149 lines
4.1 KiB
Python
149 lines
4.1 KiB
Python
"""
|
||
LangGraph 实战 - 接入 OpenAI LLM 的智能体
|
||
构建一个能"思考-行动-反思"的真实 AI 智能体
|
||
"""
|
||
import os
|
||
from langgraph.graph import StateGraph, START, END
|
||
from typing import TypedDict, Annotated
|
||
from functools import reduce
|
||
|
||
# 检查 API Key
|
||
OPENAI_KEY = os.environ.get("OPENAI_API_KEY", "")
|
||
if not OPENAI_KEY:
|
||
print("请先设置 OPENAI_API_KEY:")
|
||
print(" $env:OPENAI_API_KEY = 'sk-...' # PowerShell")
|
||
print()
|
||
print("暂时使用模拟模式演示...")
|
||
USE_MOCK = True
|
||
else:
|
||
USE_MOCK = False
|
||
|
||
# 1️⃣ 定义状态
|
||
class AgentState(TypedDict):
|
||
query: str
|
||
thought: str
|
||
action: str
|
||
observation: str
|
||
answer: str
|
||
max_iterations: int
|
||
iteration: int
|
||
|
||
# 2️⃣ 模拟 LLM 调用
|
||
def mock_llm(prompt: str) -> str:
|
||
"""模拟 LLM 响应"""
|
||
responses = {
|
||
"思考": "我需要先分析问题，然后给出解答。",
|
||
"行动": "搜索相关信息并整理答案。",
|
||
"回答": "根据分析，LangGraph 是一个用于构建有状态 AI 应用的框架。",
|
||
}
|
||
for key, value in responses.items():
|
||
if key in prompt:
|
||
return value
|
||
return "这是一个有趣的问题。"
|
||
|
||
def real_llm(prompt: str) -> str:
|
||
"""真实 LLM 调用"""
|
||
from openai import OpenAI
|
||
client = OpenAI()
|
||
response = client.chat.completions.create(
|
||
model="gpt-4o-mini",
|
||
messages=[{"role": "user", "content": prompt}],
|
||
max_tokens=200,
|
||
temperature=0.7,
|
||
)
|
||
return response.choices[0].message.content
|
||
|
||
# 3️⃣ 定义节点
|
||
def think_node(state: AgentState):
|
||
"""思考节点 - 分析当前情况"""
|
||
state = state.copy()
|
||
state['iteration'] += 1
|
||
print(f"\n{'='*40}")
|
||
print(f"[思考] 第 {state['iteration']}/{state['max_iterations']} 轮")
|
||
|
||
if USE_MOCK:
|
||
state['thought'] = mock_llm(f"思考: {state['query']}")
|
||
else:
|
||
state['thought'] = real_llm(f"思考: {state['query']}")
|
||
|
||
print(f" 想法: {state['thought']}")
|
||
return state
|
||
|
||
def act_node(state: AgentState):
|
||
"""行动节点 - 执行操作"""
|
||
state = state.copy()
|
||
print(f"[行动] 执行操作...")
|
||
|
||
if USE_MOCK:
|
||
state['action'] = mock_llm("行动")
|
||
state['observation'] = f"观察到: {state['action']}的结果"
|
||
else:
|
||
state['action'] = real_llm(f"针对问题 '{state['query']}' 我该采取什么行动？")
|
||
state['observation'] = f"执行了: {state['action']}"
|
||
|
||
print(f" 行动: {state['action']}")
|
||
print(f" 观察: {state['observation']}")
|
||
return state
|
||
|
||
def answer_node(state: AgentState):
|
||
"""回答节点 - 给出最终答案"""
|
||
state = state.copy()
|
||
print(f"\n[回答] 生成最终答案...")
|
||
|
||
if USE_MOCK:
|
||
state['answer'] = mock_llm("回答")
|
||
else:
|
||
prompt = f"""基于以下信息回答问题:
|
||
问题: {state['query']}
|
||
思考过程: {state['thought']}
|
||
行动: {state['action']}
|
||
观察: {state['observation']}
|
||
请给出简洁的答案:"""
|
||
state['answer'] = real_llm(prompt)
|
||
|
||
print(f" 答案: {state['answer']}")
|
||
return state
|
||
|
||
# 4️⃣ 路由函数
|
||
def should_continue(state: AgentState):
|
||
"""决定是否继续迭代"""
|
||
if state['iteration'] >= state['max_iterations']:
|
||
return "answer"
|
||
observation = state.get('observation', '')
|
||
if observation and len(observation) > 10:
|
||
return "answer"
|
||
return "act"
|
||
|
||
# 5️⃣ 构建图
|
||
graph = StateGraph(AgentState)
|
||
graph.add_node("think", think_node)
|
||
graph.add_node("act", act_node)
|
||
graph.add_node("answer", answer_node)
|
||
|
||
# 连接
|
||
graph.add_edge(START, "think")
|
||
graph.add_conditional_edges("think", should_continue, {"act": "act", "answer": "answer"})
|
||
graph.add_edge("act", "answer")
|
||
graph.add_edge("answer", END)
|
||
|
||
# 编译
|
||
app = graph.compile()
|
||
|
||
# 6️⃣ 运行
|
||
print("=" * 40)
|
||
print("LangGraph AI 智能体演示")
|
||
print("=" * 40)
|
||
|
||
result = app.invoke({
|
||
"query": "什么是 LangGraph？",
|
||
"thought": "",
|
||
"action": "",
|
||
"observation": "",
|
||
"answer": "",
|
||
"max_iterations": 2,
|
||
"iteration": 0,
|
||
})
|
||
|
||
print(f"\n{'='*40}")
|
||
print(f"最终答案: {result['answer']}")
|
||
print(f"{'='*40}") |