557 lines
21 KiB
Python
557 lines
21 KiB
Python
#!/usr/bin/env python3
|
||
# -*- coding: utf-8 -*-
|
||
"""
|
||
稷下学宫完整版 - 基于OpenAI Swarm的八仙辩论系统
|
||
实现完整的八仙论道 + 三清决策
|
||
"""
|
||
|
||
import os
|
||
import asyncio
|
||
import json
|
||
import subprocess
|
||
from datetime import datetime
|
||
from swarm import Swarm, Agent
|
||
from typing import Dict, List, Any, Optional
|
||
import random
|
||
|
||
class JixiaSwarmAcademy:
|
||
"""稷下学宫 - 完整的八仙辩论系统"""
|
||
|
||
def __init__(self):
|
||
# 从Doppler获取API密钥
|
||
self.api_key = self.get_secure_api_key()
|
||
|
||
# 设置环境变量
|
||
if self.api_key:
|
||
os.environ["OPENAI_API_KEY"] = self.api_key
|
||
os.environ["OPENAI_BASE_URL"] = "https://openrouter.ai/api/v1"
|
||
|
||
# 初始化Swarm客户端，传入配置
|
||
from openai import OpenAI
|
||
openai_client = OpenAI(
|
||
api_key=self.api_key,
|
||
base_url="https://openrouter.ai/api/v1",
|
||
default_headers={
|
||
"HTTP-Referer": "https://github.com/ben/cauldron",
|
||
"X-Title": "Jixia Academy Debate System" # 避免中文字符
|
||
}
|
||
)
|
||
self.client = Swarm(client=openai_client)
|
||
else:
|
||
print("❌ 无法获取有效的API密钥")
|
||
self.client = None
|
||
|
||
# 八仙配置 - 完整版
|
||
self.immortals_config = {
|
||
'吕洞宾': {
|
||
'role': '剑仙投资顾问',
|
||
'gua_position': '乾☰',
|
||
'specialty': '技术分析',
|
||
'stance': 'positive',
|
||
'style': '一剑封喉，直指要害',
|
||
'personality': '犀利直接，善于识破市场迷雾',
|
||
'weapon': '纯阳剑',
|
||
'next': '何仙姑'
|
||
},
|
||
'何仙姑': {
|
||
'role': '慈悲风控专家',
|
||
'gua_position': '坤☷',
|
||
'specialty': '风险控制',
|
||
'stance': 'negative',
|
||
'style': '荷花在手，全局在胸',
|
||
'personality': '温和坚定，关注风险控制',
|
||
'weapon': '荷花',
|
||
'next': '张果老'
|
||
},
|
||
'张果老': {
|
||
'role': '历史数据分析师',
|
||
'gua_position': '艮☶',
|
||
'specialty': '历史回测',
|
||
'stance': 'positive',
|
||
'style': '倒骑毛驴，逆向思维',
|
||
'personality': '博古通今，从历史中寻找规律',
|
||
'weapon': '鱼鼓',
|
||
'next': '韩湘子'
|
||
},
|
||
'韩湘子': {
|
||
'role': '市场情绪分析师',
|
||
'gua_position': '兑☱',
|
||
'specialty': '情绪分析',
|
||
'stance': 'negative',
|
||
'style': '笛声悠扬，感知人心',
|
||
'personality': '敏感细腻，善于捕捉市场情绪',
|
||
'weapon': '洞箫',
|
||
'next': '汉钟离'
|
||
},
|
||
'汉钟离': {
|
||
'role': '宏观经济分析师',
|
||
'gua_position': '离☲',
|
||
'specialty': '宏观分析',
|
||
'stance': 'positive',
|
||
'style': '扇子一挥，大局明了',
|
||
'personality': '气度恢宏，关注宏观大势',
|
||
'weapon': '芭蕉扇',
|
||
'next': '蓝采和'
|
||
},
|
||
'蓝采和': {
|
||
'role': '量化交易专家',
|
||
'gua_position': '巽☴',
|
||
'specialty': '量化模型',
|
||
'stance': 'negative',
|
||
'style': '花篮一抛，数据飞舞',
|
||
'personality': '逻辑严密，依赖数学模型',
|
||
'weapon': '花篮',
|
||
'next': '曹国舅'
|
||
},
|
||
'曹国舅': {
|
||
'role': '价值投资专家',
|
||
'gua_position': '坎☵',
|
||
'specialty': '基本面分析',
|
||
'stance': 'positive',
|
||
'style': '玉板一敲，价值显现',
|
||
'personality': '稳重踏实，注重内在价值',
|
||
'weapon': '玉板',
|
||
'next': '铁拐李'
|
||
},
|
||
'铁拐李': {
|
||
'role': '逆向投资大师',
|
||
'gua_position': '震☳',
|
||
'specialty': '逆向投资',
|
||
'stance': 'negative',
|
||
'style': '铁拐一点，危机毕现',
|
||
'personality': '不拘一格，挑战主流观点',
|
||
'weapon': '铁拐杖',
|
||
'next': 'summary'
|
||
}
|
||
}
|
||
|
||
# 三清决策层配置
|
||
self.sanqing_config = {
|
||
'元始天尊': {
|
||
'role': '最终决策者',
|
||
'specialty': '综合决策',
|
||
'style': '无极生太极，一言定乾坤'
|
||
},
|
||
'灵宝天尊': {
|
||
'role': '风险评估师',
|
||
'specialty': '风险量化',
|
||
'style': '太极生两仪，阴阳定风险'
|
||
},
|
||
'道德天尊': {
|
||
'role': '合规审查员',
|
||
'specialty': '合规检查',
|
||
'style': '两仪生四象，四象定规矩'
|
||
}
|
||
}
|
||
|
||
# 创建智能体
|
||
self.immortal_agents = self.create_immortal_agents()
|
||
self.sanqing_agents = self.create_sanqing_agents()
|
||
|
||
# 辩论历史
|
||
self.debate_history = []
|
||
self.current_round = 0
|
||
self.max_rounds = 2 # 每个仙人最多发言2轮
|
||
|
||
def get_secure_api_key(self):
|
||
"""获取API密钥 - 支持多种方式"""
|
||
# 从环境变量获取API密钥
|
||
available_keys = [
|
||
os.getenv("OPENROUTER_API_KEY_1"),
|
||
os.getenv("OPENROUTER_API_KEY_2"),
|
||
os.getenv("OPENROUTER_API_KEY_3"),
|
||
os.getenv("OPENROUTER_API_KEY_4")
|
||
]
|
||
# 过滤掉None值
|
||
available_keys = [key for key in available_keys if key]
|
||
|
||
# 直接使用第一个密钥进行测试
|
||
test_key = available_keys[0]
|
||
print(f"🔑 直接使用测试密钥: {test_key[:20]}...")
|
||
return test_key
|
||
|
||
|
||
|
||
def create_immortal_agents(self) -> Dict[str, Agent]:
|
||
"""创建八仙智能体"""
|
||
agents = {}
|
||
|
||
for name, config in self.immortals_config.items():
|
||
# 创建转换函数 - 使用英文名称避免特殊字符问题
|
||
next_immortal = config['next']
|
||
if next_immortal == 'summary':
|
||
transfer_func = self.transfer_to_sanqing
|
||
else:
|
||
# 创建一个简单的转换函数，避免lambda的问题
|
||
def create_transfer_func(next_name):
|
||
def transfer():
|
||
return self.transfer_to_immortal(next_name)
|
||
transfer.__name__ = f"transfer_to_{self.get_english_name(next_name)}"
|
||
return transfer
|
||
transfer_func = create_transfer_func(next_immortal)
|
||
|
||
# 构建详细的指令
|
||
instructions = self.build_immortal_instructions(name, config)
|
||
|
||
agents[name] = Agent(
|
||
name=name,
|
||
instructions=instructions,
|
||
functions=[transfer_func]
|
||
)
|
||
|
||
return agents
|
||
|
||
def create_sanqing_agents(self) -> Dict[str, Agent]:
|
||
"""创建三清决策层智能体"""
|
||
agents = {}
|
||
|
||
# 元始天尊 - 最终决策者
|
||
agents['元始天尊'] = Agent(
|
||
name="元始天尊",
|
||
instructions="""
|
||
你是元始天尊，道教三清之首，稷下学宫的最终决策者。
|
||
|
||
你的使命：
|
||
1. 综合八仙的所有观点，做出最终投资决策
|
||
2. 平衡正反两方的观点，寻找最优解
|
||
3. 给出具体的投资建议和操作指导
|
||
4. 评估决策的风险等级和预期收益
|
||
|
||
你的风格：
|
||
- 高屋建瓴，统揽全局
|
||
- 言简意赅，一锤定音
|
||
- 既不偏向乐观，也不偏向悲观
|
||
- 以数据和逻辑为准绳
|
||
|
||
请以"元始天尊曰"开头，给出最终决策。
|
||
决策格式：
|
||
- 投资建议：买入/持有/卖出
|
||
- 风险等级：低/中/高
|
||
- 预期收益：具体百分比
|
||
- 操作建议：具体的操作指导
|
||
- 决策依据：主要的决策理由
|
||
""",
|
||
functions=[]
|
||
)
|
||
|
||
return agents
|
||
|
||
def build_immortal_instructions(self, name: str, config: Dict) -> str:
|
||
"""构建仙人的详细指令"""
|
||
stance_desc = "看涨派，倾向于发现投资机会" if config['stance'] == 'positive' else "看跌派，倾向于发现投资风险"
|
||
|
||
instructions = f"""
|
||
你是{name}，八仙之一，{config['role']}。
|
||
|
||
你的身份特征：
|
||
- 位居{config['gua_position']}之位，代表{self.get_gua_meaning(config['gua_position'])}
|
||
- 持有{config['weapon']}，{config['style']}
|
||
- 擅长{config['specialty']}，{config['personality']}
|
||
- 立场倾向：{stance_desc}
|
||
|
||
在稷下学宫辩论中，你要：
|
||
|
||
1. **专业分析**：从{config['specialty']}角度深入分析
|
||
2. **立场鲜明**：作为{stance_desc}，要有明确的观点
|
||
3. **数据支撑**：用具体的数据、图表、历史案例支撑观点
|
||
4. **互动辩论**：可以质疑前面仙人的观点，但要有理有据
|
||
5. **仙风道骨**：保持古雅的表达风格，但不影响专业性
|
||
6. **承上启下**：总结前面的观点，为后面的仙人铺垫
|
||
|
||
发言格式：
|
||
- 以"{name}曰："开头
|
||
- 先简要回应前面仙人的观点（如果有）
|
||
- 然后从你的专业角度进行分析
|
||
- 最后明确表达你的投资倾向
|
||
- 结尾时说"请{config['next']}仙长继续论道"（如果不是最后一个）
|
||
|
||
记住：你是{stance_desc}，要体现这个立场，但也要保持专业和客观。
|
||
"""
|
||
|
||
return instructions
|
||
|
||
def get_gua_meaning(self, gua: str) -> str:
|
||
"""获取卦象含义"""
|
||
meanings = {
|
||
'乾☰': '天行健，自强不息',
|
||
'坤☷': '地势坤，厚德载物',
|
||
'艮☶': '艮为山，止于至善',
|
||
'兑☱': '兑为泽，和悦致祥',
|
||
'离☲': '离为火，光明磊落',
|
||
'巽☴': '巽为风，随风而化',
|
||
'坎☵': '坎为水，智慧如水',
|
||
'震☳': '震为雷，威震四方'
|
||
}
|
||
return meanings.get(gua, '神秘莫测')
|
||
|
||
def transfer_to_hexiangu(self):
|
||
"""转到何仙姑"""
|
||
return self.immortal_agents.get('何仙姑')
|
||
|
||
def transfer_to_zhangguolao(self):
|
||
"""转到张果老"""
|
||
return self.immortal_agents.get('张果老')
|
||
|
||
def transfer_to_hanxiangzi(self):
|
||
"""转到韩湘子"""
|
||
return self.immortal_agents.get('韩湘子')
|
||
|
||
def transfer_to_hanzhongli(self):
|
||
"""转到汉钟离"""
|
||
return self.immortal_agents.get('汉钟离')
|
||
|
||
def transfer_to_lancaihe(self):
|
||
"""转到蓝采和"""
|
||
return self.immortal_agents.get('蓝采和')
|
||
|
||
def transfer_to_caoguojiu(self):
|
||
"""转到曹国舅"""
|
||
return self.immortal_agents.get('曹国舅')
|
||
|
||
def transfer_to_tieguaili(self):
|
||
"""转到铁拐李"""
|
||
return self.immortal_agents.get('铁拐李')
|
||
|
||
def transfer_to_sanqing(self):
|
||
"""转到三清决策层"""
|
||
return self.sanqing_agents['元始天尊']
|
||
|
||
async def conduct_full_debate(self, topic: str, context: Dict[str, Any] = None) -> Dict[str, Any]:
|
||
"""进行完整的稷下学宫辩论"""
|
||
if not self.api_key or not self.client:
|
||
print("❌ 无法获取API密钥或初始化客户端，无法进行论道")
|
||
return None
|
||
|
||
print("🏛️ 稷下学宫八仙论道正式开始！")
|
||
print("=" * 80)
|
||
print(f"🎯 论道主题: {topic}")
|
||
print(f"⏰ 开始时间: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
|
||
print()
|
||
|
||
# 构建初始提示
|
||
initial_prompt = self.build_debate_prompt(topic, context)
|
||
|
||
try:
|
||
# 从吕洞宾开始论道
|
||
print("⚔️ 吕洞宾仙长请先发言...")
|
||
print("-" * 60)
|
||
|
||
response = self.client.run(
|
||
agent=self.immortal_agents['吕洞宾'],
|
||
messages=[{"role": "user", "content": initial_prompt}],
|
||
max_turns=20 # 允许多轮对话
|
||
)
|
||
|
||
print("\n" + "=" * 80)
|
||
print("🎊 稷下学宫八仙论道圆满结束！")
|
||
print("📊 三清决策已生成")
|
||
|
||
# 处理辩论结果
|
||
debate_result = self.process_debate_result(response, topic, context)
|
||
|
||
# 显示辩论总结
|
||
self.display_debate_summary(debate_result)
|
||
|
||
return debate_result
|
||
|
||
except Exception as e:
|
||
print(f"❌ 论道过程中出错: {e}")
|
||
import traceback
|
||
traceback.print_exc()
|
||
return None
|
||
|
||
def build_debate_prompt(self, topic: str, context: Dict[str, Any] = None) -> str:
|
||
"""构建辩论提示"""
|
||
context_str = ""
|
||
if context:
|
||
context_str = f"\n📊 市场背景:\n{json.dumps(context, indent=2, ensure_ascii=False)}\n"
|
||
|
||
# 随机选择一些市场数据作为背景
|
||
market_context = self.generate_market_context(topic)
|
||
|
||
prompt = f"""
|
||
🏛️ 稷下学宫八仙论道正式开始！
|
||
|
||
📜 论道主题: {topic}
|
||
{context_str}
|
||
📈 当前市场环境:
|
||
{market_context}
|
||
|
||
🎭 论道规则:
|
||
1. 八仙按序发言：吕洞宾 → 何仙姑 → 张果老 → 韩湘子 → 汉钟离 → 蓝采和 → 曹国舅 → 铁拐李
|
||
2. 正反方交替：正方(看涨) vs 反方(看跌)
|
||
3. 每位仙人从专业角度分析，必须提供数据支撑
|
||
4. 可以质疑前面仙人的观点，但要有理有据
|
||
5. 保持仙风道骨的表达风格
|
||
6. 最后由三清做出最终决策
|
||
|
||
🗡️ 请吕洞宾仙长首先发言，展现剑仙的犀利分析！
|
||
记住：你是看涨派，要从技术分析角度找到投资机会！
|
||
"""
|
||
return prompt
|
||
|
||
def generate_market_context(self, topic: str) -> str:
|
||
"""生成模拟的市场背景数据"""
|
||
# 这里可以集成真实的市场数据，现在先用模拟数据
|
||
contexts = {
|
||
"英伟达": "NVDA当前价格$120，P/E比率65，市值$3T，AI芯片需求旺盛",
|
||
"比特币": "BTC当前价格$43,000，24h涨幅+2.3%，机构持续买入",
|
||
"美联储": "联邦基金利率5.25%，通胀率3.2%，就业数据强劲",
|
||
"中国股市": "上证指数3100点，外资流入放缓，政策支持预期"
|
||
}
|
||
|
||
# 根据主题选择相关背景
|
||
for key, context in contexts.items():
|
||
if key in topic:
|
||
return context
|
||
|
||
return "市场情绪谨慎，波动率上升，投资者观望情绪浓厚"
|
||
|
||
def process_debate_result(self, response, topic: str, context: Dict[str, Any]) -> Dict[str, Any]:
|
||
"""处理辩论结果"""
|
||
# 提取所有消息
|
||
all_messages = response.messages if hasattr(response, 'messages') else []
|
||
|
||
# 分析发言者和内容
|
||
debate_messages = []
|
||
speakers = []
|
||
|
||
for msg in all_messages:
|
||
if msg.get('role') == 'assistant' and msg.get('content'):
|
||
content = msg['content']
|
||
speaker = self.extract_speaker_from_content(content)
|
||
|
||
debate_messages.append({
|
||
'speaker': speaker,
|
||
'content': content,
|
||
'timestamp': datetime.now().isoformat(),
|
||
'stance': self.get_speaker_stance(speaker)
|
||
})
|
||
|
||
if speaker not in speakers:
|
||
speakers.append(speaker)
|
||
|
||
# 提取最终决策（通常是最后一条消息）
|
||
final_decision = ""
|
||
if debate_messages:
|
||
final_decision = debate_messages[-1]['content']
|
||
|
||
# 构建结果
|
||
result = {
|
||
"debate_id": f"jixia_debate_{datetime.now().strftime('%Y%m%d_%H%M%S')}",
|
||
"topic": topic,
|
||
"context": context,
|
||
"participants": speakers,
|
||
"messages": debate_messages,
|
||
"final_decision": final_decision,
|
||
"summary": self.generate_debate_summary(debate_messages),
|
||
"timestamp": datetime.now().isoformat(),
|
||
"framework": "OpenAI Swarm",
|
||
"academy": "稷下学宫"
|
||
}
|
||
|
||
self.debate_history.append(result)
|
||
return result
|
||
|
||
def extract_speaker_from_content(self, content: str) -> str:
|
||
"""从内容中提取发言者"""
|
||
for name in list(self.immortals_config.keys()) + list(self.sanqing_config.keys()):
|
||
if f"{name}曰" in content or name in content[:20]:
|
||
return name
|
||
return "未知仙人"
|
||
|
||
def get_speaker_stance(self, speaker: str) -> str:
|
||
"""获取发言者立场"""
|
||
if speaker in self.immortals_config:
|
||
return self.immortals_config[speaker]['stance']
|
||
elif speaker in self.sanqing_config:
|
||
return 'neutral'
|
||
return 'unknown'
|
||
|
||
def generate_debate_summary(self, messages: List[Dict]) -> str:
|
||
"""生成辩论摘要"""
|
||
positive_count = len([m for m in messages if m.get('stance') == 'positive'])
|
||
negative_count = len([m for m in messages if m.get('stance') == 'negative'])
|
||
|
||
summary = f"""
|
||
📊 辩论统计:
|
||
- 参与仙人: {len(set(m['speaker'] for m in messages))}位
|
||
- 看涨观点: {positive_count}条
|
||
- 看跌观点: {negative_count}条
|
||
- 总发言数: {len(messages)}条
|
||
|
||
🎯 观点倾向: {'偏向看涨' if positive_count > negative_count else '偏向看跌' if negative_count > positive_count else '观点平衡'}
|
||
"""
|
||
|
||
return summary
|
||
|
||
def display_debate_summary(self, result: Dict[str, Any]):
|
||
"""显示辩论总结"""
|
||
print("\n🌟 稷下学宫辩论总结")
|
||
print("=" * 80)
|
||
print(f"📜 主题: {result['topic']}")
|
||
print(f"🎭 参与仙人: {', '.join(result['participants'])}")
|
||
print(f"⏰ 辩论时间: {result['timestamp']}")
|
||
print(f"🔧 技术框架: {result['framework']}")
|
||
|
||
print(result['summary'])
|
||
|
||
print("\n🏆 最终决策:")
|
||
print("-" * 40)
|
||
print(result['final_decision'])
|
||
|
||
print("\n✨ 稷下学宫辩论特色:")
|
||
print("🗡️ 八仙各展所长，观点多元化")
|
||
print("⚖️ 正反方交替发言，辩论更激烈")
|
||
print("🧠 三清最终决策，权威性更强")
|
||
print("🔄 基于Swarm框架，性能更优越")
|
||
|
||
# 主函数和测试
|
||
async def main():
|
||
"""主函数 - 演示完整的稷下学宫辩论"""
|
||
print("🏛️ 稷下学宫 - OpenAI Swarm完整版")
|
||
print("🔐 使用Doppler安全管理API密钥")
|
||
print("🚀 八仙论道 + 三清决策的完整体验")
|
||
print()
|
||
|
||
# 创建学宫
|
||
academy = JixiaSwarmAcademy()
|
||
|
||
if not academy.api_key:
|
||
print("❌ 无法获取API密钥，请检查Doppler配置或环境变量")
|
||
return
|
||
|
||
# 辩论主题列表
|
||
topics = [
|
||
"英伟达股价走势：AI泡沫还是技术革命？",
|
||
"美联储2024年货币政策：加息还是降息？",
|
||
"比特币vs黄金：谁是更好的避险资产？",
|
||
"中国房地产市场：触底反弹还是继续下行？",
|
||
"特斯拉股价：马斯克效应还是基本面支撑？"
|
||
]
|
||
|
||
# 随机选择主题
|
||
topic = random.choice(topics)
|
||
|
||
# 构建市场背景
|
||
context = {
|
||
"market_sentiment": "谨慎乐观",
|
||
"volatility": "中等",
|
||
"major_events": ["美联储会议", "财报季", "地缘政治紧张"],
|
||
"technical_indicators": {
|
||
"RSI": 65,
|
||
"MACD": "金叉",
|
||
"MA20": "上穿"
|
||
}
|
||
}
|
||
|
||
# 开始辩论
|
||
result = await academy.conduct_full_debate(topic, context)
|
||
|
||
if result:
|
||
print(f"\n🎉 辩论成功完成！辩论ID: {result['debate_id']}")
|
||
else:
|
||
print("❌ 辩论失败")
|
||
|
||
if __name__ == "__main__":
|
||
asyncio.run(main()) |