- 创建提示词
from langchain import hub
prompt = hub.pull("hwchase17/openai-functions-agent")
[SystemMessagePromptTemplate(prompt=PromptTemplate(input_variables=[], temp
late='You are a helpful assistant')), MessagesPlaceholder(variable_name='ch
at_history', optional=True), HumanMessagePromptTemplate(prompt=PromptTempla
te(input_variables=['input'], template='{input}')), MessagesPlaceholder(var
iable_name='agent_scratchpad')]
- 构建agent
from langchain.agents import create_tool_calling_agent
agent = create_tool_calling_agent(llm, tools, prompt)
- 将 Agent 与 AgentExecutor 中的工具结合起来
- AgentExecutor将重复调用代理并执行工具
from langchain.agents import AgentExecutor
agent_executor = AgentExecutor(agent=agent, tools=tools)
- 运行
respose = agent_executor.invoke({ "input": "你好" })
{'input': '你好', 'output': '你好！有什么我可以帮助你的吗？'}
- 创建会话记忆存储搜索函数
from langchain_community.chat_message_histories import ChatMessageHistory
from langchain_core.chat_history import BaseChatMessageHistory
from langchain_core.runnables.history import RunnableWithMessageHistory
store = {}
def get_session_history(session_id: str) -> BaseChatMessageHistory:
if session_id not in store:
store[session_id] = ChatMessageHistory()
return store[session_id]
- 使用
RunnableWithMessageHistory 构建
agent_with_history = RunnableWithMessageHistory(
agent_executor,
get_session_history,
input_messages_key="input",
history_messages_key="chat_history"
)
- 运行
response = agent_with_history.invoke(
{"input": "你好哇，我叫kingmusi"},
config={"configurable": {"session_id": "123"}}
)
print(response)
response = agent_with_history.invoke(
{"input": "我叫什么名字？"},
config={"configurable": {"session_id": "123"}}
)
print(response)
{'input': '你好哇，我叫kingmusi', 'chat_history': [], 'output': '你好，Kingmusi！很高兴认识你！有什么我可以帮你的吗？😊'}
{'input': '我叫什么名字？', 'chat_history': [HumanMessage(content='你好哇，我叫kingmusi', additional_kwargs={}, response_metadata={}), AIMessage(content='你好，Kingmusi！很高兴认识你！有什么我可以帮你的吗？😊', additional_kwargs={}, response_metadata={})], 'output': '你刚刚告诉我你的名字是 **Kingmusi**！没错吧？😄'}
