Update app.py
Browse files
app.py
CHANGED
@@ -1,517 +1,517 @@
|
|
1 |
-
from flask import Flask, request, jsonify, send_from_directory, render_template, redirect, url_for
|
2 |
-
from flask_cors import CORS
|
3 |
-
import os
|
4 |
-
import time
|
5 |
-
import traceback
|
6 |
-
import json
|
7 |
-
import re
|
8 |
-
import sys
|
9 |
-
import io
|
10 |
-
import threading
|
11 |
-
import queue
|
12 |
-
import contextlib
|
13 |
-
import signal
|
14 |
-
import psutil
|
15 |
-
from dotenv import load_dotenv
|
16 |
-
|
17 |
-
# 导入模块路由
|
18 |
-
from modules.knowledge_base.routes import knowledge_bp
|
19 |
-
from modules.code_executor.routes import code_executor_bp
|
20 |
-
from modules.visualization.routes import visualization_bp
|
21 |
-
from modules.agent_builder.routes import agent_builder_bp
|
22 |
-
|
23 |
-
# 加载环境变量
|
24 |
-
load_dotenv()
|
25 |
-
|
26 |
-
app = Flask(__name__)
|
27 |
-
CORS(app)
|
28 |
-
|
29 |
-
# 注册蓝图
|
30 |
-
app.register_blueprint(knowledge_bp, url_prefix='/api/knowledge')
|
31 |
-
app.register_blueprint(code_executor_bp, url_prefix='/api/code')
|
32 |
-
app.register_blueprint(visualization_bp, url_prefix='/api/visualization')
|
33 |
-
app.register_blueprint(agent_builder_bp, url_prefix='/api/agent')
|
34 |
-
|
35 |
-
# 确保目录存在
|
36 |
-
os.makedirs('static', exist_ok=True)
|
37 |
-
os.makedirs('uploads', exist_ok=True)
|
38 |
-
os.makedirs('agents', exist_ok=True)
|
39 |
-
|
40 |
-
# 用于代码执行的上下文
|
41 |
-
execution_contexts = {}
|
42 |
-
|
43 |
-
def get_memory_usage():
|
44 |
-
"""获取当前进程的内存使用情况"""
|
45 |
-
process = psutil.Process(os.getpid())
|
46 |
-
return f"{process.memory_info().rss / 1024 / 1024:.1f} MB"
|
47 |
-
|
48 |
-
class CustomStdin:
|
49 |
-
def __init__(self, input_queue):
|
50 |
-
self.input_queue = input_queue
|
51 |
-
self.buffer = ""
|
52 |
-
|
53 |
-
def readline(self):
|
54 |
-
if not self.buffer:
|
55 |
-
self.buffer = self.input_queue.get() + "\n"
|
56 |
-
|
57 |
-
result = self.buffer
|
58 |
-
self.buffer = ""
|
59 |
-
return result
|
60 |
-
|
61 |
-
class InteractiveExecution:
|
62 |
-
"""管理Python代码的交互式执行"""
|
63 |
-
def __init__(self, code):
|
64 |
-
self.code = code
|
65 |
-
self.context_id = str(time.time())
|
66 |
-
self.is_complete = False
|
67 |
-
self.is_waiting_for_input = False
|
68 |
-
self.stdout_buffer = io.StringIO()
|
69 |
-
self.last_read_position = 0
|
70 |
-
self.input_queue = queue.Queue()
|
71 |
-
self.error = None
|
72 |
-
self.thread = None
|
73 |
-
self.should_terminate = False
|
74 |
-
|
75 |
-
def run(self):
|
76 |
-
"""在单独的线程中启动执行"""
|
77 |
-
self.thread = threading.Thread(target=self._execute)
|
78 |
-
self.thread.daemon = True
|
79 |
-
self.thread.start()
|
80 |
-
|
81 |
-
# 给执行一点时间开始
|
82 |
-
time.sleep(0.1)
|
83 |
-
return self.context_id
|
84 |
-
|
85 |
-
def _execute(self):
|
86 |
-
"""执行代码,处理标准输入输出"""
|
87 |
-
try:
|
88 |
-
# 保存原始的stdin/stdout
|
89 |
-
orig_stdin = sys.stdin
|
90 |
-
orig_stdout = sys.stdout
|
91 |
-
|
92 |
-
# 创建自定义stdin
|
93 |
-
custom_stdin = CustomStdin(self.input_queue)
|
94 |
-
|
95 |
-
# 重定向stdin和stdout
|
96 |
-
sys.stdin = custom_stdin
|
97 |
-
sys.stdout = self.stdout_buffer
|
98 |
-
|
99 |
-
try:
|
100 |
-
# 检查终止的函数
|
101 |
-
self._last_check_time = 0
|
102 |
-
|
103 |
-
def check_termination():
|
104 |
-
if self.should_terminate:
|
105 |
-
raise KeyboardInterrupt("Execution terminated by user")
|
106 |
-
|
107 |
-
# 设置一个模拟__main__模块的命名空间
|
108 |
-
shared_namespace = {
|
109 |
-
"__builtins__": __builtins__,
|
110 |
-
"_check_termination": check_termination,
|
111 |
-
"time": time,
|
112 |
-
"__name__": "__main__"
|
113 |
-
}
|
114 |
-
|
115 |
-
# 在这个命名空间中执行用户代码
|
116 |
-
try:
|
117 |
-
exec(self.code, shared_namespace)
|
118 |
-
except KeyboardInterrupt:
|
119 |
-
print("\nExecution terminated by user")
|
120 |
-
|
121 |
-
except Exception as e:
|
122 |
-
self.error = {
|
123 |
-
"error": str(e),
|
124 |
-
"traceback": traceback.format_exc()
|
125 |
-
}
|
126 |
-
|
127 |
-
finally:
|
128 |
-
# 恢复原始stdin/stdout
|
129 |
-
sys.stdin = orig_stdin
|
130 |
-
sys.stdout = orig_stdout
|
131 |
-
|
132 |
-
# 标记执行完成
|
133 |
-
self.is_complete = True
|
134 |
-
|
135 |
-
except Exception as e:
|
136 |
-
self.error = {
|
137 |
-
"error": str(e),
|
138 |
-
"traceback": traceback.format_exc()
|
139 |
-
}
|
140 |
-
self.is_complete = True
|
141 |
-
|
142 |
-
def terminate(self):
|
143 |
-
"""终止执行"""
|
144 |
-
self.should_terminate = True
|
145 |
-
|
146 |
-
# 如果在等待输入,放入一些内容以解除阻塞
|
147 |
-
if self.is_waiting_for_input:
|
148 |
-
self.input_queue.put("\n")
|
149 |
-
|
150 |
-
# 给执行一点时间终止
|
151 |
-
time.sleep(0.2)
|
152 |
-
|
153 |
-
# 标记为完成
|
154 |
-
self.is_complete = True
|
155 |
-
|
156 |
-
return True
|
157 |
-
|
158 |
-
def provide_input(self, user_input):
|
159 |
-
"""
|
160 |
-
self.input_queue.put(user_input)
|
161 |
-
self.is_waiting_for_input = False
|
162 |
-
return True
|
163 |
-
|
164 |
-
def get_output(self):
|
165 |
-
"""获取stdout缓冲区的当前内容"""
|
166 |
-
output = self.stdout_buffer.getvalue()
|
167 |
-
return output
|
168 |
-
|
169 |
-
def get_new_output(self):
|
170 |
-
"""只获取自上次读取以来的新输出"""
|
171 |
-
current_value = self.stdout_buffer.getvalue()
|
172 |
-
if self.last_read_position < len(current_value):
|
173 |
-
new_output = current_value[self.last_read_position:]
|
174 |
-
self.last_read_position = len(current_value)
|
175 |
-
return new_output
|
176 |
-
return ""
|
177 |
-
|
178 |
-
@app.route('/')
|
179 |
-
def index():
|
180 |
-
"""主界面"""
|
181 |
-
return render_template('index.html')
|
182 |
-
@app.route('/code_execution.html')
|
183 |
-
def index2():
|
184 |
-
"""主界面"""
|
185 |
-
return render_template('code_execution.html')
|
186 |
-
@app.route('/api/progress/<task_id>', methods=['GET'])
|
187 |
-
def get_progress(task_id):
|
188 |
-
"""获取文档处理进度"""
|
189 |
-
try:
|
190 |
-
# 从知识库模块访问处理任务
|
191 |
-
from modules.knowledge_base.routes import processing_tasks
|
192 |
-
|
193 |
-
progress_data = processing_tasks.get(task_id, {
|
194 |
-
'progress': 0,
|
195 |
-
'status': '未找到任务',
|
196 |
-
'error': True
|
197 |
-
})
|
198 |
-
|
199 |
-
return jsonify({"success": True, "data": progress_data})
|
200 |
-
except Exception as e:
|
201 |
-
traceback.print_exc()
|
202 |
-
return jsonify({"success": False, "message": str(e)}), 500
|
203 |
-
|
204 |
-
@app.route('/student/<agent_id>')
|
205 |
-
def student_view(agent_id):
|
206 |
-
"""学生访问Agent界面"""
|
207 |
-
token = request.args.get('token', '')
|
208 |
-
|
209 |
-
# 验证Agent存在
|
210 |
-
agent_path = os.path.join('agents', f"{agent_id}.json")
|
211 |
-
if not os.path.exists(agent_path):
|
212 |
-
return render_template('error.html',
|
213 |
-
message="找不到指定的Agent",
|
214 |
-
error_code=404)
|
215 |
-
|
216 |
-
# 加载Agent配置
|
217 |
-
with open(agent_path, 'r', encoding='utf-8') as f:
|
218 |
-
try:
|
219 |
-
agent_config = json.load(f)
|
220 |
-
except:
|
221 |
-
return render_template('error.html',
|
222 |
-
message="Agent配置无效",
|
223 |
-
error_code=500)
|
224 |
-
|
225 |
-
# 验证访问令牌
|
226 |
-
if token:
|
227 |
-
valid_token = False
|
228 |
-
if "distributions" in agent_config:
|
229 |
-
for dist in agent_config["distributions"]:
|
230 |
-
if dist.get("token") == token:
|
231 |
-
valid_token = True
|
232 |
-
break
|
233 |
-
|
234 |
-
if not valid_token:
|
235 |
-
return render_template('error.html',
|
236 |
-
message="访问令牌无效",
|
237 |
-
error_code=403)
|
238 |
-
|
239 |
-
# 渲染学生页面
|
240 |
-
return render_template('student.html',
|
241 |
-
agent_id=agent_id,
|
242 |
-
agent_name=agent_config.get('name', 'AI学习助手'),
|
243 |
-
agent_description=agent_config.get('description', ''),
|
244 |
-
token=token)
|
245 |
-
|
246 |
-
@app.route('/code_execution.html')
|
247 |
-
def code_execution_page():
|
248 |
-
"""代码执行页面"""
|
249 |
-
return send_from_directory(os.path.dirname(os.path.abspath(__file__)), 'code_execution.html')
|
250 |
-
|
251 |
-
@app.route('/api/student/chat/<agent_id>', methods=['POST'])
|
252 |
-
def student_chat(agent_id):
|
253 |
-
"""学生与Agent聊天的API"""
|
254 |
-
try:
|
255 |
-
data = request.json
|
256 |
-
message = data.get('message', '')
|
257 |
-
token = data.get('token', '')
|
258 |
-
|
259 |
-
if not message:
|
260 |
-
return jsonify({"success": False, "message": "消息不能为空"}), 400
|
261 |
-
|
262 |
-
# 验证Agent和令牌
|
263 |
-
agent_path = os.path.join('agents', f"{agent_id}.json")
|
264 |
-
if not os.path.exists(agent_path):
|
265 |
-
return jsonify({"success": False, "message": "Agent不存在"}), 404
|
266 |
-
|
267 |
-
with open(agent_path, 'r', encoding='utf-8') as f:
|
268 |
-
agent_config = json.load(f)
|
269 |
-
|
270 |
-
# 验证令牌(如果提供)
|
271 |
-
if token and "distributions" in agent_config:
|
272 |
-
valid_token = False
|
273 |
-
for dist in agent_config["distributions"]:
|
274 |
-
if dist.get("token") == token:
|
275 |
-
valid_token = True
|
276 |
-
|
277 |
-
# 更新使用计数
|
278 |
-
dist["usage_count"] = dist.get("usage_count", 0) + 1
|
279 |
-
break
|
280 |
-
|
281 |
-
if not valid_token:
|
282 |
-
return jsonify({"success": False, "message": "访问令牌无效"}), 403
|
283 |
-
|
284 |
-
# 更新Agent使用统计
|
285 |
-
if "stats" not in agent_config:
|
286 |
-
agent_config["stats"] = {}
|
287 |
-
|
288 |
-
agent_config["stats"]["usage_count"] = agent_config["stats"].get("usage_count", 0) + 1
|
289 |
-
agent_config["stats"]["last_used"] = int(time.time())
|
290 |
-
|
291 |
-
#
|
292 |
-
with open(agent_path, 'w', encoding='utf-8') as f:
|
293 |
-
json.dump(agent_config, f, ensure_ascii=False, indent=2)
|
294 |
-
|
295 |
-
# 获取Agent
|
296 |
-
knowledge_bases = agent_config.get('knowledge_bases', [])
|
297 |
-
plugins = agent_config.get('plugins', [])
|
298 |
-
|
299 |
-
# 获取学科和指导者信息
|
300 |
-
subject = agent_config.get('subject', agent_config.get('name', '通用学科'))
|
301 |
-
instructor = agent_config.get('instructor', '教师')
|
302 |
-
|
303 |
-
# 创建Generator实例,传入学科和指导者信息
|
304 |
-
from modules.knowledge_base.generator import Generator
|
305 |
-
generator = Generator(subject=subject, instructor=instructor)
|
306 |
-
|
307 |
-
# 检测需要使用的插件
|
308 |
-
suggested_plugins = []
|
309 |
-
|
310 |
-
# 检测是否需要代码执行插件
|
311 |
-
if 'code' in plugins and ('代码' in message or 'python' in message.lower() or '编程' in message or 'code' in message.lower() or 'program' in message.lower()):
|
312 |
-
suggested_plugins.append('code')
|
313 |
-
|
314 |
-
# 检测是否需要3D可视化插件
|
315 |
-
if 'visualization' in plugins and ('3d' in message.lower() or '可视化' in message or '图形' in message):
|
316 |
-
suggested_plugins.append('visualization')
|
317 |
-
|
318 |
-
# 检测是否需要思维导图插件
|
319 |
-
if 'mindmap' in plugins and ('思维导图' in message or 'mindmap' in message.lower()):
|
320 |
-
suggested_plugins.append('mindmap')
|
321 |
-
|
322 |
-
# 检查是否有配置知识库
|
323 |
-
if not knowledge_bases:
|
324 |
-
# 没有知识库,直接使用模型进行回答
|
325 |
-
print(f"\n=== 处理查询: {message} (无知识库) ===")
|
326 |
-
|
327 |
-
# 使用空的文档列表调用生成器进行回答
|
328 |
-
final_response = ""
|
329 |
-
for chunk in generator.generate_stream(message, []):
|
330 |
-
if isinstance(chunk, dict):
|
331 |
-
continue # 跳过处理数据
|
332 |
-
final_response += chunk
|
333 |
-
|
334 |
-
# 返回生成的回答
|
335 |
-
return jsonify({
|
336 |
-
"success": True,
|
337 |
-
"message": final_response,
|
338 |
-
"tools": suggested_plugins
|
339 |
-
})
|
340 |
-
|
341 |
-
# 有知识库配置,执行知识库查询流程
|
342 |
-
try:
|
343 |
-
# 导入RAG系统组件
|
344 |
-
from modules.knowledge_base.retriever import Retriever
|
345 |
-
from modules.knowledge_base.reranker import Reranker
|
346 |
-
|
347 |
-
retriever = Retriever()
|
348 |
-
reranker = Reranker()
|
349 |
-
|
350 |
-
# 构建工具定义 - 将所有知识库作为工具
|
351 |
-
tools = []
|
352 |
-
|
353 |
-
# 创建工具名称到索引的映射
|
354 |
-
tool_to_index = {}
|
355 |
-
|
356 |
-
for i, index in enumerate(knowledge_bases):
|
357 |
-
display_name = index[4:] if index.startswith('rag_') else index
|
358 |
-
|
359 |
-
# 判断是否是视频知识库
|
360 |
-
is_video = "视频" in display_name or "video" in display_name.lower()
|
361 |
-
|
362 |
-
# 根据内容类型生成适当的工具名称
|
363 |
-
if is_video:
|
364 |
-
tool_name = f"video_knowledge_base_{i+1}"
|
365 |
-
description = f"在'{display_name}'视频知识库中搜索,返回带时间戳的视频链接。适用于需要视频讲解的问题。"
|
366 |
-
else:
|
367 |
-
tool_name = f"knowledge_base_{i+1}"
|
368 |
-
description = f"在'{display_name}'知识库中搜索专业知识、概念和原理。适用于需要文本说明的问题。"
|
369 |
-
|
370 |
-
# 添加工具名到索引的映射
|
371 |
-
tool_to_index[tool_name] = index
|
372 |
-
|
373 |
-
tools.append({
|
374 |
-
"type": "function",
|
375 |
-
"function": {
|
376 |
-
"name": tool_name,
|
377 |
-
"description": description,
|
378 |
-
"parameters": {
|
379 |
-
"type": "object",
|
380 |
-
"properties": {
|
381 |
-
"keywords": {
|
382 |
-
"type": "array",
|
383 |
-
"items": {"type": "string"},
|
384 |
-
"description": "搜索的关键词列表"
|
385 |
-
}
|
386 |
-
},
|
387 |
-
"required": ["keywords"],
|
388 |
-
"additionalProperties": False
|
389 |
-
},
|
390 |
-
"strict": True
|
391 |
-
}
|
392 |
-
})
|
393 |
-
|
394 |
-
# 第一阶段:工具选择决策
|
395 |
-
print(f"\n=== 处理查询: {message} ===")
|
396 |
-
tool_calls = generator.extract_keywords_with_tools(message, tools)
|
397 |
-
|
398 |
-
# 如果不需要调用工具,直接回答
|
399 |
-
if not tool_calls:
|
400 |
-
print("未检测到需要使用知识库,直接回答")
|
401 |
-
final_response = ""
|
402 |
-
for chunk in generator.generate_stream(message, []):
|
403 |
-
if isinstance(chunk, dict):
|
404 |
-
continue #
|
405 |
-
final_response += chunk
|
406 |
-
|
407 |
-
return jsonify({
|
408 |
-
"success": True,
|
409 |
-
"message": final_response,
|
410 |
-
"tools": suggested_plugins
|
411 |
-
})
|
412 |
-
|
413 |
-
# 收集来自工具执行的所有文档
|
414 |
-
all_docs = []
|
415 |
-
|
416 |
-
# 执行每个工具调用
|
417 |
-
for tool_call in tool_calls:
|
418 |
-
try:
|
419 |
-
tool_name = tool_call["function"]["name"]
|
420 |
-
actual_index = tool_to_index.get(tool_name)
|
421 |
-
|
422 |
-
if not actual_index:
|
423 |
-
print(f"找不到工具名称 '{tool_name}' 对应的索引")
|
424 |
-
continue
|
425 |
-
|
426 |
-
print(f"\n执行工具 '{tool_name}' -> 使用索引 '{actual_index}'")
|
427 |
-
|
428 |
-
arguments = json.loads(tool_call["function"]["arguments"])
|
429 |
-
keywords = " ".join(arguments.get("keywords", []))
|
430 |
-
|
431 |
-
if not keywords:
|
432 |
-
print("没有提供关键词,跳过检索")
|
433 |
-
continue
|
434 |
-
|
435 |
-
print(f"检索关键词: {keywords}")
|
436 |
-
|
437 |
-
# 执行检索
|
438 |
-
retrieved_docs, _ = retriever.retrieve(keywords, specific_index=actual_index)
|
439 |
-
print(f"检索到 {len(retrieved_docs)} 个文档")
|
440 |
-
|
441 |
-
# 重排序文档
|
442 |
-
reranked_docs = reranker.rerank(message, retrieved_docs, actual_index)
|
443 |
-
print(f"重排序完成,排序后有 {len(reranked_docs)} 个文档")
|
444 |
-
|
445 |
-
# 添加结果
|
446 |
-
all_docs.extend(reranked_docs)
|
447 |
-
|
448 |
-
except Exception as e:
|
449 |
-
print(f"执行工具 '{tool_call.get('function', {}).get('name', '未知')}' 调用时出错: {str(e)}")
|
450 |
-
import traceback
|
451 |
-
traceback.print_exc()
|
452 |
-
|
453 |
-
# 如果没有检索到任何文档,直接回答
|
454 |
-
if not all_docs:
|
455 |
-
print("未检索到任何相关文档,直接回答")
|
456 |
-
final_response = ""
|
457 |
-
for chunk in generator.generate_stream(message, []):
|
458 |
-
if isinstance(chunk, dict):
|
459 |
-
continue # 跳过处理数据
|
460 |
-
final_response += chunk
|
461 |
-
|
462 |
-
return jsonify({
|
463 |
-
"success": True,
|
464 |
-
"message": final_response,
|
465 |
-
"tools": suggested_plugins
|
466 |
-
})
|
467 |
-
|
468 |
-
# 按相关性排序
|
469 |
-
all_docs.sort(key=lambda x: x.get('rerank_score', 0), reverse=True)
|
470 |
-
print(f"\n最终收集到 {len(all_docs)} 个文档用于生成回答")
|
471 |
-
|
472 |
-
# 提取参考信息
|
473 |
-
references = []
|
474 |
-
for i, doc in enumerate(all_docs[:3], 1): # 只展示前3个参考来源
|
475 |
-
file_name = doc['metadata'].get('file_name', '未知文件')
|
476 |
-
content = doc['content']
|
477 |
-
|
478 |
-
# 提取大约前100字符作为摘要
|
479 |
-
summary = content[:100] + ('...' if len(content) > 100 else '')
|
480 |
-
|
481 |
-
references.append({
|
482 |
-
'index': i,
|
483 |
-
'file_name': file_name,
|
484 |
-
'content': content,
|
485 |
-
'summary': summary
|
486 |
-
})
|
487 |
-
|
488 |
-
# 第二阶段:生成最终答案
|
489 |
-
final_response = ""
|
490 |
-
for chunk in generator.generate_stream(message, all_docs):
|
491 |
-
if isinstance(chunk, dict):
|
492 |
-
continue # 跳过处理数据
|
493 |
-
final_response += chunk
|
494 |
-
|
495 |
-
# 构建回复
|
496 |
-
return jsonify({
|
497 |
-
"success": True,
|
498 |
-
"message": final_response,
|
499 |
-
"tools": suggested_plugins,
|
500 |
-
"references": references
|
501 |
-
})
|
502 |
-
|
503 |
-
except Exception as e:
|
504 |
-
import traceback
|
505 |
-
traceback.print_exc()
|
506 |
-
return jsonify({
|
507 |
-
"success": False,
|
508 |
-
"message": f"处理查询时出错: {str(e)}"
|
509 |
-
}), 500
|
510 |
-
|
511 |
-
except Exception as e:
|
512 |
-
import traceback
|
513 |
-
traceback.print_exc()
|
514 |
-
return jsonify({"success": False, "message": str(e)}), 500
|
515 |
-
|
516 |
-
if __name__ == '__main__':
|
517 |
-
app.run(debug=True, host='0.0.0.0', port=
|
|
|
1 |
+
from flask import Flask, request, jsonify, send_from_directory, render_template, redirect, url_for
|
2 |
+
from flask_cors import CORS
|
3 |
+
import os
|
4 |
+
import time
|
5 |
+
import traceback
|
6 |
+
import json
|
7 |
+
import re
|
8 |
+
import sys
|
9 |
+
import io
|
10 |
+
import threading
|
11 |
+
import queue
|
12 |
+
import contextlib
|
13 |
+
import signal
|
14 |
+
import psutil
|
15 |
+
from dotenv import load_dotenv
|
16 |
+
|
17 |
+
# 导入模块路由
|
18 |
+
from modules.knowledge_base.routes import knowledge_bp
|
19 |
+
from modules.code_executor.routes import code_executor_bp
|
20 |
+
from modules.visualization.routes import visualization_bp
|
21 |
+
from modules.agent_builder.routes import agent_builder_bp
|
22 |
+
|
23 |
+
# 加载环境变量
|
24 |
+
load_dotenv()
|
25 |
+
|
26 |
+
app = Flask(__name__)
|
27 |
+
CORS(app)
|
28 |
+
|
29 |
+
# 注册蓝图
|
30 |
+
app.register_blueprint(knowledge_bp, url_prefix='/api/knowledge')
|
31 |
+
app.register_blueprint(code_executor_bp, url_prefix='/api/code')
|
32 |
+
app.register_blueprint(visualization_bp, url_prefix='/api/visualization')
|
33 |
+
app.register_blueprint(agent_builder_bp, url_prefix='/api/agent')
|
34 |
+
|
35 |
+
# 确保目录存在
|
36 |
+
os.makedirs('static', exist_ok=True)
|
37 |
+
os.makedirs('uploads', exist_ok=True)
|
38 |
+
os.makedirs('agents', exist_ok=True)
|
39 |
+
|
40 |
+
# 用于代码执行的上下文
|
41 |
+
execution_contexts = {}
|
42 |
+
|
43 |
+
def get_memory_usage():
|
44 |
+
"""获取当前进程的内存使用情况"""
|
45 |
+
process = psutil.Process(os.getpid())
|
46 |
+
return f"{process.memory_info().rss / 1024 / 1024:.1f} MB"
|
47 |
+
|
48 |
+
class CustomStdin:
|
49 |
+
def __init__(self, input_queue):
|
50 |
+
self.input_queue = input_queue
|
51 |
+
self.buffer = ""
|
52 |
+
|
53 |
+
def readline(self):
|
54 |
+
if not self.buffer:
|
55 |
+
self.buffer = self.input_queue.get() + "\n"
|
56 |
+
|
57 |
+
result = self.buffer
|
58 |
+
self.buffer = ""
|
59 |
+
return result
|
60 |
+
|
61 |
+
class InteractiveExecution:
|
62 |
+
"""管理Python代码的交互式执行"""
|
63 |
+
def __init__(self, code):
|
64 |
+
self.code = code
|
65 |
+
self.context_id = str(time.time())
|
66 |
+
self.is_complete = False
|
67 |
+
self.is_waiting_for_input = False
|
68 |
+
self.stdout_buffer = io.StringIO()
|
69 |
+
self.last_read_position = 0
|
70 |
+
self.input_queue = queue.Queue()
|
71 |
+
self.error = None
|
72 |
+
self.thread = None
|
73 |
+
self.should_terminate = False
|
74 |
+
|
75 |
+
def run(self):
|
76 |
+
"""在单独的线程中启动执行"""
|
77 |
+
self.thread = threading.Thread(target=self._execute)
|
78 |
+
self.thread.daemon = True
|
79 |
+
self.thread.start()
|
80 |
+
|
81 |
+
# 给执行一点时间开始
|
82 |
+
time.sleep(0.1)
|
83 |
+
return self.context_id
|
84 |
+
|
85 |
+
def _execute(self):
|
86 |
+
"""执行代码,处理标准输入输出"""
|
87 |
+
try:
|
88 |
+
# 保存原始的stdin/stdout
|
89 |
+
orig_stdin = sys.stdin
|
90 |
+
orig_stdout = sys.stdout
|
91 |
+
|
92 |
+
# 创建自定义stdin
|
93 |
+
custom_stdin = CustomStdin(self.input_queue)
|
94 |
+
|
95 |
+
# 重定向stdin和stdout
|
96 |
+
sys.stdin = custom_stdin
|
97 |
+
sys.stdout = self.stdout_buffer
|
98 |
+
|
99 |
+
try:
|
100 |
+
# 检查终止的函数
|
101 |
+
self._last_check_time = 0
|
102 |
+
|
103 |
+
def check_termination():
|
104 |
+
if self.should_terminate:
|
105 |
+
raise KeyboardInterrupt("Execution terminated by user")
|
106 |
+
|
107 |
+
# 设置一个模拟__main__模块的命名空间
|
108 |
+
shared_namespace = {
|
109 |
+
"__builtins__": __builtins__,
|
110 |
+
"_check_termination": check_termination,
|
111 |
+
"time": time,
|
112 |
+
"__name__": "__main__"
|
113 |
+
}
|
114 |
+
|
115 |
+
# 在这个命名空间中执行用户代码
|
116 |
+
try:
|
117 |
+
exec(self.code, shared_namespace)
|
118 |
+
except KeyboardInterrupt:
|
119 |
+
print("\nExecution terminated by user")
|
120 |
+
|
121 |
+
except Exception as e:
|
122 |
+
self.error = {
|
123 |
+
"error": str(e),
|
124 |
+
"traceback": traceback.format_exc()
|
125 |
+
}
|
126 |
+
|
127 |
+
finally:
|
128 |
+
# 恢复原始stdin/stdout
|
129 |
+
sys.stdin = orig_stdin
|
130 |
+
sys.stdout = orig_stdout
|
131 |
+
|
132 |
+
# 标记执行完成
|
133 |
+
self.is_complete = True
|
134 |
+
|
135 |
+
except Exception as e:
|
136 |
+
self.error = {
|
137 |
+
"error": str(e),
|
138 |
+
"traceback": traceback.format_exc()
|
139 |
+
}
|
140 |
+
self.is_complete = True
|
141 |
+
|
142 |
+
def terminate(self):
|
143 |
+
"""终止执行"""
|
144 |
+
self.should_terminate = True
|
145 |
+
|
146 |
+
# 如果在等待输入,放入一些内容以解除阻塞
|
147 |
+
if self.is_waiting_for_input:
|
148 |
+
self.input_queue.put("\n")
|
149 |
+
|
150 |
+
# 给执行一点时间终止
|
151 |
+
time.sleep(0.2)
|
152 |
+
|
153 |
+
# 标记为完成
|
154 |
+
self.is_complete = True
|
155 |
+
|
156 |
+
return True
|
157 |
+
|
158 |
+
def provide_input(self, user_input):
|
159 |
+
"""为运行的代码提供���入"""
|
160 |
+
self.input_queue.put(user_input)
|
161 |
+
self.is_waiting_for_input = False
|
162 |
+
return True
|
163 |
+
|
164 |
+
def get_output(self):
|
165 |
+
"""获取stdout缓冲区的当前内容"""
|
166 |
+
output = self.stdout_buffer.getvalue()
|
167 |
+
return output
|
168 |
+
|
169 |
+
def get_new_output(self):
|
170 |
+
"""只获取自上次读取以来的新输出"""
|
171 |
+
current_value = self.stdout_buffer.getvalue()
|
172 |
+
if self.last_read_position < len(current_value):
|
173 |
+
new_output = current_value[self.last_read_position:]
|
174 |
+
self.last_read_position = len(current_value)
|
175 |
+
return new_output
|
176 |
+
return ""
|
177 |
+
|
178 |
+
@app.route('/')
|
179 |
+
def index():
|
180 |
+
"""主界面"""
|
181 |
+
return render_template('index.html')
|
182 |
+
@app.route('/code_execution.html')
|
183 |
+
def index2():
|
184 |
+
"""主界面"""
|
185 |
+
return render_template('code_execution.html')
|
186 |
+
@app.route('/api/progress/<task_id>', methods=['GET'])
|
187 |
+
def get_progress(task_id):
|
188 |
+
"""获取文档处理进度"""
|
189 |
+
try:
|
190 |
+
# 从知识库模块访问处理任务
|
191 |
+
from modules.knowledge_base.routes import processing_tasks
|
192 |
+
|
193 |
+
progress_data = processing_tasks.get(task_id, {
|
194 |
+
'progress': 0,
|
195 |
+
'status': '未找到任务',
|
196 |
+
'error': True
|
197 |
+
})
|
198 |
+
|
199 |
+
return jsonify({"success": True, "data": progress_data})
|
200 |
+
except Exception as e:
|
201 |
+
traceback.print_exc()
|
202 |
+
return jsonify({"success": False, "message": str(e)}), 500
|
203 |
+
|
204 |
+
@app.route('/student/<agent_id>')
|
205 |
+
def student_view(agent_id):
|
206 |
+
"""学生访问Agent界面"""
|
207 |
+
token = request.args.get('token', '')
|
208 |
+
|
209 |
+
# 验证Agent存在
|
210 |
+
agent_path = os.path.join('agents', f"{agent_id}.json")
|
211 |
+
if not os.path.exists(agent_path):
|
212 |
+
return render_template('error.html',
|
213 |
+
message="找不到指定的Agent",
|
214 |
+
error_code=404)
|
215 |
+
|
216 |
+
# 加载Agent配置
|
217 |
+
with open(agent_path, 'r', encoding='utf-8') as f:
|
218 |
+
try:
|
219 |
+
agent_config = json.load(f)
|
220 |
+
except:
|
221 |
+
return render_template('error.html',
|
222 |
+
message="Agent配置无效",
|
223 |
+
error_code=500)
|
224 |
+
|
225 |
+
# 验证访问令牌
|
226 |
+
if token:
|
227 |
+
valid_token = False
|
228 |
+
if "distributions" in agent_config:
|
229 |
+
for dist in agent_config["distributions"]:
|
230 |
+
if dist.get("token") == token:
|
231 |
+
valid_token = True
|
232 |
+
break
|
233 |
+
|
234 |
+
if not valid_token:
|
235 |
+
return render_template('error.html',
|
236 |
+
message="访问令牌无效",
|
237 |
+
error_code=403)
|
238 |
+
|
239 |
+
# 渲染学生页面
|
240 |
+
return render_template('student.html',
|
241 |
+
agent_id=agent_id,
|
242 |
+
agent_name=agent_config.get('name', 'AI学习助手'),
|
243 |
+
agent_description=agent_config.get('description', ''),
|
244 |
+
token=token)
|
245 |
+
|
246 |
+
@app.route('/code_execution.html')
|
247 |
+
def code_execution_page():
|
248 |
+
"""代码执行页面"""
|
249 |
+
return send_from_directory(os.path.dirname(os.path.abspath(__file__)), 'code_execution.html')
|
250 |
+
|
251 |
+
@app.route('/api/student/chat/<agent_id>', methods=['POST'])
|
252 |
+
def student_chat(agent_id):
|
253 |
+
"""学生与Agent聊天的API"""
|
254 |
+
try:
|
255 |
+
data = request.json
|
256 |
+
message = data.get('message', '')
|
257 |
+
token = data.get('token', '')
|
258 |
+
|
259 |
+
if not message:
|
260 |
+
return jsonify({"success": False, "message": "消息不能为空"}), 400
|
261 |
+
|
262 |
+
# 验证Agent和令牌
|
263 |
+
agent_path = os.path.join('agents', f"{agent_id}.json")
|
264 |
+
if not os.path.exists(agent_path):
|
265 |
+
return jsonify({"success": False, "message": "Agent不存在"}), 404
|
266 |
+
|
267 |
+
with open(agent_path, 'r', encoding='utf-8') as f:
|
268 |
+
agent_config = json.load(f)
|
269 |
+
|
270 |
+
# 验证令牌(如果提供)
|
271 |
+
if token and "distributions" in agent_config:
|
272 |
+
valid_token = False
|
273 |
+
for dist in agent_config["distributions"]:
|
274 |
+
if dist.get("token") == token:
|
275 |
+
valid_token = True
|
276 |
+
|
277 |
+
# 更新使用计数
|
278 |
+
dist["usage_count"] = dist.get("usage_count", 0) + 1
|
279 |
+
break
|
280 |
+
|
281 |
+
if not valid_token:
|
282 |
+
return jsonify({"success": False, "message": "访问令牌无效"}), 403
|
283 |
+
|
284 |
+
# 更新Agent使用统计
|
285 |
+
if "stats" not in agent_config:
|
286 |
+
agent_config["stats"] = {}
|
287 |
+
|
288 |
+
agent_config["stats"]["usage_count"] = agent_config["stats"].get("usage_count", 0) + 1
|
289 |
+
agent_config["stats"]["last_used"] = int(time.time())
|
290 |
+
|
291 |
+
# 保存更新后的Agent配置
|
292 |
+
with open(agent_path, 'w', encoding='utf-8') as f:
|
293 |
+
json.dump(agent_config, f, ensure_ascii=False, indent=2)
|
294 |
+
|
295 |
+
# 获取Agent关联的知识库���插件
|
296 |
+
knowledge_bases = agent_config.get('knowledge_bases', [])
|
297 |
+
plugins = agent_config.get('plugins', [])
|
298 |
+
|
299 |
+
# 获取学科和指导者信息
|
300 |
+
subject = agent_config.get('subject', agent_config.get('name', '通用学科'))
|
301 |
+
instructor = agent_config.get('instructor', '教师')
|
302 |
+
|
303 |
+
# 创建Generator实例,传入学科和指导者信息
|
304 |
+
from modules.knowledge_base.generator import Generator
|
305 |
+
generator = Generator(subject=subject, instructor=instructor)
|
306 |
+
|
307 |
+
# 检测需要使用的插件
|
308 |
+
suggested_plugins = []
|
309 |
+
|
310 |
+
# 检测是否需要代码执行插件
|
311 |
+
if 'code' in plugins and ('代码' in message or 'python' in message.lower() or '编程' in message or 'code' in message.lower() or 'program' in message.lower()):
|
312 |
+
suggested_plugins.append('code')
|
313 |
+
|
314 |
+
# 检测是否需要3D可视化插件
|
315 |
+
if 'visualization' in plugins and ('3d' in message.lower() or '可视化' in message or '图形' in message):
|
316 |
+
suggested_plugins.append('visualization')
|
317 |
+
|
318 |
+
# 检测是否需要思维导图插件
|
319 |
+
if 'mindmap' in plugins and ('思维导图' in message or 'mindmap' in message.lower()):
|
320 |
+
suggested_plugins.append('mindmap')
|
321 |
+
|
322 |
+
# 检查是否有配置知识库
|
323 |
+
if not knowledge_bases:
|
324 |
+
# 没有知识库,直接使用模型进行回答
|
325 |
+
print(f"\n=== 处理查询: {message} (无知识库) ===")
|
326 |
+
|
327 |
+
# 使用空的文档列表调用生成器进行回答
|
328 |
+
final_response = ""
|
329 |
+
for chunk in generator.generate_stream(message, []):
|
330 |
+
if isinstance(chunk, dict):
|
331 |
+
continue # 跳过处理数据
|
332 |
+
final_response += chunk
|
333 |
+
|
334 |
+
# 返回生成的回答
|
335 |
+
return jsonify({
|
336 |
+
"success": True,
|
337 |
+
"message": final_response,
|
338 |
+
"tools": suggested_plugins
|
339 |
+
})
|
340 |
+
|
341 |
+
# 有知识库配置,执行知识库查询流程
|
342 |
+
try:
|
343 |
+
# 导入RAG系统组件
|
344 |
+
from modules.knowledge_base.retriever import Retriever
|
345 |
+
from modules.knowledge_base.reranker import Reranker
|
346 |
+
|
347 |
+
retriever = Retriever()
|
348 |
+
reranker = Reranker()
|
349 |
+
|
350 |
+
# 构建工具定义 - 将所有知识库作为工具
|
351 |
+
tools = []
|
352 |
+
|
353 |
+
# 创建工具名称到索引的映射
|
354 |
+
tool_to_index = {}
|
355 |
+
|
356 |
+
for i, index in enumerate(knowledge_bases):
|
357 |
+
display_name = index[4:] if index.startswith('rag_') else index
|
358 |
+
|
359 |
+
# 判断是否是视频知识库
|
360 |
+
is_video = "视频" in display_name or "video" in display_name.lower()
|
361 |
+
|
362 |
+
# 根据内容类型生成适当的工具名称
|
363 |
+
if is_video:
|
364 |
+
tool_name = f"video_knowledge_base_{i+1}"
|
365 |
+
description = f"在'{display_name}'视频知识库中搜索,返回带时间戳的视频链接。适用于需要视频讲解的问题。"
|
366 |
+
else:
|
367 |
+
tool_name = f"knowledge_base_{i+1}"
|
368 |
+
description = f"在'{display_name}'知识库中搜索专业知识、概念和原理。适用于需要文本说明的问题。"
|
369 |
+
|
370 |
+
# 添加工具名到索引的映射
|
371 |
+
tool_to_index[tool_name] = index
|
372 |
+
|
373 |
+
tools.append({
|
374 |
+
"type": "function",
|
375 |
+
"function": {
|
376 |
+
"name": tool_name,
|
377 |
+
"description": description,
|
378 |
+
"parameters": {
|
379 |
+
"type": "object",
|
380 |
+
"properties": {
|
381 |
+
"keywords": {
|
382 |
+
"type": "array",
|
383 |
+
"items": {"type": "string"},
|
384 |
+
"description": "搜索的关键词列表"
|
385 |
+
}
|
386 |
+
},
|
387 |
+
"required": ["keywords"],
|
388 |
+
"additionalProperties": False
|
389 |
+
},
|
390 |
+
"strict": True
|
391 |
+
}
|
392 |
+
})
|
393 |
+
|
394 |
+
# 第一阶段:工具选择决策
|
395 |
+
print(f"\n=== 处理查询: {message} ===")
|
396 |
+
tool_calls = generator.extract_keywords_with_tools(message, tools)
|
397 |
+
|
398 |
+
# 如果不需要调用工具,直接回答
|
399 |
+
if not tool_calls:
|
400 |
+
print("未检测到需要使用知识库,直接回答")
|
401 |
+
final_response = ""
|
402 |
+
for chunk in generator.generate_stream(message, []):
|
403 |
+
if isinstance(chunk, dict):
|
404 |
+
continue # 跳���处理数据
|
405 |
+
final_response += chunk
|
406 |
+
|
407 |
+
return jsonify({
|
408 |
+
"success": True,
|
409 |
+
"message": final_response,
|
410 |
+
"tools": suggested_plugins
|
411 |
+
})
|
412 |
+
|
413 |
+
# 收集来自工具执行的所有文档
|
414 |
+
all_docs = []
|
415 |
+
|
416 |
+
# 执行每个工具调用
|
417 |
+
for tool_call in tool_calls:
|
418 |
+
try:
|
419 |
+
tool_name = tool_call["function"]["name"]
|
420 |
+
actual_index = tool_to_index.get(tool_name)
|
421 |
+
|
422 |
+
if not actual_index:
|
423 |
+
print(f"找不到工具名称 '{tool_name}' 对应的索引")
|
424 |
+
continue
|
425 |
+
|
426 |
+
print(f"\n执行工具 '{tool_name}' -> 使用索引 '{actual_index}'")
|
427 |
+
|
428 |
+
arguments = json.loads(tool_call["function"]["arguments"])
|
429 |
+
keywords = " ".join(arguments.get("keywords", []))
|
430 |
+
|
431 |
+
if not keywords:
|
432 |
+
print("没有提供关键词,跳过检索")
|
433 |
+
continue
|
434 |
+
|
435 |
+
print(f"检索关键词: {keywords}")
|
436 |
+
|
437 |
+
# 执行检索
|
438 |
+
retrieved_docs, _ = retriever.retrieve(keywords, specific_index=actual_index)
|
439 |
+
print(f"检索到 {len(retrieved_docs)} 个文档")
|
440 |
+
|
441 |
+
# 重排序文档
|
442 |
+
reranked_docs = reranker.rerank(message, retrieved_docs, actual_index)
|
443 |
+
print(f"重排序完成,排序后有 {len(reranked_docs)} 个文档")
|
444 |
+
|
445 |
+
# 添加结果
|
446 |
+
all_docs.extend(reranked_docs)
|
447 |
+
|
448 |
+
except Exception as e:
|
449 |
+
print(f"执行工具 '{tool_call.get('function', {}).get('name', '未知')}' 调用时出错: {str(e)}")
|
450 |
+
import traceback
|
451 |
+
traceback.print_exc()
|
452 |
+
|
453 |
+
# 如果没有检索到任何文档,直接回答
|
454 |
+
if not all_docs:
|
455 |
+
print("未检索到任何相关文档,直接回答")
|
456 |
+
final_response = ""
|
457 |
+
for chunk in generator.generate_stream(message, []):
|
458 |
+
if isinstance(chunk, dict):
|
459 |
+
continue # 跳过处理数据
|
460 |
+
final_response += chunk
|
461 |
+
|
462 |
+
return jsonify({
|
463 |
+
"success": True,
|
464 |
+
"message": final_response,
|
465 |
+
"tools": suggested_plugins
|
466 |
+
})
|
467 |
+
|
468 |
+
# 按相关性排序
|
469 |
+
all_docs.sort(key=lambda x: x.get('rerank_score', 0), reverse=True)
|
470 |
+
print(f"\n最终收集到 {len(all_docs)} 个文档用于生成回答")
|
471 |
+
|
472 |
+
# 提取参考信息
|
473 |
+
references = []
|
474 |
+
for i, doc in enumerate(all_docs[:3], 1): # 只展示前3个参考来源
|
475 |
+
file_name = doc['metadata'].get('file_name', '未知文件')
|
476 |
+
content = doc['content']
|
477 |
+
|
478 |
+
# 提取大约前100字符作为摘要
|
479 |
+
summary = content[:100] + ('...' if len(content) > 100 else '')
|
480 |
+
|
481 |
+
references.append({
|
482 |
+
'index': i,
|
483 |
+
'file_name': file_name,
|
484 |
+
'content': content,
|
485 |
+
'summary': summary
|
486 |
+
})
|
487 |
+
|
488 |
+
# 第二阶段:生成最终答案
|
489 |
+
final_response = ""
|
490 |
+
for chunk in generator.generate_stream(message, all_docs):
|
491 |
+
if isinstance(chunk, dict):
|
492 |
+
continue # 跳过处理数据
|
493 |
+
final_response += chunk
|
494 |
+
|
495 |
+
# 构建回复
|
496 |
+
return jsonify({
|
497 |
+
"success": True,
|
498 |
+
"message": final_response,
|
499 |
+
"tools": suggested_plugins,
|
500 |
+
"references": references
|
501 |
+
})
|
502 |
+
|
503 |
+
except Exception as e:
|
504 |
+
import traceback
|
505 |
+
traceback.print_exc()
|
506 |
+
return jsonify({
|
507 |
+
"success": False,
|
508 |
+
"message": f"处理查询时出错: {str(e)}"
|
509 |
+
}), 500
|
510 |
+
|
511 |
+
except Exception as e:
|
512 |
+
import traceback
|
513 |
+
traceback.print_exc()
|
514 |
+
return jsonify({"success": False, "message": str(e)}), 500
|
515 |
+
|
516 |
+
if __name__ == '__main__':
|
517 |
+
app.run(debug=True, host='0.0.0.0', port=7860)
|