {
+            resultDiv.innerHTML = `<span style="color:var(--success)"><i class="fa-solid fa-check"></i> ${data.message}</span>`;
+        } else {
+            resultDiv.innerHTML = `<span style="color:var(--warning)">${data.message}</span>`;
+        }
+    } catch (e) {
+        resultDiv.innerHTML = `<span style="color:red">Error: ${e.message}</span>`;
+    }
+}
+
+// --- Help Functions ---
+window.fetchHelp = async function () {
+    const container = document.getElementById('helpContainer');
+    container.innerHTML = 'Loading...';
+
+    try {
+        const res = await fetch('/api/help/troubleshooting');
+        const data = await res.json();
+        container.innerHTML = marked.parse(data.content);
+    } catch (e) {
+        container.innerHTML = 'Failed to load guide.';
+    }
+}
diff --git a/web/static/style.css b/web/static/style.css
new file mode 100644
index 0000000..656bde4
--- /dev/null
+++ b/web/static/style.css
@@ -0,0 +1,387 @@
+:root {
+    --bg-color: #0f172a;
+    --sidebar-bg: rgba(30, 41, 59, 0.7);
+    --glass-border: 1px solid rgba(255, 255, 255, 0.1);
+    --primary-color: #6366f1;
+    --primary-hover: #4f46e5;
+    --text-main: #f8fafc;
+    --text-muted: #94a3b8;
+    --card-bg: rgba(30, 41, 59, 0.4);
+    --success: #10b981;
+    --warning: #f59e0b;
+}
+
+* {
+    box-sizing: border-box;
+    margin: 0;
+    padding: 0;
+}
+
+body {
+    font-family: 'Inter', sans-serif;
+    background-color: var(--bg-color);
+    background-image:
+        radial-gradient(at 0% 0%, rgba(99, 102, 241, 0.15) 0px, transparent 50%),
+        radial-gradient(at 100% 0%, rgba(16, 185, 129, 0.15) 0px, transparent 50%);
+    color: var(--text-main);
+    height: 100vh;
+    overflow: hidden;
+}
+
+.app-container {
+    display: flex;
+    height: 100vh;
+}
+
+/* Sidebar */
+.sidebar {
+    width: 320px;
+    background: var(--sidebar-bg);
+    backdrop-filter: blur(12px);
+    border-right: var(--glass-border);
+    padding: 1.5rem;
+    display: flex;
+    flex-direction: column;
+    gap: 2rem;
+}
+
+.brand {
+    display: flex;
+    align-items: center;
+    gap: 0.75rem;
+    color: var(--text-main);
+}
+
+.brand i {
+    font-size: 1.5rem;
+    color: var(--primary-color);
+}
+
+.brand h1 {
+    font-size: 1.25rem;
+    font-weight: 700;
+}
+
+.control-group h3 {
+    font-size: 0.875rem;
+    color: var(--text-muted);
+    text-transform: uppercase;
+    letter-spacing: 0.05em;
+    margin-bottom: 0.75rem;
+}
+
+/* Upload Zone */
+.upload-zone {
+    border: 2px dashed rgba(255, 255, 255, 0.1);
+    border-radius: 0.75rem;
+    padding: 2rem 1rem;
+    text-align: center;
+    transition: all 0.2s;
+    background: rgba(255, 255, 255, 0.02);
+}
+
+.upload-zone.dragover {
+    border-color: var(--primary-color);
+    background: rgba(99, 102, 241, 0.1);
+}
+
+.upload-zone i {
+    font-size: 2rem;
+    color: var(--text-muted);
+    margin-bottom: 0.5rem;
+}
+
+.upload-zone p {
+    font-size: 0.875rem;
+    color: var(--text-muted);
+    margin-bottom: 1rem;
+}
+
+/* Inputs */
+textarea {
+    width: 100%;
+    height: 120px;
+    background: rgba(0, 0, 0, 0.2);
+    border: var(--glass-border);
+    border-radius: 0.5rem;
+    padding: 0.75rem;
+    color: var(--text-main);
+    font-family: inherit;
+    resize: none;
+    font-size: 0.875rem;
+    margin-bottom: 1rem;
+}
+
+textarea:focus {
+    outline: none;
+    border-color: var(--primary-color);
+}
+
+/* Buttons */
+.btn {
+    width: 100%;
+    padding: 0.75rem;
+    border-radius: 0.5rem;
+    border: none;
+    cursor: pointer;
+    font-weight: 600;
+    transition: all 0.2s;
+    display: flex;
+    align-items: center;
+    justify-content: center;
+    gap: 0.5rem;
+}
+
+.btn.primary {
+    background: var(--primary-color);
+    color: white;
+}
+
+.btn.primary:hover {
+    background: var(--primary-hover);
+    transform: translateY(-1px);
+}
+
+.btn.secondary {
+    background: rgba(255, 255, 255, 0.1);
+    color: var(--text-main);
+    font-size: 0.8rem;
+    padding: 0.5rem;
+    width: auto;
+    margin: 0 auto;
+}
+
+.btn:disabled {
+    opacity: 0.5;
+    cursor: not-allowed;
+    transform: none;
+}
+
+/* Status */
+.status-indicator {
+    margin-top: auto;
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    font-size: 0.875rem;
+    padding-top: 1rem;
+    border-top: var(--glass-border);
+}
+
+.dot {
+    width: 8px;
+    height: 8px;
+    border-radius: 50%;
+    background-color: var(--text-muted);
+}
+
+.dot.running {
+    background-color: var(--warning);
+    box-shadow: 0 0 8px var(--warning);
+    animation: pulse 1s infinite;
+}
+
+.dot.done {
+    background-color: var(--success);
+}
+
+@keyframes pulse {
+    0% {
+        opacity: 1;
+    }
+
+    50% {
+        opacity: 0.5;
+    }
+
+    100% {
+        opacity: 1;
+    }
+}
+
+/* Main Content */
+.main-content {
+    flex: 1;
+    display: flex;
+    flex-direction: column;
+    padding: 1.5rem;
+    gap: 1rem;
+    overflow: hidden;
+    min-height: 0;
+    /* Crucial for nested flex scroll */
+}
+
+/* View Section needs to fill space for analysis view */
+.view-section {
+    flex: 1;
+    display: none;
+    overflow: hidden;
+    min-height: 0;
+}
+
+.view-section.active {
+    display: flex;
+    /* Changed from block to flex for analysis view layout */
+    flex-direction: column;
+}
+
+.tabs {
+    flex-shrink: 0;
+    /* Prevent tabs from shrinking */
+    display: flex;
+    gap: 1rem;
+    border-bottom: var(--glass-border);
+    padding-bottom: 0.5rem;
+}
+
+/* ... existing tab-btn styles ... */
+
+.tab-content {
+    display: none;
+    flex: 1;
+    overflow: hidden;
+    min-height: 0;
+    /* Crucial for nested flex scroll */
+    animation: fadeIn 0.3s ease;
+}
+
+.tab-content.active {
+    display: flex;
+    flex-direction: column;
+}
+
+/* Logs */
+.terminal-window {
+    background: #000;
+    border-radius: 0.75rem;
+    padding: 1rem;
+    flex: 1;
+    overflow-y: auto;
+    min-height: 0;
+    /* Crucial for nested flex scroll */
+    border: var(--glass-border);
+    font-family: 'JetBrains Mono', 'Menlo', monospace;
+    font-size: 0.85rem;
+    color: #4ade80;
+    line-height: 1.5;
+}
+
+#logOutput {
+    white-space: pre-wrap;
+}
+
+/* Report */
+#reportContainer {
+    flex: 1;
+    overflow-y: auto;
+    background: white;
+    color: #1e293b;
+    border-radius: 0.75rem;
+    padding: 2rem;
+}
+
+.empty-state {
+    height: 100%;
+    display: flex;
+    flex-direction: column;
+    align-items: center;
+    justify-content: center;
+    color: #94a3b8;
+    gap: 1rem;
+}
+
+.empty-state i {
+    font-size: 3rem;
+    opacity: 0.5;
+}
+
+/* Markdown Styles Override for Report */
+.markdown-body img {
+    max-width: 100%;
+    border-radius: 0.5rem;
+    box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.1);
+    margin: 1rem 0;
+}
+
+.markdown-body h1,
+.markdown-body h2,
+.markdown-body h3 {
+    margin-top: 1.5em;
+    margin-bottom: 0.5em;
+    color: #111827;
+}
+
+.markdown-body p {
+    margin-bottom: 1em;
+    line-height: 1.7;
+}
+
+/* New: Tools Grid */
+.tools-grid {
+    display: grid;
+    grid-template-columns: repeat(auto-fit, minmax(300px, 1fr));
+    gap: 1.5rem;
+    padding: 1rem 0;
+    overflow-y: auto;
+}
+
+.tool-card {
+    background: var(--card-bg);
+    border: var(--glass-border);
+    border-radius: 1rem;
+    padding: 1.5rem;
+    transition: transform 0.2s;
+}
+
+.tool-card:hover {
+    transform: translateY(-2px);
+    border-color: var(--primary-color);
+}
+
+.tool-icon {
+    font-size: 2rem;
+    color: var(--primary-color);
+    margin-bottom: 1rem;
+}
+
+.tool-card h3 {
+    margin-bottom: 0.5rem;
+}
+
+.tool-card p {
+    font-size: 0.9rem;
+    color: var(--text-muted);
+    margin-bottom: 1.5rem;
+    min-height: 3em;
+}
+
+.tool-actions {
+    display: flex;
+    gap: 0.5rem;
+}
+
+.input-sm {
+    background: rgba(0, 0, 0, 0.3);
+    border: var(--glass-border);
+    color: white;
+    padding: 0.5rem;
+    border-radius: 0.5rem;
+    flex: 1;
+}
+
+/* Help Styles */
+.help-header {
+    display: flex;
+    justify-content: space-between;
+    align-items: center;
+    margin-bottom: 1.5rem;
+}
+
+.help-body {
+    background: white;
+    color: #1e293b;
+    padding: 2rem;
+    border-radius: 0.75rem;
+    overflow-y: auto;
+    flex: 1;
+}
\ No newline at end of file