AI框架

引言 LangChain从2022年的一个LLM调用库，发展到2026年的完整AI应用平台。LangGraph、LangSmith、LangServe构成了LangChain生态系统。本文将全面介绍2026年LangChain的演进。 LangChain 2026架构 LangChain生态系统 ├── LangChain (核心库) │ ├── Models (模型抽象) │ ├── Prompts (提示管理) │ ├── Chains (链式调用) │ ├── Agents (智能体) │ ├── Memory (记忆) │ ├── Retrievers (检索器) │ └── Tools (工具集) ├── LangGraph (Agent编排) │ ├── State Graph (状态图) │ ├── Checkpointing (检查点) │ └── Human-in-loop (人机协作) ├── LangSmith (可观测性) │ ├── Tracing (追踪) │ ├── Evaluation (评估) │ └── Monitoring (监控) └── LangServe (部署) ├── API Server └── Streaming (流式) LangChain核心库 模型抽象 from langchain_community.llms import Ollama from langchain_openai import ChatOpenAI from langchain_anthropic import ChatAnthropic # 统一接口，不同后端 models = { "gpt5": ChatOpenAI(model="gpt-5"), "claude4": ChatAnthropic(model="claude-4-opus"), "glm5": Ollama(model="glm-5:32b"), } # 统一调用 for name, model in models.items(): response = model.invoke("你好") print(f"{name}: {response.content}") 提示管理 from langchain.prompts import ChatPromptTemplate # 提示模板 prompt = ChatPromptTemplate.from_messages([ ("system", "你是一个{role}。"), ("human", "{question}") ]) # 链式调用 chain = prompt | model | output_parser response = chain.invoke({"role": "数学老师", "question": "1+1=?"}) RAG from langchain_community.embeddings import OllamaEmbeddings from langchain_community.vectorstores import Chroma from langchain.text_splitter import RecursiveCharacterTextSplitter # 文档处理 splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50) chunks = splitter.split_text(document) # 嵌入和存储 embeddings = OllamaEmbeddings(model="bge-large-zh") vectorstore = Chroma.from_texts(chunks, embeddings) # RAG链 retriever = vectorstore.as_retriever(search_kwargs={"k": 5}) rag_chain = ( {"context": retriever, "question": RunnablePassthrough()} | prompt | model | output_parser ) LangGraph 2026年最重要的Agent编排工具： ...