新增客服重定向接口

添加专用脚本用于启动rag2_0.api.intent_recognition_api服务
脚本功能包括检测并结束现有screen会话，清理占用端口，最后启动新服务

.env

@@ -1,27 +1,27 @@
 OPENAI_API_BASE=https://api.siliconflow.cn/v1/
 MODEL_NAME=deepseek-ai/DeepSeek-V3
 
-RERANKER_BASE_URL=http://10.1.16.39:9995
+# RERANKER_BASE_URL=http://10.1.16.39:9995
-RERANKER_MODEL_NAME=bge-reranker-v2-m3
+# RERANKER_MODEL_NAME=bge-reranker-v2-m3
-RERANKER_API_KEY=test
+# RERANKER_API_KEY=test
 
-EMBEDDING_BASE_URL=http://10.1.16.39:9995
+# EMBEDDING_BASE_URL=http://10.1.16.39:9995
-EMBEDDING_MODEL_NAME=bge-m3
+# EMBEDDING_MODEL_NAME=bge-m3
-EMBEDDING_API_KEY=test
+# EMBEDDING_API_KEY=test
 
 
-DIFY_BSAE_URL=http://10.1.16.39/v1
+# DIFY_BSAE_URL=http://10.1.16.39/v1
-DIFY_APP_KEY=app-CPoOMaGDsLRPAe9TW7Xjhszy
+# DIFY_APP_KEY=app-CPoOMaGDsLRPAe9TW7Xjhszy
-DIFY_DATASET_KEY=dataset-skLjmPVonjHo119OWNf3kAmY
+# DIFY_DATASET_KEY=dataset-skLjmPVonjHo119OWNf3kAmY
 
-DIFY_PG_HOST = 10.1.16.39
+# DIFY_PG_HOST = 10.1.16.39
-DIFY_PG_PORT = 5432
+# DIFY_PG_PORT = 5432
-DIFY_PG_USER = postgres
+# DIFY_PG_USER = postgres
-DIFY_PG_PASSWORD = difyai123456
+# DIFY_PG_PASSWORD = difyai123456
-DIFY_PG_DATABASE = dify
+# DIFY_PG_DATABASE = dify
 
 
-ENABLE_LANGFUSE=true
+# ENABLE_LANGFUSE=true
-LANGFUSE_PUBLIC_KEY=pk-lf-4e9b7cbe-528c-4697-b73c-33257a60072c
+# LANGFUSE_PUBLIC_KEY=pk-lf-4e9b7cbe-528c-4697-b73c-33257a60072c
-LANGFUSE_SECRET_KEY=sk-lf-cd8a78c5-2538-455e-a85a-87b6e1aa69d0
+# LANGFUSE_SECRET_KEY=sk-lf-cd8a78c5-2538-455e-a85a-87b6e1aa69d0
-LANGFUSE_HOST=http://10.1.6.34:3000
+# LANGFUSE_HOST=http://10.1.6.34:3000

.vscode/launch.json

@@ -10,7 +10,10 @@
             "request": "launch",
             "program": "${file}",
             "console": "integratedTerminal",
-            "justMyCode": true
+            "justMyCode": true,
+            "env": {
+                "PYTHONPATH": "${workspaceFolder}"
+            }
         },
         {
             "name": "IntentRecognition",
@@ -18,7 +21,10 @@
             "request": "launch",
             "program": "${workspaceFolder}/rag2_0/demo/intent_recognition_example.py",
             "console": "integratedTerminal",
-            "justMyCode": true
+            "justMyCode": true,
+            "env": {
+                "PYTHONPATH": "${workspaceFolder}"
+            }
         }
     ]
 }

api_key.txt

@@ -57,7 +57,6 @@ sk-benuasjbhbxvdmgxishibmtpfyieamlfclmdclfbqloqsmaf
 sk-ufmqbuplpjvzzlzohvsxertwgnguhipsbajxnxecvvccozly
 sk-rypfoscrczeelowmrsixiuyunyqmqvknaprsnzmdguwzrkzx
 sk-lucemnosmcxuwedvzilpefuxjnyvaxldpbgaqwnwalxmntul
-sk-niymkyuzpyovndvvqvpaniiqfgoofnxczhdmjjessiocbeul
 sk-cxlvgeuxavxfcajprxietuqyqjngtbrwrmrmrioxmgtbkpci
 sk-vjjsuzntqbhcmelfsuquqyoxjivxcfwyxnrhpwzobgxlpmrv
 sk-hbgctnpvntsnelveaudpekyncfgstdfazezboxmcgjvudzyg

rag2_0/api/AnswerType_api.py

@@ -29,7 +29,6 @@ def main(query: str) -> dict:
 
 
 import sys
-sys.path.append(os.getcwd())
 from rag2_0.dify.DifyQueryRetrieval import DifyQueryRetrieval
 
 # 定义数据库路径

rag2_0/api/DifyQueryRetrieval_api.py

@@ -18,7 +18,6 @@ import logging
 load_dotenv()
 
 import sys
-sys.path.append(os.getcwd())
 from rag2_0.dify.DifyQueryRetrieval import DifyQueryRetrieval
 
 # 确保日志目录存在

rag2_0/api/create_qingdan_dinge_database.py

@@ -5,7 +5,6 @@ import pandas as pd
 from openpyxl import load_workbook
 import logging
 import numpy as np
-sys.path.append(os.getcwd())
 from rag2_0.tool.ModelTool import XinferenceEmbeddings
 from langchain_community.vectorstores import SQLiteVSS
 

rag2_0/api/intent_recognition_api.py

@@ -15,8 +15,8 @@ import logging
 load_dotenv()
 
 import sys
-sys.path.append(os.getcwd())
 from rag2_0.intent_recognition import AsyncIntentRecognizer
+from rag2_0.api.kefu_redirect_url import router as kefu_router
 
 # 确保日志目录存在
 os.makedirs('data/logs', exist_ok=True)
@@ -85,6 +85,9 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
+# 注册外部路由
+app.include_router(kefu_router)
+
 # 全局变量存储AsyncIntentRecognizer实例
 _instance = None
 

rag2_0/api/kefu_redirect_url.py

@@ -0,0 +1,92 @@
+from fastapi import APIRouter
+from fastapi.responses import RedirectResponse
+import os
+import sqlite3
+import threading
+import time
+from queue import Queue, Full
+
+
+router = APIRouter()
+
+# 以当前文件为基准的相对路径：../../data/db
+PROJECT_ROOT = os.getcwd()
+DB_DIR = os.path.join(PROJECT_ROOT, "data", "db")
+DB_FILE = os.path.join(DB_DIR, "redirects.sqlite3")
+TABLE_SQL = (
+    "CREATE TABLE IF NOT EXISTS redirects ("
+    " msg_id TEXT PRIMARY KEY,"
+    " url TEXT NOT NULL"
+    ")"
+)
+
+
+def _ensure_db():
+    """确保数据库与表存在。"""
+    os.makedirs(DB_DIR, exist_ok=True)
+    with sqlite3.connect(DB_FILE) as conn:
+        cur = conn.cursor()
+        cur.execute(TABLE_SQL)
+        conn.commit()
+
+
+def save_redirect(msg_id: str, url: str) -> None:
+    """将 msg_id 与 url 写入 SQLite，若已存在则忽略。
+
+    使用 INSERT OR IGNORE 结合 PRIMARY KEY(msg_id) 来避免重复写入。
+    """
+    _ensure_db()
+    with sqlite3.connect(DB_FILE) as conn:
+        cur = conn.cursor()
+        cur.execute(
+            "INSERT OR IGNORE INTO redirects (msg_id, url) VALUES (?, ?)",
+            (msg_id, url),
+        )
+        conn.commit()
+
+
+# ========= 异步写库队列与后台线程 =========
+_write_queue: "Queue[tuple[str, str]]" = Queue(maxsize=10000)
+
+
+def _write_worker():
+    _ensure_db()
+    while True:
+        try:
+            msg_id, url = _write_queue.get()
+            try:
+                with sqlite3.connect(DB_FILE) as conn:
+                    cur = conn.cursor()
+                    cur.execute(
+                        "INSERT OR IGNORE INTO redirects (msg_id, url) VALUES (?, ?)",
+                        (msg_id, url),
+                    )
+                    conn.commit()
+            except Exception:
+                # 失败忽略，避免阻断工作线程
+                pass
+            finally:
+                _write_queue.task_done()
+        except Exception:
+            # 防御性 sleep，避免异常导致CPU空转
+            time.sleep(0.1)
+
+
+_worker_thread = threading.Thread(target=_write_worker, daemon=True)
+_worker_thread.start()
+
+
+@router.get("/kefu_login", summary="客服登录页重定向")
+async def kefu_redirect(msg_id:str):
+    """重定向到客服登录页。"""
+    target_url = "https://www.booway.com.cn/kefu/toLoginPage"
+    # 写入 SQLite：若 msg_id 已存在将不会重复写入
+    try:
+        if msg_id:
+            # 走异步队列
+            _write_queue.put_nowait((msg_id, target_url))
+    except Exception:
+        # 出于稳健性考虑，即使写库失败也不影响重定向
+        pass
+    return RedirectResponse(target_url, status_code=302)
+

rag2_0/api/query_dinge_qingdan_api.py

@@ -10,7 +10,6 @@ import sys
 import os
 
 # 导入ExcelToSQLiteProcessor类
-sys.path.append(os.getcwd())
 from rag2_0.api.create_qingdan_dinge_database import ExcelToSQLiteProcessor, create_db
 # 导入向量检索相关类
 from rag2_0.tool.ModelTool import XinferenceEmbeddings

rag2_0/demo/dialogue_to_workorder.py

@@ -18,8 +18,6 @@ from tqdm import tqdm
 import glob
 import shutil
 
-# 将项目根目录添加到Python路径
-sys.path.append(os.getcwd())
 from rag2_0.tool.ModelTool import OpenAiLLM
 
 load_dotenv()

rag2_0/demo/intent_recognition_example.py

@@ -20,7 +20,6 @@ import argparse
 from typing import List, Dict, Any
 from langchain.output_parsers import PydanticOutputParser
 from pydantic import BaseModel, Field
-sys.path.append(os.getcwd())
 from rag2_0.intent_recognition import AsyncIntentRecognizer
 from rag2_0.dify.DifyQueryRetrieval import DifyQueryRetrieval
 from rag2_0.intent_recognition.DataModels import Classification

rag2_0/demo/vectorize_save_noun.py

@@ -10,7 +10,6 @@ import os
 import json
 from dotenv import load_dotenv
 import sys
-sys.path.append(os.getcwd())
 from rag2_0.intent_recognition import ProfessionalNounVectorizer
 import logging
 

rag2_0/dify/DifyCompareTest.py

@@ -15,7 +15,6 @@ from datetime import datetime
 import os
 from langchain_core.output_parsers import JsonOutputParser
 
-sys.path.append(os.getcwd())
 from rag2_0.dify.dify_client import ChatClient
 from rag2_0.tool.ModelTool import OpenAiLLM
 from rag2_0.dify.dify_tool import DifyTool

rag2_0/dify/DifyQueryRetrieval.py

@@ -6,7 +6,6 @@ import logging
 import time
 import asyncio
 import httpx
-sys.path.append(os.getcwd())
 
 from rag2_0.dify.dify_client.client import DifyClient, KnowledgeBaseClient
 from rag2_0.tool.ModelTool import XinferenceReRankerModel

rag2_0/dify/GenerateSoftwareWikiLibrary.py

@@ -5,7 +5,6 @@ import sys
 from dotenv import load_dotenv
 
 load_dotenv()
-sys.path.append(os.getcwd())
 
 from rag2_0.dify.dify_client import DifyApi
 

rag2_0/dify/WorkorderToDify.py

@@ -17,7 +17,6 @@ logging.basicConfig(
     ]
 )
 
-sys.path.append(os.getcwd())
 import rag2_0.dify.dify_client.dify_api as DifyApi
 import pandas as pd
 

rag2_0/dify/dify_tool.py

@@ -6,7 +6,6 @@ import json
 from concurrent.futures import ThreadPoolExecutor, as_completed
 
 import sys
-sys.path.append(os.getcwd())
 from rag2_0.dify.dify_client import ChatClient
 from pydantic import BaseModel, Field
 from langchain.output_parsers import PydanticOutputParser
@@ -270,6 +269,51 @@ class DifyTool:
                 except (Exception, psycopg2.Error) as error:
                     raise Exception(f"Error while getting conversation_messages: {error}")
                 return None
+                
+        def execute_custom_sql(self, sql, params=None, fetch_type='all'):
+            """
+            执行自定义的SQL查询或命令。
+            
+            Args:
+                sql: 要执行的SQL语句
+                params: SQL参数（可选），用于参数化查询
+                fetch_type: 结果获取类型，可选值：'all'(返回所有行), 'one'(返回单行), 'none'(不返回结果)
+                
+            Returns:
+                根据fetch_type返回查询结果：
+                - fetch_type='all': 返回包含所有行的列表，每行是一个字典
+                - fetch_type='one': 返回单行字典，如果没有结果则返回None
+                - fetch_type='none': 返回受影响的行数
+                
+            Raises:
+                Exception: 如果执行SQL时发生错误
+            """
+            with self.pg_sql_lock:
+                try:
+                    with self.connection.cursor() as cursor:
+                        # 执行SQL语句
+                        cursor.execute(sql, params or ())
+                        
+                        # 根据fetch_type处理结果
+                        if fetch_type == 'all':
+                            result = cursor.fetchall()
+                            if result:
+                                colnames = [desc[0] for desc in cursor.description]
+                                return [dict(zip(colnames, row)) for row in result]
+                            return []
+                        elif fetch_type == 'one':
+                            result = cursor.fetchone()
+                            if result:
+                                colnames = [desc[0] for desc in cursor.description]
+                                return dict(zip(colnames, result))
+                            return None
+                        elif fetch_type == 'none':
+                            # 对于UPDATE, INSERT, DELETE等操作，返回受影响的行数
+                            return cursor.rowcount
+                        else:
+                            raise ValueError(f"不支持的fetch_type: {fetch_type}")
+                except (Exception, psycopg2.Error) as error:
+                    raise Exception(f"Error executing custom SQL: {error}")
 
     """
     提供用于获取 Dify 应用调试信息的工具类。

rag2_0/dify/export_new_dify.py

@@ -6,7 +6,6 @@ import pandas as pd
 
 
 import sys
-sys.path.append(os.getcwd())
 from rag2_0.dify.dify_tool import DifyTool
 import requests
 

rag2_0/intent_recognition/IntentRecognition.py

@@ -355,7 +355,11 @@ class AsyncIntentRecognizer:
     "dinge_info_list":{{"dinge_code_list":["xxxx","xxxx"], "dinge_name_list":["xxxx","xxxx"]}},
     "qingdan_info":{{"qingdan_code_list":["xxxx","xxxx"], "qingdan_name_list":["xxxx","xxxx"]}}
 }}"""
-
+        # 暂时跳过提取定额清单信息，本环节还未梳理清楚套用规则。
+        return {
+    "dinge_info_list":{"dinge_code_list":[], "dinge_name_list":[]},
+    "qingdan_info":{"qingdan_code_list":[], "qingdan_name_list":[]}
+}
         try:
             # response = await self._llm.ainvoke(prompt, response_format={"type": "json_object"}, extra_body={"enable_thinking": False})
             response = await self._llm.ainvoke(prompt, response_format={"type": "json_object"})

rag2_0/intent_recognition/PromptTemplates.py

@@ -100,16 +100,14 @@ query_rewrite_prompt_pro="""# 问答优化工程师
 2. 所有新增内容必须源于历史对话或聊天背景，禁止捏造。  
 3. 归一化替换需严格全词匹配：查询中的词必须与术语库同义词完全一致(不区分大小写)。部分匹配（如子字符串）或不匹配，保留原词 
 
-
 ## 核心原则
-1. **指代消除 → 当指示代词（"那"/"这"）出现时，强制继承历史对话的最新核心主题（如功能或任务），并应用到当前主体。**  
+1. **指代消除 → 当指示代词（"那"/"这"）出现时，继承历史对话的最新核心主题（如功能或任务），并应用到当前主体。**  
 2. 术语规范 → 提问中出现的同义词(synonymous)替换为标准词(name)并【】标记
 3. 语义保真 → 保持问题核心意图，允许指代消除
 
 ## 归一化替换规则
 1. 只有当问题中的词与术语库中某一项的同义词列表中的某个词完全相同时，才替换为对应的标准词
 
-
 ## 处理流程
 ### 一、输入解析
 - 原始问题(需保留核心语义):
@@ -125,14 +123,14 @@ query_rewrite_prompt_pro="""# 问答优化工程师
 {chat_history}
 </history>
 
-### 一、重构流程
+### 二、重构流程
 1、问题是否指代不明，指代不明时根据历史对话补充上下文
 2、问题是否包含同义词，包含同义词时进行同义词转标准词
 
 ### 三、重构优先级
 1. **指代消除 → 当指示代词出现时，结合历史对话补充上下文**  
 2. 同义词转标准词 → 将提问中出现的同义词(synonymous)替换为对应标准词(name) 并使用【】标记
-3. 结构优化 → 保持原问题的5W2H特征，指代消除、背景继承下允许微调意图。
+3. 结构优化 → 指代消除、背景继承下允许微调提问。
 
 ## 输出规范
 {output_format}
@@ -207,14 +205,6 @@ step_back_prompt = """# 后退提示生成器
 {output_format}
 
 ## 示例
-原始问题: "2023版本如何在Windows 11系统上导入单位工程量清单？"
-后退问题: 
-{{
-    "original_query": "2023版本如何在Windows 11系统上导入单位工程量清单？",
-    "can_use_back_prompt": true,
-    "step_back_query": ["如何在Windows 11系统上导入单位工程量清单？", "如何导入单位工程量清单？"]
-}}
-
 原始问题: "某个设备更换后，如何在系统中更新对应的定额？"
 后退问题: 
 {{

start_intent_api.sh

@@ -0,0 +1,111 @@
+#!/usr/bin/env bash
+
+# 专用脚本：启动 rag2_0.api.intent_recognition_api 服务
+# 功能：启动前检测screen是否存在，存在则结束，最后启动服务
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+SESSION_NAME="intent_recognition_api"
+SERVICE_PORT="8001"
+START_COMMAND="cd \"$SCRIPT_DIR\" && uv run uvicorn rag2_0.api.intent_recognition_api:app --host 0.0.0.0 --port 8001 --workers 4"
+
+echo "[脚本] 启动 intent_recognition_api 服务..."
+
+# 检查screen会话是否存在
+exists_session() {
+  # 使用严格匹配，避免误判
+  if screen -ls 2>/dev/null | grep -q "\\.${SESSION_NAME}\\s"; then
+    return 0
+  fi
+  return 1
+}
+
+# 按端口获取监听该端口的任意一个PID，优先用 ss，其次 lsof
+pids_on_port() {
+  # 从 ss 提取 pid 列表
+  local ss_pids
+  ss_pids=$(ss -lptn 2>/dev/null \
+    | grep -E ":${SERVICE_PORT}\\b" \
+    | awk '{print $NF}' \
+    | sed -n 's/.*pid=\([0-9]\+\),.*/\1/p' \
+    | sort -u)
+  if [[ -n "$ss_pids" ]]; then
+    echo "$ss_pids"
+    return 0
+  fi
+  # 从 lsof 提取 pid 列表
+  if command -v lsof >/dev/null 2>&1; then
+    local lsof_pids
+    lsof_pids=$(lsof -nP -i :"$SERVICE_PORT" -sTCP:LISTEN -t 2>/dev/null | sort -u)
+    if [[ -n "$lsof_pids" ]]; then
+      echo "$lsof_pids"
+      return 0
+    fi
+  fi
+  return 1
+}
+
+# 根据端口优雅终止(TERM)并在必要时强制(KILL)清理进程
+kill_by_port() {
+  local pids
+  pids=$(pids_on_port || true)
+  if [[ -z "$pids" ]]; then
+    return 0
+  fi
+  echo "[清理] 端口 $SERVICE_PORT 仍被占用，发送 SIGTERM 到: $pids"
+  kill -TERM $pids 2>/dev/null || true
+  sleep 2
+  # 再次检查
+  local left
+  left=$(pids_on_port || true)
+  if [[ -n "$left" ]]; then
+    echo "[强制] 端口 $SERVICE_PORT 仍占用，发送 SIGKILL 到: $left"
+    kill -KILL $left 2>/dev/null || true
+  fi
+}
+
+# 停止已存在的服务
+stop_existing_service() {
+  # 1) 先尝试关闭 screen 会话
+  if exists_session "$SESSION_NAME"; then
+    echo "[停止] 发现已存在的 screen 会话 '$SESSION_NAME'，正在结束..."
+    screen -S "$SESSION_NAME" -X quit || true
+    echo "[停止] screen 会话 '$SESSION_NAME' 已结束"
+  else
+    echo "[提示] 未发现 screen 会话: $SESSION_NAME"
+  fi
+
+  # 2) 等待释放端口
+  sleep 2
+
+  # 3) 如果仍占用，按端口清理
+  if ss -lptn 2>/dev/null | grep -E -q ":${SERVICE_PORT}\\b" || (command -v lsof >/dev/null 2>&1 && lsof -i :"$SERVICE_PORT" -sTCP:LISTEN >/dev/null 2>&1); then
+    echo "[清理] 端口 $SERVICE_PORT 仍被占用，正在清理..."
+    kill_by_port
+    echo "[清理] 端口 $SERVICE_PORT 清理完成"
+  fi
+}
+
+# 启动服务
+start_new_service() {
+  echo "[启动] 准备启动 intent_recognition_api 服务..."
+  echo "[启动] 启动命令: $START_COMMAND"
+  screen -dmS "$SESSION_NAME" bash -c "$START_COMMAND"
+  echo "[启动] intent_recognition_api 服务已启动，screen 会话名: '$SESSION_NAME'"
+  echo "[启动] 服务运行在端口: $SERVICE_PORT"
+  echo "[提示] 可使用 'screen -r $SESSION_NAME' 查看服务输出"
+}
+
+# 主流程
+main() {
+  # 1. 停止已存在的服务
+  stop_existing_service
+  
+  # 2. 启动新服务
+  start_new_service
+  
+  echo "[完成] intent_recognition_api 服务启动脚本执行完成"
+}
+
+main