memory.py

from langchain.agents import create_agent
from langchain_community.chat_models.tongyi import ChatTongyi
import os
from langchain.tools import tool
from langchain.agents.middleware import SummarizationMiddleware
from langchain_core.runnables import RunnableConfig
from langgraph.checkpoint.memory import InMemorySaver  # Checkpoint


# 工具函数，用于获取天气
@tool
def weather_tool(city: str) -> str:
    """Get the weather for a city."""
    return f"it's sunny and 70 degrees in {city}"

# 使用 InMemorySaver，也就是在内存中来保存对话状态。如果你希望持久化状态，可以考虑使用数据库或文件存储。
checkpointer = InMemorySaver()  

# 创建聊天模型
llm = ChatTongyi(model="qwen-plus", api_key=os.environ["DASHSCOPE_API_KEY"])

# 创建代理，使用state_schema
agent = create_agent(
    model=llm, 
    tools=[weather_tool], 
    system_prompt="You are a helpful assistant",
    checkpointer=checkpointer,  # 使用自定义的 checkpoint
    middleware=[SummarizationMiddleware(
        model=llm,
        trigger=("tokens", 4000),  # 设置触发条件,当对话内容的 token 数达到 4000 时触发摘要
        keep=("messages", 20)      # 保留最后的20条消息
    )]
)
config: RunnableConfig = {"configurable": {"thread_id": "1"}}
# 多轮对话示例，每一次invoke都会更新checkpointer状态
state = agent.invoke({"messages": [{"role": "user", "content": "我的名字是韩"}]}, config) 
state = agent.invoke({"messages": [{"role": "user", "content": "北京天气怎么样"}]}, config)
state = agent.invoke({"messages": [{"role": "user", "content": "南京呢"}]}, config)
state = agent.invoke({"messages": [{"role": "user", "content": "什么是计算机网络"}]}, config)
state = agent.invoke({"messages": [{"role": "user", "content": "推荐几本计算机网络的课程"}]}, config)
state = agent.invoke({"messages": [{"role": "user", "content": "我应该如何学习计算机网络"}]}, config)
# 查看最后的总结结果
print(state["messages"]) 
# from langchain.agents import create_agent
# from langchain.agents.middleware import SummarizationMiddleware
# from langgraph.checkpoint.memory import InMemorySaver
# from langchain_core.runnables import RunnableConfig
# from langchain_community.chat_models.tongyi import ChatTongyi
# import os

# checkpointer = InMemorySaver()
# llm = ChatTongyi(model="qwen-plus", api_key=os.environ["DASHSCOPE_API_KEY"])
# agent = create_agent(
#     model=llm,
#     tools=[],
#     middleware=[
#         SummarizationMiddleware(
#             model=llm,
#             trigger=("tokens", 4000),
#             keep=("messages", 20)
#         )
#     ],
#     checkpointer=checkpointer,
# )

# config: RunnableConfig = {"configurable": {"thread_id": "1"}}
# agent.invoke({"messages": "hi, my name is bob"}, config)
# agent.invoke({"messages": "write a short poem about cats"}, config)
# agent.invoke({"messages": "now do the same but for dogs"}, config)
# agent.invoke({"messages": "I like dogs and cats"}, config)
# final_response = agent.invoke({"messages": "what's my name?"}, config)
# print(f"fianal response:{final_response}")

# final_response["messages"][-1].pretty_print()