Phase 3: Memory & Growth - Multi-file fusion, Entity alignment with embedding, Document import, Knowledge base panel

Dockerfile

@@ -1,29 +1,33 @@
+# InsightFlow - Audio to Knowledge Graph Platform
+# Phase 3: Memory & Growth
+
 FROM python:3.11-slim
 
 WORKDIR /app
 
-# Install uv
+# Install system dependencies
-COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
-
-# Install system deps
 RUN apt-get update && apt-get install -y \
-    ffmpeg \
+    gcc \
-    git \
+    libpq-dev \
     && rm -rf /var/lib/apt/lists/*
 
-# Copy project files
+# Copy backend requirements
-COPY backend/pyproject.toml backend/uv.lock ./
+COPY backend/requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
 
-# Install dependencies using uv sync
+# Copy application code
-RUN uv sync --frozen --no-install-project
-
-# Copy code
 COPY backend/ ./backend/
 COPY frontend/ ./frontend/
 
-# Install project
+# Create data directory
-RUN uv sync --frozen
+RUN mkdir -p /app/data
 
+# Set environment variables
+ENV PYTHONPATH=/app
+ENV DB_PATH=/app/data/insightflow.db
+
+# Expose port
 EXPOSE 8000
 
-CMD ["uv", "run", "python", "backend/main.py"]
+# Run the application
+CMD ["python", "-m", "uvicorn", "backend.main:app", "--host", "0.0.0.0", "--port", "8000"]

README.md

@@ -1,27 +1,88 @@
-# InsightFlow
+# InsightFlow - Audio to Knowledge Graph Platform
 
-音频与文档的领域知识构建平台
+## Phase 3: Memory & Growth - Completed ✅
 
-## 产品定位
+### 新增功能
-将会议录音和文档转化为结构化的知识图谱，通过人机回圈(Human-in-the-Loop)实现知识持续生长。
 
-## 核心特性
+#### 1. 多文件图谱融合 ✅
-- 🎙️ ASR 语音识别 + 热词注入
+- 支持上传多个音频文件到同一项目
-- 🧠 LLM 实体抽取与解释
+- 系统自动对齐实体，合并图谱
-- 🔗 双视图联动（文档视图 + 图谱视图）
+- 实体提及跨文件追踪
-- 📈 知识生长（多文件实体对齐）
+- 文件选择器切换不同转录内容
 
-## 技术栈
+#### 2. 实体对齐算法优化 ✅
-- 前端: Next.js + Tailwind
+- 新增 `entity_aligner.py` 模块
-- 后端: Node.js / Python
+- 支持使用 Kimi API embedding 进行语义相似度匹配
-- 数据库: MySQL + Neo4j
+- 余弦相似度计算
-- ASR: Whisper
+- 自动别名建议
-- LLM: OpenAI / Kimi
+- 批量实体对齐 API
 
-## 开发阶段
+#### 3. PDF/DOCX 文档导入 ✅
-- [ ] Phase 1: 骨架与单体分析 (MVP)
+- 新增 `document_processor.py` 模块
-- [ ] Phase 2: 交互与纠错工作台
+- 支持 PDF、DOCX、TXT、MD 格式
-- [ ] Phase 3: 记忆与生长
+- 文档文本提取并参与实体提取
+- 文档类型标记（音频/文档）
 
-## 文档
+#### 4. 项目知识库面板 ✅
-- [PRD v2.0](docs/PRD-v2.0.md)
+- 全新的知识库视图
+- 统计面板：实体数、关系数、文件数、术语数
+- 实体网格展示（带提及统计）
+- 关系列表展示
+- 术语表管理（添加/删除）
+- 文件列表展示
+
+### 技术栈
+- 后端: FastAPI + SQLite
+- 前端: 原生 HTML/JS + D3.js
+- ASR: 阿里云听悟
+- LLM: Kimi API
+- 文档处理: PyPDF2, python-docx
+
+### 部署
+
+```bash
+# 构建 Docker 镜像
+docker build -t insightflow:phase3 .
+
+# 运行容器
+docker run -d \
+  -p 18000:8000 \
+  -v /opt/data:/app/data \
+  -e KIMI_API_KEY=your_key \
+  -e ALIYUN_ACCESS_KEY_ID=your_key \
+  -e ALIYUN_ACCESS_KEY_SECRET=your_secret \
+  insightflow:phase3
+```
+
+### API 文档
+
+#### 新增 API
+
+**文档上传**
+```
+POST /api/v1/projects/{project_id}/upload-document
+Content-Type: multipart/form-data
+file: <文件>
+```
+
+**知识库查询**
+```
+GET /api/v1/projects/{project_id}/knowledge-base
+```
+
+**术语表管理**
+```
+POST /api/v1/projects/{project_id}/glossary
+GET /api/v1/projects/{project_id}/glossary
+DELETE /api/v1/glossary/{term_id}
+```
+
+**实体对齐**
+```
+POST /api/v1/projects/{project_id}/align-entities?threshold=0.85
+```
+
+### 数据库 Schema 更新
+- `transcripts` 表新增 `type` 字段（audio/document）
+- `entities` 表新增 `embedding` 字段
+- 新增索引优化查询性能

STATUS.md

@@ -4,7 +4,7 @@
 
 ## 当前阶段
 
-Phase 2: 交互与纠错工作台 - **已完成 ✅**
+Phase 3: 记忆与生长 - **已完成 ✅**
 
 ## 已完成
 
@@ -64,22 +64,70 @@ Phase 2: 交互与纠错工作台 - **已完成 ✅**
 - ✅ update_relation() - 更新关系
 - ✅ update_transcript() - 更新转录文本
 
-## Phase 3 计划 (记忆与生长) - **即将开始**
+### Phase 3: 记忆与生长 ✅
 
-- 多文件图谱融合
+#### 多文件图谱融合
-- 实体对齐算法优化
+- ✅ 支持上传多个音频文件到同一项目
-- PDF/DOCX 文档导入
+- ✅ 系统自动对齐实体，合并图谱
-- 项目知识库面板
+- ✅ 实体提及跨文件追踪
+- ✅ 文件选择器切换不同转录内容
+- ✅ 转录列表 API 返回文件类型
+
+#### 实体对齐算法优化
+- ✅ 新增 `entity_aligner.py` 模块
+- ✅ 使用 Kimi API embedding 进行语义相似度匹配
+- ✅ 余弦相似度计算
+- ✅ 自动别名建议
+- ✅ 批量实体对齐 API
+- ✅ 实体对齐回退机制（字符串匹配）
+
+#### PDF/DOCX 文档导入
+- ✅ 新增 `document_processor.py` 模块
+- ✅ 支持 PDF、DOCX、TXT、MD 格式
+- ✅ 文档文本提取并参与实体提取
+- ✅ 文档上传 API (/api/v1/projects/{id}/upload-document)
+- ✅ 文档类型标记（audio/document）
+
+#### 项目知识库面板
+- ✅ 全新的知识库视图
+- ✅ 侧边栏导航切换（工作台/知识库）
+- ✅ 统计面板：实体数、关系数、文件数、术语数
+- ✅ 实体网格展示（带提及统计）
+- ✅ 关系列表展示
+- ✅ 术语表管理（添加/删除）
+- ✅ 文件列表展示（区分音频/文档）
+
+#### 术语表功能
+- ✅ 术语表数据库表 (glossary)
+- ✅ 添加术语 API
+- ✅ 获取术语列表 API
+- ✅ 删除术语 API
+- ✅ 前端术语表管理界面
+
+#### 数据库更新
+- ✅ transcripts 表新增 `type` 字段
+- ✅ entities 表新增 `embedding` 字段
+- ✅ 新增 glossary 表
+- ✅ 新增索引优化查询性能
 
 ## 技术债务
 
 - 听悟 SDK fallback 到 mock 需要更好的错误处理
-- 实体相似度匹配目前只是简单字符串包含，需要 embedding 方案
 - 前端需要状态管理（目前使用全局变量）
 - 需要添加 API 文档 (OpenAPI/Swagger)
+- Embedding 缓存需要持久化
+- 实体对齐算法需要更多测试
 
 ## 部署信息
 
 - 服务器: 122.51.127.111
 - 项目路径: /opt/projects/insightflow
 - 端口: 18000
+- Docker 镜像: insightflow:phase3
+
+## 下一步 (Phase 4)
+
+- 知识推理与问答
+- 实体属性扩展
+- 时间线视图
+- 导出功能（PDF/图片）

backend/db_manager.py

@@ -1,7 +1,8 @@
 #!/usr/bin/env python3
 """
-InsightFlow Database Manager
+InsightFlow Database Manager - Phase 3
 处理项目、实体、关系的持久化
+支持文档类型和多文件融合
 """
 
 import os
@@ -166,6 +167,18 @@ class DatabaseManager:
             (target_id, source_id)
         )
         
+        # 更新关系 - source 作为 source_entity_id
+        conn.execute(
+            "UPDATE entity_relations SET source_entity_id = ? WHERE source_entity_id = ?",
+            (target_id, source_id)
+        )
+        
+        # 更新关系 - source 作为 target_entity_id
+        conn.execute(
+            "UPDATE entity_relations SET target_entity_id = ? WHERE target_entity_id = ?",
+            (target_id, source_id)
+        )
+        
         # 删除源实体
         conn.execute("DELETE FROM entities WHERE id = ?", (source_id,))
         
@@ -222,13 +235,13 @@ class DatabaseManager:
         return [EntityMention(**dict(r)) for r in rows]
 
     # Transcript operations
-    def save_transcript(self, transcript_id: str, project_id: str, filename: str, full_text: str):
+    def save_transcript(self, transcript_id: str, project_id: str, filename: str, full_text: str, transcript_type: str = "audio"):
         """保存转录记录"""
         conn = self.get_conn()
         now = datetime.now().isoformat()
         conn.execute(
-            "INSERT INTO transcripts (id, project_id, filename, full_text, created_at) VALUES (?, ?, ?, ?, ?)",
+            "INSERT INTO transcripts (id, project_id, filename, full_text, type, created_at) VALUES (?, ?, ?, ?, ?, ?)",
-            (transcript_id, project_id, filename, full_text, now)
+            (transcript_id, project_id, filename, full_text, transcript_type, now)
         )
         conn.commit()
         conn.close()
@@ -388,6 +401,58 @@ class DatabaseManager:
         conn.close()
         
         return dict(row) if row else None
+    
+    # Phase 3: Glossary operations
+    def add_glossary_term(self, project_id: str, term: str, pronunciation: str = "") -> str:
+        """添加术语到术语表"""
+        conn = self.get_conn()
+        
+        # 检查是否已存在
+        existing = conn.execute(
+            "SELECT * FROM glossary WHERE project_id = ? AND term = ?",
+            (project_id, term)
+        ).fetchone()
+        
+        if existing:
+            # 更新频率
+            conn.execute(
+                "UPDATE glossary SET frequency = frequency + 1 WHERE id = ?",
+                (existing['id'],)
+            )
+            conn.commit()
+            conn.close()
+            return existing['id']
+        
+        term_id = str(uuid.uuid4())[:8]
+        conn.execute(
+            "INSERT INTO glossary (id, project_id, term, pronunciation, frequency) VALUES (?, ?, ?, ?, ?)",
+            (term_id, project_id, term, pronunciation, 1)
+        )
+        conn.commit()
+        conn.close()
+        return term_id
+    
+    def list_glossary(self, project_id: str) -> List[dict]:
+        """列出项目术语表"""
+        conn = self.get_conn()
+        rows = conn.execute(
+            "SELECT * FROM glossary WHERE project_id = ? ORDER BY frequency DESC",
+            (project_id,)
+        ).fetchall()
+        conn.close()
+        return [dict(r) for r in rows]
+    
+    def delete_glossary_term(self, term_id: str):
+        """删除术语"""
+        conn = self.get_conn()
+        conn.execute("DELETE FROM glossary WHERE id = ?", (term_id,))
+        conn.commit()
+        conn.close()
+    
+    # Phase 3: Get all entities for embedding
+    def get_all_entities_for_embedding(self, project_id: str) -> List[Entity]:
+        """获取所有实体用于 embedding 计算"""
+        return self.list_project_entities(project_id)
 
 
 # Singleton instance

backend/document_processor.py

@@ -0,0 +1,180 @@
+#!/usr/bin/env python3
+"""
+Document Processor - Phase 3
+支持 PDF 和 DOCX 文档导入
+"""
+
+import os
+import io
+from typing import Dict, Optional
+
+class DocumentProcessor:
+    """文档处理器 - 提取 PDF/DOCX 文本"""
+    
+    def __init__(self):
+        self.supported_formats = {
+            '.pdf': self._extract_pdf,
+            '.docx': self._extract_docx,
+            '.doc': self._extract_docx,
+            '.txt': self._extract_txt,
+            '.md': self._extract_txt,
+        }
+    
+    def process(self, content: bytes, filename: str) -> Dict[str, str]:
+        """
+        处理文档并提取文本
+        
+        Args:
+            content: 文件二进制内容
+            filename: 文件名
+            
+        Returns:
+            {"text": "提取的文本内容", "format": "文件格式"}
+        """
+        ext = os.path.splitext(filename.lower())[1]
+        
+        if ext not in self.supported_formats:
+            raise ValueError(f"Unsupported file format: {ext}. Supported: {list(self.supported_formats.keys())}")
+        
+        extractor = self.supported_formats[ext]
+        text = extractor(content)
+        
+        # 清理文本
+        text = self._clean_text(text)
+        
+        return {
+            "text": text,
+            "format": ext,
+            "filename": filename
+        }
+    
+    def _extract_pdf(self, content: bytes) -> str:
+        """提取 PDF 文本"""
+        try:
+            import PyPDF2
+            pdf_file = io.BytesIO(content)
+            reader = PyPDF2.PdfReader(pdf_file)
+            
+            text_parts = []
+            for page in reader.pages:
+                page_text = page.extract_text()
+                if page_text:
+                    text_parts.append(page_text)
+            
+            return "\n\n".join(text_parts)
+        except ImportError:
+            # Fallback: 尝试使用 pdfplumber
+            try:
+                import pdfplumber
+                text_parts = []
+                with pdfplumber.open(io.BytesIO(content)) as pdf:
+                    for page in pdf.pages:
+                        page_text = page.extract_text()
+                        if page_text:
+                            text_parts.append(page_text)
+                return "\n\n".join(text_parts)
+            except ImportError:
+                raise ImportError("PDF processing requires PyPDF2 or pdfplumber. Install with: pip install PyPDF2")
+        except Exception as e:
+            raise ValueError(f"PDF extraction failed: {str(e)}")
+    
+    def _extract_docx(self, content: bytes) -> str:
+        """提取 DOCX 文本"""
+        try:
+            import docx
+            doc_file = io.BytesIO(content)
+            doc = docx.Document(doc_file)
+            
+            text_parts = []
+            for para in doc.paragraphs:
+                if para.text.strip():
+                    text_parts.append(para.text)
+            
+            # 提取表格中的文本
+            for table in doc.tables:
+                for row in table.rows:
+                    row_text = []
+                    for cell in row.cells:
+                        if cell.text.strip():
+                            row_text.append(cell.text.strip())
+                    if row_text:
+                        text_parts.append(" | ".join(row_text))
+            
+            return "\n\n".join(text_parts)
+        except ImportError:
+            raise ImportError("DOCX processing requires python-docx. Install with: pip install python-docx")
+        except Exception as e:
+            raise ValueError(f"DOCX extraction failed: {str(e)}")
+    
+    def _extract_txt(self, content: bytes) -> str:
+        """提取纯文本"""
+        # 尝试多种编码
+        encodings = ['utf-8', 'gbk', 'gb2312', 'latin-1']
+        
+        for encoding in encodings:
+            try:
+                return content.decode(encoding)
+            except UnicodeDecodeError:
+                continue
+        
+        # 如果都失败了，使用 latin-1 并忽略错误
+        return content.decode('latin-1', errors='ignore')
+    
+    def _clean_text(self, text: str) -> str:
+        """清理提取的文本"""
+        if not text:
+            return ""
+        
+        # 移除多余的空白字符
+        lines = text.split('\n')
+        cleaned_lines = []
+        
+        for line in lines:
+            line = line.strip()
+            # 移除空行，但保留段落分隔
+            if line:
+                cleaned_lines.append(line)
+        
+        # 合并行，保留段落结构
+        text = '\n\n'.join(cleaned_lines)
+        
+        # 移除多余的空格
+        text = ' '.join(text.split())
+        
+        # 移除控制字符
+        text = ''.join(char for char in text if ord(char) >= 32 or char in '\n\r\t')
+        
+        return text.strip()
+    
+    def is_supported(self, filename: str) -> bool:
+        """检查文件格式是否支持"""
+        ext = os.path.splitext(filename.lower())[1]
+        return ext in self.supported_formats
+
+
+# 简单的文本提取器（不需要外部依赖）
+class SimpleTextExtractor:
+    """简单的文本提取器，用于测试"""
+    
+    def extract(self, content: bytes, filename: str) -> str:
+        """尝试提取文本"""
+        encodings = ['utf-8', 'gbk', 'latin-1']
+        
+        for encoding in encodings:
+            try:
+                return content.decode(encoding)
+            except UnicodeDecodeError:
+                continue
+        
+        return content.decode('latin-1', errors='ignore')
+
+
+if __name__ == "__main__":
+    # 测试
+    processor = DocumentProcessor()
+    
+    # 测试文本提取
+    test_text = "Hello World\n\nThis is a test document.\n\nMultiple paragraphs."
+    result = processor.process(test_text.encode('utf-8'), "test.txt")
+    print(f"Text extraction test: {len(result['text'])} chars")
+    print(result['text'][:100])

backend/entity_aligner.py

@@ -0,0 +1,372 @@
+#!/usr/bin/env python3
+"""
+Entity Aligner - Phase 3
+使用 embedding 进行实体对齐
+"""
+
+import os
+import json
+import httpx
+import numpy as np
+from typing import List, Optional, Dict
+from dataclasses import dataclass
+
+# API Keys
+KIMI_API_KEY = os.getenv("KIMI_API_KEY", "")
+KIMI_BASE_URL = os.getenv("KIMI_BASE_URL", "https://api.kimi.com/coding")
+
+@dataclass
+class EntityEmbedding:
+    entity_id: str
+    name: str
+    definition: str
+    embedding: List[float]
+
+class EntityAligner:
+    """实体对齐器 - 使用 embedding 进行相似度匹配"""
+    
+    def __init__(self, similarity_threshold: float = 0.85):
+        self.similarity_threshold = similarity_threshold
+        self.embedding_cache: Dict[str, List[float]] = {}
+        
+    def get_embedding(self, text: str) -> Optional[List[float]]:
+        """
+        使用 Kimi API 获取文本的 embedding
+        
+        Args:
+            text: 输入文本
+            
+        Returns:
+            embedding 向量或 None
+        """
+        if not KIMI_API_KEY:
+            return None
+        
+        # 检查缓存
+        cache_key = hash(text)
+        if cache_key in self.embedding_cache:
+            return self.embedding_cache[cache_key]
+        
+        try:
+            response = httpx.post(
+                f"{KIMI_BASE_URL}/v1/embeddings",
+                headers={"Authorization": f"Bearer {KIMI_API_KEY}", "Content-Type": "application/json"},
+                json={
+                    "model": "k2p5",
+                    "input": text[:500]  # 限制长度
+                },
+                timeout=30.0
+            )
+            response.raise_for_status()
+            result = response.json()
+            
+            embedding = result["data"][0]["embedding"]
+            self.embedding_cache[cache_key] = embedding
+            return embedding
+            
+        except Exception as e:
+            print(f"Embedding API failed: {e}")
+            return None
+    
+    def compute_similarity(self, embedding1: List[float], embedding2: List[float]) -> float:
+        """
+        计算两个 embedding 的余弦相似度
+        
+        Args:
+            embedding1: 第一个向量
+            embedding2: 第二个向量
+            
+        Returns:
+            相似度分数 (0-1)
+        """
+        vec1 = np.array(embedding1)
+        vec2 = np.array(embedding2)
+        
+        # 余弦相似度
+        dot_product = np.dot(vec1, vec2)
+        norm1 = np.linalg.norm(vec1)
+        norm2 = np.linalg.norm(vec2)
+        
+        if norm1 == 0 or norm2 == 0:
+            return 0.0
+        
+        return float(dot_product / (norm1 * norm2))
+    
+    def get_entity_text(self, name: str, definition: str = "") -> str:
+        """
+        构建用于 embedding 的实体文本
+        
+        Args:
+            name: 实体名称
+            definition: 实体定义
+            
+        Returns:
+            组合文本
+        """
+        if definition:
+            return f"{name}: {definition}"
+        return name
+    
+    def find_similar_entity(
+        self, 
+        project_id: str, 
+        name: str, 
+        definition: str = "",
+        exclude_id: Optional[str] = None,
+        threshold: Optional[float] = None
+    ) -> Optional[object]:
+        """
+        查找相似的实体
+        
+        Args:
+            project_id: 项目 ID
+            name: 实体名称
+            definition: 实体定义
+            exclude_id: 要排除的实体 ID
+            threshold: 相似度阈值
+            
+        Returns:
+            相似的实体或 None
+        """
+        if threshold is None:
+            threshold = self.similarity_threshold
+        
+        try:
+            from db_manager import get_db_manager
+            db = get_db_manager()
+        except ImportError:
+            return None
+        
+        # 获取项目的所有实体
+        entities = db.get_all_entities_for_embedding(project_id)
+        
+        if not entities:
+            return None
+        
+        # 获取查询实体的 embedding
+        query_text = self.get_entity_text(name, definition)
+        query_embedding = self.get_embedding(query_text)
+        
+        if query_embedding is None:
+            # 如果 embedding API 失败，回退到简单匹配
+            return self._fallback_similarity_match(entities, name, exclude_id)
+        
+        best_match = None
+        best_score = threshold
+        
+        for entity in entities:
+            if exclude_id and entity.id == exclude_id:
+                continue
+            
+            # 获取实体的 embedding
+            entity_text = self.get_entity_text(entity.name, entity.definition)
+            entity_embedding = self.get_embedding(entity_text)
+            
+            if entity_embedding is None:
+                continue
+            
+            # 计算相似度
+            similarity = self.compute_similarity(query_embedding, entity_embedding)
+            
+            if similarity > best_score:
+                best_score = similarity
+                best_match = entity
+        
+        return best_match
+    
+    def _fallback_similarity_match(
+        self, 
+        entities: List[object], 
+        name: str, 
+        exclude_id: Optional[str] = None
+    ) -> Optional[object]:
+        """
+        回退到简单的相似度匹配（不使用 embedding）
+        
+        Args:
+            entities: 实体列表
+            name: 查询名称
+            exclude_id: 要排除的实体 ID
+            
+        Returns:
+            最相似的实体或 None
+        """
+        name_lower = name.lower()
+        
+        # 1. 精确匹配
+        for entity in entities:
+            if exclude_id and entity.id == exclude_id:
+                continue
+            if entity.name.lower() == name_lower:
+                return entity
+            if entity.aliases and name_lower in [a.lower() for a in entity.aliases]:
+                return entity
+        
+        # 2. 包含匹配
+        for entity in entities:
+            if exclude_id and entity.id == exclude_id:
+                continue
+            if name_lower in entity.name.lower() or entity.name.lower() in name_lower:
+                return entity
+        
+        return None
+    
+    def batch_align_entities(
+        self, 
+        project_id: str, 
+        new_entities: List[Dict],
+        threshold: Optional[float] = None
+    ) -> List[Dict]:
+        """
+        批量对齐实体
+        
+        Args:
+            project_id: 项目 ID
+            new_entities: 新实体列表 [{"name": "...", "definition": "..."}]
+            threshold: 相似度阈值
+            
+        Returns:
+            对齐结果列表 [{"new_entity": {...}, "matched_entity": {...}, "similarity": 0.9}]
+        """
+        if threshold is None:
+            threshold = self.similarity_threshold
+        
+        results = []
+        
+        for new_ent in new_entities:
+            matched = self.find_similar_entity(
+                project_id,
+                new_ent["name"],
+                new_ent.get("definition", ""),
+                threshold=threshold
+            )
+            
+            result = {
+                "new_entity": new_ent,
+                "matched_entity": None,
+                "similarity": 0.0,
+                "should_merge": False
+            }
+            
+            if matched:
+                # 计算相似度
+                query_text = self.get_entity_text(new_ent["name"], new_ent.get("definition", ""))
+                matched_text = self.get_entity_text(matched.name, matched.definition)
+                
+                query_emb = self.get_embedding(query_text)
+                matched_emb = self.get_embedding(matched_text)
+                
+                if query_emb and matched_emb:
+                    similarity = self.compute_similarity(query_emb, matched_emb)
+                    result["matched_entity"] = {
+                        "id": matched.id,
+                        "name": matched.name,
+                        "type": matched.type,
+                        "definition": matched.definition
+                    }
+                    result["similarity"] = similarity
+                    result["should_merge"] = similarity >= threshold
+            
+            results.append(result)
+        
+        return results
+    
+    def suggest_entity_aliases(self, entity_name: str, entity_definition: str = "") -> List[str]:
+        """
+        使用 LLM 建议实体的别名
+        
+        Args:
+            entity_name: 实体名称
+            entity_definition: 实体定义
+            
+        Returns:
+            建议的别名列表
+        """
+        if not KIMI_API_KEY:
+            return []
+        
+        prompt = f"""为以下实体生成可能的别名或简称：
+
+实体名称：{entity_name}
+定义：{entity_definition}
+
+请返回 JSON 格式的别名列表：
+{{"aliases": ["别名1", "别名2", "别名3"]}}
+
+只返回 JSON，不要其他内容。"""
+        
+        try:
+            response = httpx.post(
+                f"{KIMI_BASE_URL}/v1/chat/completions",
+                headers={"Authorization": f"Bearer {KIMI_API_KEY}", "Content-Type": "application/json"},
+                json={
+                    "model": "k2p5",
+                    "messages": [{"role": "user", "content": prompt}],
+                    "temperature": 0.3
+                },
+                timeout=30.0
+            )
+            response.raise_for_status()
+            result = response.json()
+            content = result["choices"][0]["message"]["content"]
+            
+            import re
+            json_match = re.search(r'\{{.*?\}}', content, re.DOTALL)
+            if json_match:
+                data = json.loads(json_match.group())
+                return data.get("aliases", [])
+        except Exception as e:
+            print(f"Alias suggestion failed: {e}")
+        
+        return []
+
+
+# 简单的字符串相似度计算（不使用 embedding）
+def simple_similarity(str1: str, str2: str) -> float:
+    """
+    计算两个字符串的简单相似度
+    
+    Args:
+        str1: 第一个字符串
+        str2: 第二个字符串
+        
+    Returns:
+        相似度分数 (0-1)
+    """
+    if str1 == str2:
+        return 1.0
+    
+    if not str1 or not str2:
+        return 0.0
+    
+    # 转换为小写
+    s1 = str1.lower()
+    s2 = str2.lower()
+    
+    # 包含关系
+    if s1 in s2 or s2 in s1:
+        return 0.8
+    
+    # 计算编辑距离相似度
+    from difflib import SequenceMatcher
+    return SequenceMatcher(None, s1, s2).ratio()
+
+
+if __name__ == "__main__":
+    # 测试
+    aligner = EntityAligner()
+    
+    # 测试 embedding
+    test_text = "Kubernetes 容器编排平台"
+    embedding = aligner.get_embedding(test_text)
+    if embedding:
+        print(f"Embedding dimension: {len(embedding)}")
+        print(f"First 5 values: {embedding[:5]}")
+    else:
+        print("Embedding API not available")
+    
+    # 测试相似度计算
+    emb1 = [1.0, 0.0, 0.0]
+    emb2 = [0.9, 0.1, 0.0]
+    sim = aligner.compute_similarity(emb1, emb2)
+    print(f"Similarity: {sim:.4f}")

backend/main.py

@@ -1,7 +1,7 @@
 #!/usr/bin/env python3
 """
-InsightFlow Backend - Phase 3 (Production Ready)
+InsightFlow Backend - Phase 3 (Memory & Growth)
-Knowledge Growth: Multi-file fusion + Entity Alignment
+Knowledge Growth: Multi-file fusion + Entity Alignment + Document Import
 ASR: 阿里云听悟 + OSS
 """
 
@@ -9,6 +9,7 @@ import os
 import json
 import httpx
 import uuid
+import re
 from fastapi import FastAPI, File, UploadFile, HTTPException, Form
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.staticfiles import StaticFiles
@@ -35,6 +36,18 @@ try:
 except ImportError:
     DB_AVAILABLE = False
 
+try:
+    from document_processor import DocumentProcessor
+    DOC_PROCESSOR_AVAILABLE = True
+except ImportError:
+    DOC_PROCESSOR_AVAILABLE = False
+
+try:
+    from entity_aligner import EntityAligner
+    ALIGNER_AVAILABLE = True
+except ImportError:
+    ALIGNER_AVAILABLE = False
+
 app = FastAPI(title="InsightFlow", version="0.3.0")
 
 app.add_middleware(
@@ -90,9 +103,29 @@ class EntityMergeRequest(BaseModel):
     source_entity_id: str
     target_entity_id: str
 
+class GlossaryTermCreate(BaseModel):
+    term: str
+    pronunciation: Optional[str] = ""
+
 # API Keys
 KIMI_API_KEY = os.getenv("KIMI_API_KEY", "")
-KIMI_BASE_URL = "https://api.kimi.com/coding"
+KIMI_BASE_URL = os.getenv("KIMI_BASE_URL", "https://api.kimi.com/coding")
+
+# Phase 3: Entity Aligner singleton
+_aligner = None
+def get_aligner():
+    global _aligner
+    if _aligner is None and ALIGNER_AVAILABLE:
+        _aligner = EntityAligner()
+    return _aligner
+
+# Phase 3: Document Processor singleton
+_doc_processor = None
+def get_doc_processor():
+    global _doc_processor
+    if _doc_processor is None and DOC_PROCESSOR_AVAILABLE:
+        _doc_processor = DocumentProcessor()
+    return _doc_processor
 
 # Phase 2: Entity Edit API
 @app.put("/api/v1/entities/{entity_id}")
@@ -406,12 +439,21 @@ def extract_entities_with_llm(text: str) -> tuple[List[dict], List[dict]]:
     
     return [], []
 
-def align_entity(project_id: str, name: str, db) -> Optional[Entity]:
+def align_entity(project_id: str, name: str, db, definition: str = "") -> Optional[Entity]:
-    """实体对齐"""
+    """实体对齐 - Phase 3: 使用 embedding 对齐"""
+    # 1. 首先尝试精确匹配
     existing = db.get_entity_by_name(project_id, name)
     if existing:
         return existing
     
+    # 2. 使用 embedding 对齐（如果可用）
+    aligner = get_aligner()
+    if aligner:
+        similar = aligner.find_similar_entity(project_id, name, definition)
+        if similar:
+            return similar
+    
+    # 3. 回退到简单相似度匹配
     similar = db.find_similar_entities(project_id, name)
     if similar:
         return similar[0]
@@ -443,7 +485,7 @@ async def list_projects():
 
 @app.post("/api/v1/projects/{project_id}/upload", response_model=AnalysisResult)
 async def upload_audio(project_id: str, file: UploadFile = File(...)):
-    """上传音频到指定项目"""
+    """上传音频到指定项目 - Phase 3: 支持多文件融合"""
     if not DB_AVAILABLE:
         raise HTTPException(status_code=500, detail="Database not available")
     
@@ -471,12 +513,12 @@ async def upload_audio(project_id: str, file: UploadFile = File(...)):
         full_text=tw_result["full_text"]
     )
     
-    # 实体对齐并保存
+    # 实体对齐并保存 - Phase 3: 使用增强对齐
     aligned_entities = []
     entity_name_to_id = {}  # 用于关系映射
     
     for raw_ent in raw_entities:
-        existing = align_entity(project_id, raw_ent["name"], db)
+        existing = align_entity(project_id, raw_ent["name"], db, raw_ent.get("definition", ""))
         
         if existing:
             ent_model = EntityModel(
@@ -551,6 +593,302 @@ async def upload_audio(project_id: str, file: UploadFile = File(...)):
         created_at=datetime.now().isoformat()
     )
 
+# Phase 3: Document Upload API
+@app.post("/api/v1/projects/{project_id}/upload-document")
+async def upload_document(project_id: str, file: UploadFile = File(...)):
+    """上传 PDF/DOCX 文档到指定项目"""
+    if not DB_AVAILABLE:
+        raise HTTPException(status_code=500, detail="Database not available")
+    
+    if not DOC_PROCESSOR_AVAILABLE:
+        raise HTTPException(status_code=500, detail="Document processor not available")
+    
+    db = get_db_manager()
+    project = db.get_project(project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="Project not found")
+    
+    content = await file.read()
+    
+    # 处理文档
+    processor = get_doc_processor()
+    try:
+        result = processor.process(content, file.filename)
+    except Exception as e:
+        raise HTTPException(status_code=400, detail=f"Document processing failed: {str(e)}")
+    
+    # 保存文档转录记录
+    transcript_id = str(uuid.uuid4())[:8]
+    db.save_transcript(
+        transcript_id=transcript_id,
+        project_id=project_id,
+        filename=file.filename,
+        full_text=result["text"],
+        transcript_type="document"
+    )
+    
+    # 提取实体和关系
+    raw_entities, raw_relations = extract_entities_with_llm(result["text"])
+    
+    # 实体对齐并保存
+    aligned_entities = []
+    entity_name_to_id = {}
+    
+    for raw_ent in raw_entities:
+        existing = align_entity(project_id, raw_ent["name"], db, raw_ent.get("definition", ""))
+        
+        if existing:
+            entity_name_to_id[raw_ent["name"]] = existing.id
+            aligned_entities.append(EntityModel(
+                id=existing.id,
+                name=existing.name,
+                type=existing.type,
+                definition=existing.definition,
+                aliases=existing.aliases
+            ))
+        else:
+            new_ent = db.create_entity(Entity(
+                id=str(uuid.uuid4())[:8],
+                project_id=project_id,
+                name=raw_ent["name"],
+                type=raw_ent.get("type", "OTHER"),
+                definition=raw_ent.get("definition", "")
+            ))
+            entity_name_to_id[raw_ent["name"]] = new_ent.id
+            aligned_entities.append(EntityModel(
+                id=new_ent.id,
+                name=new_ent.name,
+                type=new_ent.type,
+                definition=new_ent.definition
+            ))
+        
+        # 保存实体提及位置
+        full_text = result["text"]
+        name = raw_ent["name"]
+        start_pos = 0
+        while True:
+            pos = full_text.find(name, start_pos)
+            if pos == -1:
+                break
+            mention = EntityMention(
+                id=str(uuid.uuid4())[:8],
+                entity_id=entity_name_to_id[name],
+                transcript_id=transcript_id,
+                start_pos=pos,
+                end_pos=pos + len(name),
+                text_snippet=full_text[max(0, pos-20):min(len(full_text), pos+len(name)+20)],
+                confidence=1.0
+            )
+            db.add_mention(mention)
+            start_pos = pos + 1
+    
+    # 保存关系
+    for rel in raw_relations:
+        source_id = entity_name_to_id.get(rel.get("source", ""))
+        target_id = entity_name_to_id.get(rel.get("target", ""))
+        if source_id and target_id:
+            db.create_relation(
+                project_id=project_id,
+                source_entity_id=source_id,
+                target_entity_id=target_id,
+                relation_type=rel.get("type", "related"),
+                evidence=result["text"][:200],
+                transcript_id=transcript_id
+            )
+    
+    return {
+        "transcript_id": transcript_id,
+        "project_id": project_id,
+        "filename": file.filename,
+        "text_length": len(result["text"]),
+        "entities": [e.dict() for e in aligned_entities],
+        "created_at": datetime.now().isoformat()
+    }
+
+# Phase 3: Knowledge Base API
+@app.get("/api/v1/projects/{project_id}/knowledge-base")
+async def get_knowledge_base(project_id: str):
+    """获取项目知识库 - 包含所有实体、关系、术语表"""
+    if not DB_AVAILABLE:
+        raise HTTPException(status_code=500, detail="Database not available")
+    
+    db = get_db_manager()
+    project = db.get_project(project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="Project not found")
+    
+    # 获取所有实体
+    entities = db.list_project_entities(project_id)
+    
+    # 获取所有关系
+    relations = db.list_project_relations(project_id)
+    
+    # 获取所有转录
+    transcripts = db.list_project_transcripts(project_id)
+    
+    # 获取术语表
+    glossary = db.list_glossary(project_id)
+    
+    # 构建实体统计
+    entity_stats = {}
+    for ent in entities:
+        mentions = db.get_entity_mentions(ent.id)
+        entity_stats[ent.id] = {
+            "mention_count": len(mentions),
+            "transcript_ids": list(set([m.transcript_id for m in mentions]))
+        }
+    
+    # 构建实体名称映射
+    entity_map = {e.id: e.name for e in entities}
+    
+    return {
+        "project": {
+            "id": project.id,
+            "name": project.name,
+            "description": project.description
+        },
+        "stats": {
+            "entity_count": len(entities),
+            "relation_count": len(relations),
+            "transcript_count": len(transcripts),
+            "glossary_count": len(glossary)
+        },
+        "entities": [
+            {
+                "id": e.id,
+                "name": e.name,
+                "type": e.type,
+                "definition": e.definition,
+                "aliases": e.aliases,
+                "mention_count": entity_stats.get(e.id, {}).get("mention_count", 0),
+                "appears_in": entity_stats.get(e.id, {}).get("transcript_ids", [])
+            }
+            for e in entities
+        ],
+        "relations": [
+            {
+                "id": r["id"],
+                "source_id": r["source_entity_id"],
+                "source_name": entity_map.get(r["source_entity_id"], "Unknown"),
+                "target_id": r["target_entity_id"],
+                "target_name": entity_map.get(r["target_entity_id"], "Unknown"),
+                "type": r["relation_type"],
+                "evidence": r["evidence"]
+            }
+            for r in relations
+        ],
+        "glossary": [
+            {
+                "id": g["id"],
+                "term": g["term"],
+                "pronunciation": g["pronunciation"],
+                "frequency": g["frequency"]
+            }
+            for g in glossary
+        ],
+        "transcripts": [
+            {
+                "id": t["id"],
+                "filename": t["filename"],
+                "type": t.get("type", "audio"),
+                "created_at": t["created_at"]
+            }
+            for t in transcripts
+        ]
+    }
+
+# Phase 3: Glossary API
+@app.post("/api/v1/projects/{project_id}/glossary")
+async def add_glossary_term(project_id: str, term: GlossaryTermCreate):
+    """添加术语到项目术语表"""
+    if not DB_AVAILABLE:
+        raise HTTPException(status_code=500, detail="Database not available")
+    
+    db = get_db_manager()
+    project = db.get_project(project_id)
+    if not project:
+        raise HTTPException(status_code=404, detail="Project not found")
+    
+    term_id = db.add_glossary_term(
+        project_id=project_id,
+        term=term.term,
+        pronunciation=term.pronunciation
+    )
+    
+    return {
+        "id": term_id,
+        "term": term.term,
+        "pronunciation": term.pronunciation,
+        "success": True
+    }
+
+@app.get("/api/v1/projects/{project_id}/glossary")
+async def get_glossary(project_id: str):
+    """获取项目术语表"""
+    if not DB_AVAILABLE:
+        raise HTTPException(status_code=500, detail="Database not available")
+    
+    db = get_db_manager()
+    glossary = db.list_glossary(project_id)
+    return glossary
+
+@app.delete("/api/v1/glossary/{term_id}")
+async def delete_glossary_term(term_id: str):
+    """删除术语"""
+    if not DB_AVAILABLE:
+        raise HTTPException(status_code=500, detail="Database not available")
+    
+    db = get_db_manager()
+    db.delete_glossary_term(term_id)
+    return {"success": True}
+
+# Phase 3: Entity Alignment API
+@app.post("/api/v1/projects/{project_id}/align-entities")
+async def align_project_entities(project_id: str, threshold: float = 0.85):
+    """运行实体对齐算法，合并相似实体"""
+    if not DB_AVAILABLE:
+        raise HTTPException(status_code=500, detail="Database not available")
+    
+    aligner = get_aligner()
+    if not aligner:
+        raise HTTPException(status_code=500, detail="Entity aligner not available")
+    
+    db = get_db_manager()
+    entities = db.list_project_entities(project_id)
+    
+    merged_count = 0
+    merged_pairs = []
+    
+    # 使用 embedding 对齐
+    for i, entity in enumerate(entities):
+        # 跳过已合并的实体
+        existing = db.get_entity(entity.id)
+        if not existing:
+            continue
+            
+        similar = aligner.find_similar_entity(
+            project_id, 
+            entity.name, 
+            entity.definition,
+            exclude_id=entity.id,
+            threshold=threshold
+        )
+        
+        if similar:
+            # 合并实体
+            db.merge_entities(similar.id, entity.id)
+            merged_count += 1
+            merged_pairs.append({
+                "source": entity.name,
+                "target": similar.name
+            })
+    
+    return {
+        "success": True,
+        "merged_count": merged_count,
+        "merged_pairs": merged_pairs
+    }
+
 @app.get("/api/v1/projects/{project_id}/entities")
 async def get_project_entities(project_id: str):
     """获取项目的全局实体列表"""
@@ -559,7 +897,7 @@ async def get_project_entities(project_id: str):
     
     db = get_db_manager()
     entities = db.list_project_entities(project_id)
-    return [{"id": e.id, "name": e.name, "type": e.type, "definition": e.definition} for e in entities]
+    return [{"id": e.id, "name": e.name, "type": e.type, "definition": e.definition, "aliases": e.aliases} for e in entities]
 
 
 @app.get("/api/v1/projects/{project_id}/relations")
@@ -597,6 +935,7 @@ async def get_project_transcripts(project_id: str):
     return [{
         "id": t["id"],
         "filename": t["filename"],
+        "type": t.get("type", "audio"),
         "created_at": t["created_at"],
         "preview": t["full_text"][:100] + "..." if len(t["full_text"]) > 100 else t["full_text"]
     } for t in transcripts]
@@ -619,42 +958,18 @@ async def get_entity_mentions(entity_id: str):
         "confidence": m.confidence
     } for m in mentions]
 
-@app.post("/api/v1/entities/{entity_id}/merge")
-async def merge_entities_endpoint(entity_id: str, merge_req: EntityMergeRequest):
-    """合并两个实体"""
-    if not DB_AVAILABLE:
-        raise HTTPException(status_code=500, detail="Database not available")
-    
-    db = get_db_manager()
-    
-    # 验证两个实体都存在
-    source = db.get_entity(merge_req.source_entity_id)
-    target = db.get_entity(merge_req.target_entity_id)
-    
-    if not source or not target:
-        raise HTTPException(status_code=404, detail="Entity not found")
-    
-    result = db.merge_entities(merge_req.target_entity_id, merge_req.source_entity_id)
-    return {
-        "success": True,
-        "merged_entity": {
-            "id": result.id,
-            "name": result.name,
-            "type": result.type,
-            "definition": result.definition,
-            "aliases": result.aliases
-        }
-    }
-
 # Health check
 @app.get("/health")
 async def health_check():
     return {
         "status": "ok",
         "version": "0.3.0",
+        "phase": "Phase 3 - Memory & Growth",
         "oss_available": OSS_AVAILABLE,
         "tingwu_available": TINGWU_AVAILABLE,
-        "db_available": DB_AVAILABLE
+        "db_available": DB_AVAILABLE,
+        "doc_processor_available": DOC_PROCESSOR_AVAILABLE,
+        "aligner_available": ALIGNER_AVAILABLE
     }
 
 # Serve frontend

backend/requirements.txt

@@ -0,0 +1,24 @@
+# InsightFlow Backend Dependencies
+
+# Web Framework
+fastapi==0.109.0
+uvicorn[standard]==0.27.0
+python-multipart==0.0.6
+
+# HTTP Client
+httpx==0.26.0
+
+# Document Processing
+PyPDF2==3.0.1
+python-docx==1.1.0
+
+# Data Processing
+numpy==1.26.3
+
+# Aliyun SDK
+aliyun-python-sdk-core==2.14.0
+aliyun-python-sdk-oss==2.18.5
+oss2==2.18.5
+
+# Utilities
+python-dotenv==1.0.0

backend/schema.sql

@@ -16,7 +16,9 @@ CREATE TABLE IF NOT EXISTS transcripts (
     project_id TEXT NOT NULL,
     filename TEXT,
     full_text TEXT,
+    type TEXT DEFAULT 'audio',  -- 'audio' 或 'document'
     created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
+    updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
     FOREIGN KEY (project_id) REFERENCES projects(id)
 );
 
@@ -29,6 +31,7 @@ CREATE TABLE IF NOT EXISTS entities (
     type TEXT,
     definition TEXT,
     aliases TEXT,  -- JSON 数组：["别名1", "别名2"]
+    embedding TEXT,  -- JSON 数组：实体名称+定义的 embedding
     created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
     updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
     FOREIGN KEY (project_id) REFERENCES projects(id)
@@ -71,3 +74,12 @@ CREATE TABLE IF NOT EXISTS glossary (
     frequency INTEGER DEFAULT 1,
     FOREIGN KEY (project_id) REFERENCES projects(id)
 );
+
+-- 创建索引以提高查询性能
+CREATE INDEX IF NOT EXISTS idx_entities_project ON entities(project_id);
+CREATE INDEX IF NOT EXISTS idx_entities_name ON entities(name);
+CREATE INDEX IF NOT EXISTS idx_transcripts_project ON transcripts(project_id);
+CREATE INDEX IF NOT EXISTS idx_mentions_entity ON entity_mentions(entity_id);
+CREATE INDEX IF NOT EXISTS idx_mentions_transcript ON entity_mentions(transcript_id);
+CREATE INDEX IF NOT EXISTS idx_relations_project ON entity_relations(project_id);
+CREATE INDEX IF NOT EXISTS idx_glossary_project ON glossary(project_id);

frontend/app.js

@@ -1,4 +1,5 @@
-// InsightFlow Frontend - Phase 2 (Interactive Workbench)
+// InsightFlow Frontend - Phase 3 (Memory & Growth)
+// Knowledge Growth: Multi-file fusion + Entity Alignment + Document Import
 const API_BASE = '/api/v1';
 
 let currentProject = null;
@@ -7,8 +8,11 @@ let selectedEntity = null;
 let projectRelations = [];
 let projectEntities = [];
 let currentTranscript = null;
+let projectTranscripts = [];
 let editMode = false;
 let contextMenuTarget = null;
+let currentUploadTab = 'audio';
+let knowledgeBaseData = null;
 
 // Init
 document.addEventListener('DOMContentLoaded', () => {
@@ -70,6 +74,49 @@ async function uploadAudio(file) {
     return await res.json();
 }
 
+// Phase 3: Document Upload API
+async function uploadDocument(file) {
+    const formData = new FormData();
+    formData.append('file', file);
+    
+    const res = await fetch(`${API_BASE}/projects/${currentProject.id}/upload-document`, {
+        method: 'POST',
+        body: formData
+    });
+    
+    if (!res.ok) {
+        const error = await res.json();
+        throw new Error(error.detail || 'Document upload failed');
+    }
+    return await res.json();
+}
+
+// Phase 3: Knowledge Base API
+async function fetchKnowledgeBase() {
+    const res = await fetch(`${API_BASE}/projects/${currentProject.id}/knowledge-base`);
+    if (!res.ok) throw new Error('Failed to fetch knowledge base');
+    return await res.json();
+}
+
+// Phase 3: Glossary API
+async function addGlossaryTerm(term, pronunciation = '') {
+    const res = await fetch(`${API_BASE}/projects/${currentProject.id}/glossary`, {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify({ term, pronunciation })
+    });
+    if (!res.ok) throw new Error('Failed to add glossary term');
+    return await res.json();
+}
+
+async function deleteGlossaryTerm(termId) {
+    const res = await fetch(`${API_BASE}/glossary/${termId}`, {
+        method: 'DELETE'
+    });
+    if (!res.ok) throw new Error('Failed to delete glossary term');
+    return await res.json();
+}
+
 // Phase 2: Entity Edit API
 async function updateEntity(entityId, data) {
     const res = await fetch(`${API_BASE}/entities/${entityId}`, {
@@ -147,7 +194,7 @@ async function updateTranscript(transcriptId, fullText) {
 
 async function loadProjectData() {
     try {
-        // 并行加载实体和关系
+        // 并行加载实体、关系和转录列表
         const [entitiesRes, relationsRes, transcriptsRes] = await Promise.all([
             fetch(`${API_BASE}/projects/${currentProject.id}/entities`),
             fetch(`${API_BASE}/projects/${currentProject.id}/relations`),
@@ -160,32 +207,173 @@ async function loadProjectData() {
         if (relationsRes.ok) {
             projectRelations = await relationsRes.json();
         }
+        if (transcriptsRes.ok) {
+            projectTranscripts = await transcriptsRes.json();
+        }
         
         // 加载最新的转录
-        if (transcriptsRes.ok) {
+        if (projectTranscripts.length > 0) {
-            const transcripts = await transcriptsRes.json();
+            currentTranscript = await getTranscript(projectTranscripts[0].id);
-            if (transcripts.length > 0) {
+            currentData = {
-                currentTranscript = await getTranscript(transcripts[0].id);
+                transcript_id: currentTranscript.id,
-                currentData = {
+                project_id: currentProject.id,
-                    transcript_id: currentTranscript.id,
+                segments: [{ speaker: '全文', text: currentTranscript.full_text }],
-                    project_id: currentProject.id,
+                entities: projectEntities,
-                    segments: [{ speaker: '全文', text: currentTranscript.full_text }],
+                full_text: currentTranscript.full_text,
-                    entities: projectEntities,
+                created_at: currentTranscript.created_at
-                    full_text: currentTranscript.full_text,
+            };
-                    created_at: currentTranscript.created_at
+            renderTranscript();
-                };
-                renderTranscript();
-            }
         }
         
         renderGraph();
         renderEntityList();
+        renderTranscriptDropdown();
         
     } catch (err) {
         console.error('Load project data failed:', err);
     }
 }
 
+// Phase 3: View Switching
+window.switchView = function(viewName) {
+    // Update sidebar buttons
+    document.querySelectorAll('.sidebar-btn').forEach(btn => {
+        btn.classList.remove('active');
+    });
+    event.target.classList.add('active');
+    
+    if (viewName === 'workbench') {
+        document.getElementById('workbenchView').style.display = 'flex';
+        document.getElementById('knowledgeBaseView').classList.remove('show');
+    } else if (viewName === 'knowledge-base') {
+        document.getElementById('workbenchView').style.display = 'none';
+        document.getElementById('knowledgeBaseView').classList.add('show');
+        loadKnowledgeBase();
+    }
+};
+
+// Phase 3: Load Knowledge Base
+async function loadKnowledgeBase() {
+    try {
+        knowledgeBaseData = await fetchKnowledgeBase();
+        renderKnowledgeBase();
+    } catch (err) {
+        console.error('Load knowledge base failed:', err);
+    }
+}
+
+// Phase 3: Render Knowledge Base
+function renderKnowledgeBase() {
+    if (!knowledgeBaseData) return;
+    
+    // Update stats
+    document.getElementById('kbEntityCount').textContent = knowledgeBaseData.stats.entity_count;
+    document.getElementById('kbRelationCount').textContent = knowledgeBaseData.stats.relation_count;
+    document.getElementById('kbTranscriptCount').textContent = knowledgeBaseData.stats.transcript_count;
+    document.getElementById('kbGlossaryCount').textContent = knowledgeBaseData.stats.glossary_count;
+    
+    // Render entities
+    const entityGrid = document.getElementById('kbEntityGrid');
+    entityGrid.innerHTML = knowledgeBaseData.entities.map(e => `
+        <div class="kb-entity-card" onclick="selectEntity('${e.id}'); switchView('workbench');">
+            <span class="entity-type-badge type-${e.type}">${e.type}</span>
+            <div class="kb-entity-name">${e.name}</div>
+            <div class="kb-entity-def">${e.definition || '暂无定义'}</div>
+            <div class="kb-entity-meta">提及 ${e.mention_count} 次 | 出现在 ${e.appears_in.length} 个文件中</div>
+        </div>
+    `).join('');
+    
+    // Render relations
+    const relationsList = document.getElementById('kbRelationsList');
+    relationsList.innerHTML = knowledgeBaseData.relations.map(r => `
+        <div class="kb-glossary-item">
+            <div>
+                <strong>${r.source_name}</strong> 
+                <span style="color:#666;">→ ${r.type} →</span> 
+                <strong>${r.target_name}</strong>
+                <div style="font-size:0.8rem;color:#666;margin-top:4px;">${r.evidence || '无证据'}</div>
+            </div>
+        </div>
+    `).join('');
+    
+    // Render glossary
+    const glossaryList = document.getElementById('kbGlossaryList');
+    glossaryList.innerHTML = knowledgeBaseData.glossary.map(g => `
+        <div class="kb-glossary-item">
+            <div>
+                <strong>${g.term}</strong>
+                ${g.pronunciation ? `<span style="color:#666;font-size:0.85rem;"> (${g.pronunciation})</span>` : ''}
+                <span style="color:#00d4ff;font-size:0.8rem;margin-left:8px;">出现 ${g.frequency} 次</span>
+            </div>
+            <button class="btn-icon" onclick="deleteGlossaryTerm('${g.id}').then(loadKnowledgeBase)">删除</button>
+        </div>
+    `).join('');
+    
+    // Render transcripts
+    const transcriptsList = document.getElementById('kbTranscriptsList');
+    transcriptsList.innerHTML = knowledgeBaseData.transcripts.map(t => `
+        <div class="kb-transcript-item">
+            <div>
+                <span class="file-type-icon type-${t.type}">${t.type === 'audio' ? '🎵' : '📄'}</span>
+                <span style="margin-left:8px;">${t.filename}</span>
+            </div>
+            <span style="color:#666;font-size:0.8rem;">${new Date(t.created_at).toLocaleDateString()}</span>
+        </div>
+    `).join('');
+}
+
+// Phase 3: KB Tab Switching
+window.switchKBTab = function(tabName) {
+    document.querySelectorAll('.kb-nav-item').forEach(item => {
+        item.classList.remove('active');
+    });
+    event.target.classList.add('active');
+    
+    document.querySelectorAll('.kb-section').forEach(section => {
+        section.classList.remove('active');
+    });
+    document.getElementById(`kb${tabName.charAt(0).toUpperCase() + tabName.slice(1)}Section`).classList.add('active');
+};
+
+// Phase 3: Transcript Dropdown
+window.toggleTranscriptDropdown = function() {
+    const dropdown = document.getElementById('transcriptDropdown');
+    dropdown.classList.toggle('show');
+};
+
+function renderTranscriptDropdown() {
+    const dropdown = document.getElementById('transcriptDropdown');
+    if (!dropdown || projectTranscripts.length === 0) return;
+    
+    dropdown.innerHTML = projectTranscripts.map(t => `
+        <div class="transcript-option ${currentTranscript && currentTranscript.id === t.id ? 'active' : ''}" 
+             onclick="switchTranscript('${t.id}')">
+            <span class="file-type-icon type-${t.type || 'audio'}">${(t.type || 'audio') === 'audio' ? '🎵' : '📄'}</span>
+            <span style="margin-left:4px;">${t.filename}</span>
+        </div>
+    `).join('');
+}
+
+window.switchTranscript = async function(transcriptId) {
+    try {
+        currentTranscript = await getTranscript(transcriptId);
+        currentData = {
+            transcript_id: currentTranscript.id,
+            project_id: currentProject.id,
+            segments: [{ speaker: '全文', text: currentTranscript.full_text }],
+            entities: projectEntities,
+            full_text: currentTranscript.full_text,
+            created_at: currentTranscript.created_at
+        };
+        renderTranscript();
+        renderTranscriptDropdown();
+        document.getElementById('transcriptDropdown').classList.remove('show');
+    } catch (err) {
+        console.error('Switch transcript failed:', err);
+        alert('切换文件失败');
+    }
+};
+
 // Phase 2: Transcript Edit Mode
 window.toggleEditMode = function() {
     editMode = !editMode;
@@ -255,7 +443,7 @@ function renderTranscript() {
     const div = document.createElement('div');
     div.className = 'segment';
     div.innerHTML = `
-        <div class="speaker">转录文本</div>
+        <div class="speaker">${currentTranscript.filename || '转录文本'}</div>
         <div class="segment-text">${text}</div>
     `;
     
@@ -311,7 +499,7 @@ function renderGraph() {
             .attr('y', '50%')
             .attr('text-anchor', 'middle')
             .attr('fill', '#666')
-            .text('暂无实体数据，请上传音频');
+            .text('暂无实体数据，请上传音频或文档');
         return;
     }
     
@@ -458,7 +646,7 @@ function renderEntityList() {
     container.innerHTML = '<h3 style="margin-bottom:12px;color:#888;font-size:0.9rem;">项目实体</h3>';
     
     if (!projectEntities || projectEntities.length === 0) {
-        container.innerHTML += '<p style="color:#666;font-size:0.85rem;">暂无实体，请上传音频文件</p>';
+        container.innerHTML += '<p style="color:#666;font-size:0.85rem;">暂无实体，请上传音频或文档文件</p>';
         return;
     }
     
@@ -788,6 +976,28 @@ window.createEntityFromSelection = async function() {
     }
 };
 
+// Phase 3: Upload Tab Switching
+window.switchUploadTab = function(tab) {
+    currentUploadTab = tab;
+    document.querySelectorAll('.upload-tab').forEach(t => t.classList.remove('active'));
+    event.target.classList.add('active');
+    
+    const hint = document.getElementById('uploadHint');
+    if (tab === 'audio') {
+        hint.textContent = '支持 MP3, WAV, M4A (最大 500MB)';
+    } else {
+        hint.textContent = '支持 PDF, DOCX, DOC, TXT, MD';
+    }
+};
+
+window.triggerFileSelect = function() {
+    if (currentUploadTab === 'audio') {
+        document.getElementById('fileInput').click();
+    } else {
+        document.getElementById('docInput').click();
+    }
+};
+
 // Show/hide upload
 window.showUpload = function() {
     const el = document.getElementById('uploadOverlay');
@@ -799,49 +1009,120 @@ window.hideUpload = function() {
     if (el) el.classList.remove('show');
 };
 
+// Phase 3: Glossary Modal
+window.showAddTermModal = function() {
+    document.getElementById('glossaryModal').classList.add('show');
+};
+
+window.hideGlossaryModal = function() {
+    document.getElementById('glossaryModal').classList.remove('show');
+    document.getElementById('glossaryTerm').value = '';
+    document.getElementById('glossaryPronunciation').value = '';
+};
+
+window.saveGlossaryTerm = async function() {
+    const term = document.getElementById('glossaryTerm').value.trim();
+    const pronunciation = document.getElementById('glossaryPronunciation').value.trim();
+    
+    if (!term) {
+        alert('请输入术语');
+        return;
+    }
+    
+    try {
+        await addGlossaryTerm(term, pronunciation);
+        hideGlossaryModal();
+        loadKnowledgeBase();
+    } catch (err) {
+        console.error('Add term failed:', err);
+        alert('添加术语失败: ' + err.message);
+    }
+};
+
 // Upload handling
 function initUpload() {
-    const input = document.getElementById('fileInput');
+    // Audio upload
+    const audioInput = document.getElementById('fileInput');
+    if (audioInput) {
+        audioInput.addEventListener('change', async (e) => {
+            if (!e.target.files.length) return;
+            await handleFileUpload(e.target.files[0], 'audio');
+        });
+    }
+    
+    // Document upload
+    const docInput = document.getElementById('docInput');
+    if (docInput) {
+        docInput.addEventListener('change', async (e) => {
+            if (!e.target.files.length) return;
+            await handleFileUpload(e.target.files[0], 'document');
+        });
+    }
+}
+
+async function handleFileUpload(file, type) {
     const overlay = document.getElementById('uploadOverlay');
     
-    if (!input) return;
+    overlay.innerHTML = `
+        <div style="text-align:center;">
+            <h2>正在分析...</h2>
+            <p style="color:#666;margin-top:10px;">${file.name}</p>
+            <p style="color:#888;margin-top:20px;font-size:0.9rem;">${type === 'audio' ? 'ASR转录 + 实体提取中' : '文档解析 + 实体提取中'}</p>
+        </div>
+    `;
     
-    input.addEventListener('change', async (e) => {
+    try {
-        if (!e.target.files.length) return;
+        let result;
+        if (type === 'audio') {
+            result = await uploadAudio(file);
+        } else {
+            result = await uploadDocument(file);
+        }
         
-        const file = e.target.files[0];
+        // 更新当前数据
-        if (overlay) {
+        currentData = result;
-            overlay.innerHTML = `
+        
-                <div style="text-align:center;">
+        // 重新加载项目数据
-                    <h2>正在分析...</h2>
+        await loadProjectData();
-                    <p style="color:#666;margin-top:10px;">${file.name}</p>
+        
-                    <p style="color:#888;margin-top:20px;font-size:0.9rem;">ASR转录 + 实体提取中</p>
+        // 重置上传界面
+        overlay.innerHTML = `
+            <div class="upload-box">
+                <h2 style="margin-bottom:10px;">上传文件</h2>
+                <div class="upload-tabs">
+                    <div class="upload-tab active" onclick="switchUploadTab('audio')">🎵 音频</div>
+                    <div class="upload-tab" onclick="switchUploadTab('document')">📄 文档</div>
                 </div>
-            `;
+                <p style="color:#666;" id="uploadHint">支持 MP3, WAV, M4A (最大 500MB)</p>
-        }
+                <input type="file" id="fileInput" accept="audio/*" hidden>
+                <input type="file" id="docInput" accept=".pdf,.docx,.doc,.txt,.md" hidden>
+                <button class="btn" onclick="triggerFileSelect()">选择文件</button>
+                <br><br>
+                <button class="btn btn-secondary" onclick="hideUpload()">取消</button>
+            </div>
+        `;
         
-        try {
+        // 重新绑定事件
-            const result = await uploadAudio(file);
+        initUpload();
-            
+        overlay.classList.remove('show');
-            // 更新当前数据
+        
-            currentData = result;
+    } catch (err) {
-            
+        console.error('Upload failed:', err);
-            // 重新加载项目数据
+        overlay.innerHTML = `
-            await loadProjectData();
+            <div style="text-align:center;">
-            
+                <h2 style="color:#ff6b6b;">分析失败</h2>
-            if (overlay) overlay.classList.remove('show');
+                <p style="color:#666;margin-top:10px;">${err.message}</p>
-            
+                <button class="btn" onclick="location.reload()" style="margin-top:20px;">重试</button>
-        } catch (err) {
+            </div>
-            console.error('Upload failed:', err);
+        `;
-            if (overlay) {
+    }
-                overlay.innerHTML = `
-                    <div style="text-align:center;">
-                        <h2 style="color:#ff6b6b;">分析失败</h2>
-                        <p style="color:#666;margin-top:10px;">${err.message}</p>
-                        <button class="btn" onclick="location.reload()" style="margin-top:20px;">重试</button>
-                    </div>
-                `;
-            }
-        }
-    });
 }
+
+// Close dropdown when clicking outside
+document.addEventListener('click', (e) => {
+    const dropdown = document.getElementById('transcriptDropdown');
+    const selector = document.querySelector('.transcript-selector');
+    if (dropdown && selector && !selector.contains(e.target)) {
+        dropdown.classList.remove('show');
+    }
+});

frontend/workbench.html

@@ -3,7 +3,7 @@
 <head>
     <meta charset="UTF-8">
     <meta name="viewport" content="width=device-width, initial-scale=1.0">
-    <title>InsightFlow - 知识工作台 (Phase 2)</title>
+    <title>InsightFlow - 知识工作台 (Phase 3)</title>
     <script src="https://d3js.org/d3.v7.min.js"></script>
     <style>
         * { margin: 0; padding: 0; box-sizing: border-box; }
@@ -46,10 +46,44 @@
             color: #888;
             font-size: 0.9rem;
         }
+        .header-actions {
+            display: flex;
+            gap: 10px;
+        }
         .main {
             display: flex;
             height: calc(100vh - 50px);
         }
+        .sidebar {
+            width: 60px;
+            background: #111;
+            border-right: 1px solid #222;
+            display: flex;
+            flex-direction: column;
+            align-items: center;
+            padding: 10px 0;
+        }
+        .sidebar-btn {
+            width: 44px;
+            height: 44px;
+            background: transparent;
+            border: none;
+            color: #666;
+            font-size: 1.2rem;
+            cursor: pointer;
+            border-radius: 8px;
+            margin-bottom: 8px;
+            transition: all 0.2s;
+        }
+        .sidebar-btn:hover, .sidebar-btn.active {
+            background: #1a1a1a;
+            color: #00d4ff;
+        }
+        .content-area {
+            flex: 1;
+            display: flex;
+            overflow: hidden;
+        }
         .editor-panel {
             width: 50%;
             border-right: 1px solid #222;
@@ -199,10 +233,29 @@
             border-radius: 16px;
             padding: 60px;
             text-align: center;
+            max-width: 500px;
         }
         .upload-box:hover {
             border-color: #00d4ff;
         }
+        .upload-tabs {
+            display: flex;
+            gap: 10px;
+            margin-bottom: 20px;
+            justify-content: center;
+        }
+        .upload-tab {
+            padding: 8px 16px;
+            background: #1a1a1a;
+            border: 1px solid #333;
+            border-radius: 6px;
+            cursor: pointer;
+            color: #888;
+        }
+        .upload-tab.active {
+            border-color: #00d4ff;
+            color: #00d4ff;
+        }
         .btn {
             background: linear-gradient(90deg, #00d4ff, #7b2cbf);
             color: white;
@@ -373,6 +426,164 @@
         .node-label {
             pointer-events: none;
         }
+        /* Phase 3: Knowledge Base Panel */
+        .kb-panel {
+            width: 100%;
+            height: 100%;
+            display: none;
+            flex-direction: column;
+            background: #0a0a0a;
+        }
+        .kb-panel.show {
+            display: flex;
+        }
+        .kb-header {
+            padding: 16px 20px;
+            background: #141414;
+            border-bottom: 1px solid #222;
+            display: flex;
+            justify-content: space-between;
+            align-items: center;
+        }
+        .kb-stats {
+            display: flex;
+            gap: 24px;
+        }
+        .kb-stat {
+            text-align: center;
+        }
+        .kb-stat-value {
+            font-size: 1.5rem;
+            font-weight: 600;
+            color: #00d4ff;
+        }
+        .kb-stat-label {
+            font-size: 0.75rem;
+            color: #666;
+        }
+        .kb-content {
+            flex: 1;
+            display: flex;
+            overflow: hidden;
+        }
+        .kb-sidebar {
+            width: 200px;
+            background: #111;
+            border-right: 1px solid #222;
+            padding: 16px 0;
+        }
+        .kb-nav-item {
+            padding: 12px 20px;
+            cursor: pointer;
+            color: #888;
+            border-left: 3px solid transparent;
+        }
+        .kb-nav-item:hover {
+            background: #1a1a1a;
+            color: #e0e0e0;
+        }
+        .kb-nav-item.active {
+            background: #1a1a1a;
+            color: #00d4ff;
+            border-left-color: #00d4ff;
+        }
+        .kb-main {
+            flex: 1;
+            padding: 20px;
+            overflow-y: auto;
+        }
+        .kb-section {
+            display: none;
+        }
+        .kb-section.active {
+            display: block;
+        }
+        .kb-entity-grid {
+            display: grid;
+            grid-template-columns: repeat(auto-fill, minmax(280px, 1fr));
+            gap: 16px;
+        }
+        .kb-entity-card {
+            background: #141414;
+            border: 1px solid #222;
+            border-radius: 8px;
+            padding: 16px;
+            cursor: pointer;
+            transition: all 0.2s;
+        }
+        .kb-entity-card:hover {
+            border-color: #00d4ff;
+        }
+        .kb-entity-name {
+            font-weight: 600;
+            margin-bottom: 4px;
+        }
+        .kb-entity-def {
+            font-size: 0.85rem;
+            color: #888;
+            margin-bottom: 8px;
+        }
+        .kb-entity-meta {
+            font-size: 0.75rem;
+            color: #666;
+        }
+        .kb-glossary-item {
+            display: flex;
+            justify-content: space-between;
+            align-items: center;
+            padding: 12px 16px;
+            background: #141414;
+            border-radius: 6px;
+            margin-bottom: 8px;
+        }
+        .kb-transcript-item {
+            padding: 12px 16px;
+            background: #141414;
+            border-radius: 6px;
+            margin-bottom: 8px;
+            display: flex;
+            justify-content: space-between;
+            align-items: center;
+        }
+        .file-type-icon {
+            padding: 4px 8px;
+            border-radius: 4px;
+            font-size: 0.7rem;
+            font-weight: 600;
+        }
+        .type-audio { background: #7b2cbf; }
+        .type-document { background: #00d4ff; color: #000; }
+        /* Transcript selector */
+        .transcript-selector {
+            position: relative;
+        }
+        .transcript-dropdown {
+            position: absolute;
+            top: 100%;
+            right: 0;
+            background: #1a1a1a;
+            border: 1px solid #333;
+            border-radius: 8px;
+            min-width: 200px;
+            max-height: 300px;
+            overflow-y: auto;
+            display: none;
+            z-index: 100;
+        }
+        .transcript-dropdown.show {
+            display: block;
+        }
+        .transcript-option {
+            padding: 10px 16px;
+            cursor: pointer;
+            border-bottom: 1px solid #222;
+        }
+        .transcript-option:hover {
+            background: #2a2a2a;
+        }
+        .transcript-option.active {
+            background: #00d4ff22;
+        }
     </style>
 </head>
 <body>
@@ -381,35 +592,113 @@
             <a href="/" class="back-link">← 返回项目列表</a>
             <span class="project-name" id="projectName">加载中...</span>
         </div>
-        <button class="btn btn-small" onclick="showUpload()">+ 上传音频</button>
+        <div class="header-actions">
+            <button class="btn btn-small" onclick="showUpload()">+ 上传文件</button>
+        </div>
     </div>
     
     <div class="main">
-        <div class="editor-panel">
+        <!-- Sidebar -->
-            <div class="panel-header">
+        <div class="sidebar">
-                <span>📄 转录文本</span>
+            <button class="sidebar-btn active" onclick="switchView('workbench')" title="工作台">📝</button>
-                <div class="panel-actions">
+            <button class="sidebar-btn" onclick="switchView('knowledge-base')" title="知识库">📚</button>
-                    <button class="btn-icon" onclick="toggleEditMode()" id="editBtn">✏️ 编辑</button>
-                    <button class="btn-icon" onclick="saveTranscript()" id="saveBtn" style="display:none;">💾 保存</button>
-                </div>
-            </div>
-            <div class="transcript-content" id="transcriptContent">
-                <div class="empty-state">
-                    <p style="color:#666;">暂无转录内容</p>
-                    <button class="btn" onclick="showUpload()">上传音频</button>
-                </div>
-            </div>
         </div>
         
-        <div class="graph-panel">
+        <!-- Content Area -->
-            <div class="panel-header">
+        <div class="content-area">
-                <span>🔗 知识图谱</span>
+            <!-- Workbench View -->
-                <span style="font-size:0.8rem;color:#666;">右键节点编辑 | 拖拽建立关系</span>
+            <div id="workbenchView" class="workbench-view" style="display: flex; width: 100%;">
+                <div class="editor-panel">
+                    <div class="panel-header">
+                        <div style="display: flex; align-items: center; gap: 12px;">
+                            <span>📄 转录文本</span>
+                            <div class="transcript-selector">
+                                <button class="btn-icon" onclick="toggleTranscriptDropdown()">📁 选择文件</button>
+                                <div class="transcript-dropdown" id="transcriptDropdown"></div>
+                            </div>
+                        </div>
+                        <div class="panel-actions">
+                            <button class="btn-icon" onclick="toggleEditMode()" id="editBtn">✏️ 编辑</button>
+                            <button class="btn-icon" onclick="saveTranscript()" id="saveBtn" style="display:none;">💾 保存</button>
+                        </div>
+                    </div>
+                    <div class="transcript-content" id="transcriptContent">
+                        <div class="empty-state">
+                            <p style="color:#666;">暂无转录内容</p>
+                            <button class="btn" onclick="showUpload()">上传音频或文档</button>
+                        </div>
+                    </div>
+                </div>
+                
+                <div class="graph-panel">
+                    <div class="panel-header">
+                        <span>🔗 知识图谱</span>
+                        <span style="font-size:0.8rem;color:#666;">右键节点编辑 | 拖拽建立关系</span>
+                    </div>
+                    <svg id="graph-svg"></svg>
+                    <div class="entity-list" id="entityList">
+                        <h3 style="margin-bottom:12px;color:#888;font-size:0.9rem;">项目实体</h3>
+                        <p style="color:#666;font-size:0.85rem;">暂无实体数据</p>
+                    </div>
+                </div>
             </div>
-            <svg id="graph-svg"></svg>
+            
-            <div class="entity-list" id="entityList">
+            <!-- Knowledge Base View -->
-                <h3 style="margin-bottom:12px;color:#888;font-size:0.9rem;">项目实体</h3>
+            <div id="knowledgeBaseView" class="kb-panel">
-                <p style="color:#666;font-size:0.85rem;">暂无实体数据</p>
+                <div class="kb-header">
+                    <h2>📚 项目知识库</h2>
+                    <div class="kb-stats">
+                        <div class="kb-stat">
+                            <div class="kb-stat-value" id="kbEntityCount">0</div>
+                            <div class="kb-stat-label">实体</div>
+                        </div>
+                        <div class="kb-stat">
+                            <div class="kb-stat-value" id="kbRelationCount">0</div>
+                            <div class="kb-stat-label">关系</div>
+                        </div>
+                        <div class="kb-stat">
+                            <div class="kb-stat-value" id="kbTranscriptCount">0</div>
+                            <div class="kb-stat-label">文件</div>
+                        </div>
+                        <div class="kb-stat">
+                            <div class="kb-stat-value" id="kbGlossaryCount">0</div>
+                            <div class="kb-stat-label">术语</div>
+                        </div>
+                    </div>
+                </div>
+                <div class="kb-content">
+                    <div class="kb-sidebar">
+                        <div class="kb-nav-item active" onclick="switchKBTab('entities')">🏷️ 实体</div>
+                        <div class="kb-nav-item" onclick="switchKBTab('relations')">🔗 关系</div>
+                        <div class="kb-nav-item" onclick="switchKBTab('glossary')">📖 术语表</div>
+                        <div class="kb-nav-item" onclick="switchKBTab('transcripts')">📁 文件</div>
+                    </div>
+                    <div class="kb-main">
+                        <!-- Entities Section -->
+                        <div class="kb-section active" id="kbEntitiesSection">
+                            <h3 style="margin-bottom:16px;">所有实体</h3>
+                            <div class="kb-entity-grid" id="kbEntityGrid"></div>
+                        </div>
+                        <!-- Relations Section -->
+                        <div class="kb-section" id="kbRelationsSection">
+                            <h3 style="margin-bottom:16px;">所有关系</h3>
+                            <div id="kbRelationsList"></div>
+                        </div>
+                        <!-- Glossary Section -->
+                        <div class="kb-section" id="kbGlossarySection">
+                            <div style="display:flex;justify-content:space-between;align-items:center;margin-bottom:16px;">
+                                <h3>术语表</h3>
+                                <button class="btn btn-small" onclick="showAddTermModal()">+ 添加术语</button>
+                            </div>
+                            <div id="kbGlossaryList"></div>
+                        </div>
+                        <!-- Transcripts Section -->
+                        <div class="kb-section" id="kbTranscriptsSection">
+                            <h3 style="margin-bottom:16px;">所有文件</h3>
+                            <div id="kbTranscriptsList"></div>
+                        </div>
+                    </div>
+                </div>
             </div>
         </div>
     </div>
@@ -417,10 +706,15 @@
     <!-- Upload Modal -->
     <div class="upload-overlay" id="uploadOverlay">
         <div class="upload-box">
-            <h2 style="margin-bottom:10px;">上传音频分析</h2>
+            <h2 style="margin-bottom:10px;">上传文件</h2>
-            <p style="color:#666;">支持 MP3, WAV, M4A (最大 500MB)</p>
+            <div class="upload-tabs">
+                <div class="upload-tab active" onclick="switchUploadTab('audio')">🎵 音频</div>
+                <div class="upload-tab" onclick="switchUploadTab('document')">📄 文档</div>
+            </div>
+            <p style="color:#666;" id="uploadHint">支持 MP3, WAV, M4A (最大 500MB)</p>
             <input type="file" id="fileInput" accept="audio/*" hidden>
-            <button class="btn" onclick="document.getElementById('fileInput').click()">选择文件</button>
+            <input type="file" id="docInput" accept=".pdf,.docx,.doc,.txt,.md" hidden>
+            <button class="btn" onclick="triggerFileSelect()">选择文件</button>
             <br><br>
             <button class="btn btn-secondary" onclick="hideUpload()">取消</button>
         </div>
@@ -514,6 +808,25 @@
         </div>
     </div>
     
+    <!-- Add Glossary Term Modal -->
+    <div class="modal-overlay" id="glossaryModal">
+        <div class="modal">
+            <h3 class="modal-header">添加术语</h3>
+            <div class="form-group">
+                <label>术语</label>
+                <input type="text" id="glossaryTerm" placeholder="术语名称">
+            </div>
+            <div class="form-group">
+                <label>发音提示 (可选)</label>
+                <input type="text" id="glossaryPronunciation" placeholder="如: K8s 发音为 Kubernetes">
+            </div>
+            <div class="modal-actions">
+                <button class="btn btn-secondary" onclick="hideGlossaryModal()">取消</button>
+                <button class="btn" onclick="saveGlossaryTerm()">添加</button>
+            </div>
+        </div>
+    </div>
+    
     <!-- Context Menu -->
     <div class="context-menu" id="contextMenu">
         <div class="context-menu-item" onclick="editEntity()">✏️ 编辑实体</div>