from langchain_community.document_loaders import TextLoaderfrom langchain_community.vectorstores.faiss import FAISSfrom langchain_core.tools import toolfrom langchain_openai import ChatOpenAI, OpenAIEmbeddingsfrom langchain.agents import AgentExecutor, create_openai_tools_agentfrom langchain import hubfrom plantuml import PlantUML
# 声明模型llm = ChatOpenAI()# # 1. 读取文件。loader = TextLoader("./需求文档.md")data = loader.load()# 3. embeddingembeddings = OpenAIEmbeddings()# # 4. 向量存储vector = FAISS.from_documents(data, embeddings)retriever = vector.as_retriever()
from langchain.tools.retriever import create_retriever_tool
retriever_tool = create_retriever_tool(    retriever,    "search_demand",    "找到需求文档中具体说明需求的地方",)@tooldef generate_png(uml_code, filename):    """输入plantuml代码生成图像并保存为文件"""    plantuml = PlantUML(url='https://plantuml.ceshiren.com/img/')    image_bytes = plantuml.processes(uml_code)    with open(f'{filename}.png', 'wb') as f:        f.write(image_bytes)
tools = [retriever_tool, generate_png]llm_with_tools = llm.bind_tools(tools)
prompt = hub.pull("hwchase17/openai-tools-agent")agent = create_openai_tools_agent(llm, tools, prompt, )agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)result = agent_executor.invoke({    "input": """我是一个测试工程师，我需要从以上的需求文档中梳理出来需求信息，请帮我将所有的需求梳理出来，"             "思维导图的第一级是需求文档中的4.x开头的标题信息，表示功能模块，第二级是该功能模块的测试点，"             "请先输出一个 plantuml 格式的源码，源码格式如代码内所示            @startmindmap            * root node                * some first level node                    * second level node                    * another second level node                * another first level node            @endmindmap             然后再根据源码信息输出一个plantuml格式的思维导图文件。生成一个图片文件，文件名为 hogwarts加任意随机数"""})