LangChain现状反思：框架的野心与局限

LangChain    → LLM应用的Spring Framework
LangGraph    → 有状态Agent编排引擎
LangSmith    → 可观测性与评测平台
LangServe    → 一键部署API
LangCloud    → 托管服务

# LangChain方式
from langchain_community.document_loaders import PyPDFLoader
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import Chroma
from langchain.chains import RetrievalQA
from langchain_openai import ChatOpenAI

loader = PyPDFLoader("doc.pdf")
documents = loader.load()
splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
chunks = splitter.split_documents(documents)
embeddings = HuggingFaceEmbeddings(model_name="BAAI/bge-m3")
vectorstore = Chroma.from_documents(chunks, embeddings)
llm = ChatOpenAI(model="gpt-5")
qa_chain = RetrievalQA.from_chain_type(llm=llm, retriever=vectorstore.as_retriever())
result = qa_chain.invoke({"query": "文档说了什么？"})

# 直接用原生库
import chromadb
from openai import OpenAI

client = OpenAI()
db = chromadb.PersistentClient(path="./chroma")
collection = db.get_or_create_collection("docs")

# 读取→切块→嵌入→存储→检索→生成
# 每一步都可控、可调试、可替换

from langgraph.graph import StateGraph, END
from typing import TypedDict, Annotated
import operator

class AgentState(TypedDict):
    messages: Annotated[list, operator.add]
    next_agent: str

def researcher(state: AgentState):
    # 调用搜索工具
    return {"messages": [search_result], "next_agent": "writer"}

def writer(state: AgentState):
    # 撰写报告
    return {"messages": [report], "next_agent": END}

workflow = StateGraph(AgentState)
workflow.add_node("researcher", researcher)
workflow.add_node("writer", writer)
workflow.set_entry_point("researcher")
workflow.add_conditional_edges("researcher", lambda s: s["next_agent"])
workflow.add_edge("writer", END)

app = workflow.compile()

推荐替代方案：
- LLM调用：直接用各Provider的官方SDK
- 向量存储：直接用Chroma/Qdrant/Milvus的SDK
- 工作流编排：LangGraph或自研状态机
- 可观测性：LangSmith（这个确实好用）或OpenTelemetry
- 评测：自建评测集 + DeepEval/RAGAS

from langsmith import Client

client = Client()

# 自动追踪每次LLM调用
@client.trace
def my_rag_pipeline(question: str):
    docs = retrieve(question)
    answer = generate(question, docs)
    return answer

# 创建评测集
dataset = client.create_dataset("rag_eval_v1")
client.create_examples(
    inputs=[{"question": "什么是RAG？"}],
    outputs=[{"answer": "检索增强生成..."}],
    dataset_id=dataset.id
)

# 运行评测
results = client.run_on_dataset(
    dataset_name="rag_eval_v1",
    llm_or_chain_factory=my_rag_pipeline,
    evaluation="auto",
)