Langchain RAG

处理流程：

1. 索引（Index）：加载 → 切分 → 嵌入向量化 → 存储
2. 检索（Retrieve）：查询 → 嵌入向量化 → 检索 → 返回文档
3. 生成（Generate）：文档 + 查询 → LLM → 回答

核心组件：

• 文档加载器（Document Loaders）：从文件、网页、数据库中摄取数据
• 文本切分器（Text Splitters）：将文档切分为块（chunks）
• 嵌入模型（Embeddings）：将文本转换为向量
• 向量存储（Vector Stores）：存储并检索向量

完整的 RAG 流程

from langchain_openai import ChatOpenAI, OpenAIEmbeddings
from langchain_community.vectorstores import InMemoryVectorStore
from langchain_text_splitters import RecursiveCharacterTextSplitter
from langchain_core.documents import Document

# 1. 加载文档
docs = [
    Document(page_content="LangChain is a framework for LLM apps.", metadata={}),
    Document(page_content="RAG = Retrieval Augmented Generation.", metadata={}),
]

# 2. 切分文档
splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
splits = splitter.split_documents(docs)

# 3. 创建嵌入并存储
embeddings = OpenAIEmbeddings(model="text-embedding-3-small")
vectorstore = InMemoryVectorStore.from_documents(splits, embeddings)

# 4. 创建检索器
retriever = vectorstore.as_retriever(search_kwargs={"k": 4})

# 5. 在 RAG 中使用
model = ChatOpenAI(model="gpt-4.1")
query = "What is RAG?"
relevant_docs = retriever.invoke(query)

context = "\n\n".join([doc.page_content for doc in relevant_docs])
response = model.invoke([
    {"role": "system", "content": f"Use this context:\n\n{context}"},
    {"role": "user", "content": query},
])

import { ChatOpenAI, OpenAIEmbeddings } from "@langchain/openai";
import { MemoryVectorStore } from "@langchain/classic/vectorstores/memory";
import { RecursiveCharacterTextSplitter } from "@langchain/textsplitters";
import { Document } from "@langchain/core/documents";

// 1. 加载文档
const docs = [
  new Document({ pageContent: "LangChain is a framework for LLM apps.", metadata: {} }),
  new Document({ pageContent: "RAG = Retrieval Augmented Generation.", metadata: {} }),
];

// 2. 切分文档
const splitter = new RecursiveCharacterTextSplitter({ chunkSize: 500, chunkOverlap: 50 });
const splits = await splitter.splitDocuments(docs);

// 3. 创建嵌入并存储
const embeddings = new OpenAIEmbeddings({ model: "text-embedding-3-small" });
const vectorstore = await MemoryVectorStore.fromDocuments(splits, embeddings);

// 4. 创建检索器
const retriever = vectorstore.asRetriever({ k: 4 });

// 5. 在 RAG 中使用
const model = new ChatOpenAI({ model: "gpt-4.1" });
const query = "What is RAG?";
const relevantDocs = await retriever.invoke(query);

const context = relevantDocs.map(doc => doc.pageContent).join("\n\n");
const response = await model.invoke([
  { role: "system", content: `Use this context:\n\n${context}` },
  { role: "user", content: query },
]);

文档加载器

from langchain_community.document_loaders import PyPDFLoader

loader = PyPDFLoader("./document.pdf")
docs = loader.load()
print(f"Loaded {len(docs)} pages")

import { PDFLoader } from "@langchain/community/document_loaders/fs/pdf";

const loader = new PDFLoader("./document.pdf");
const docs = await loader.load();
console.log(`Loaded ${docs.length} pages`);

from langchain_community.document_loaders import WebBaseLoader

loader = WebBaseLoader("https://docs.langchain.com")
docs = loader.load()

import { CheerioWebBaseLoader } from "@langchain/community/document_loaders/web/cheerio";

const loader = new CheerioWebBaseLoader("https://docs.langchain.com");
const docs = await loader.load();

from langchain_community.document_loaders import DirectoryLoader, TextLoader

# 从目录中加载所有文本文件
loader = DirectoryLoader(
    "path/to/documents",
    glob="**/*.txt",  # 要加载的文件模式
    loader_cls=TextLoader
)
docs = loader.load()

文本切分

from langchain_text_splitters import RecursiveCharacterTextSplitter

splitter = RecursiveCharacterTextSplitter(
    chunk_size=1000,        # 每个块的字符数
    chunk_overlap=200,      # 用于保持上下文连贯性的重叠字符数
    separators=["\n\n", "\n", " ", ""],  # 切分层级优先级
)

splits = splitter.split_documents(docs)

向量存储

from langchain_chroma import Chroma
from langchain_openai import OpenAIEmbeddings

vectorstore = Chroma.from_documents(
    documents=splits,
    embedding=OpenAIEmbeddings(),
    persist_directory="./chroma_db",
    collection_name="my-collection",
)

# 加载已有存储
vectorstore = Chroma(
    persist_directory="./chroma_db",
    embedding_function=OpenAIEmbeddings(),
    collection_name="my-collection",
)

import { Chroma } from "@langchain/community/vectorstores/chroma";
import { OpenAIEmbeddings } from "@langchain/openai";

const vectorstore = await Chroma.fromDocuments(
  splits,
  new OpenAIEmbeddings(),
  { collectionName: "my-collection", url: "http://localhost:8000" }
);

from langchain_community.vectorstores import FAISS

vectorstore = FAISS.from_documents(splits, embeddings)
vectorstore.save_local("./faiss_index")

# 仅加载由您创建且完全受控的 FAISS 索引。
# Python FAISS 加载器使用 pickle 存储元数据，因此切勿加载
# 下载的、共享的或其他不受信任的索引目录。
loaded = FAISS.load_local(
    "./faiss_index",
    embeddings,
    allow_dangerous_deserialization=True,
)

import { FaissStore } from "@langchain/community/vectorstores/faiss";

const vectorstore = await FaissStore.fromDocuments(splits, embeddings);
await vectorstore.save("./faiss_index");

const loaded = await FaissStore.load("./faiss_index", embeddings);

检索

# 基础检索
results = vectorstore.similarity_search(query, k=5)

# 带有得分的检索
results_with_score = vectorstore.similarity_search_with_score(query, k=5)
for doc, score in results_with_score:
    print(f"Score: {score}, Content: {doc.page_content}")

// 基础检索
const results = await vectorstore.similaritySearch(query, 5);

// 带有得分的检索
const resultsWithScore = await vectorstore.similaritySearchWithScore(query, 5);
for (const [doc, score] of resultsWithScore) {
  console.log(`Score: ${score}, Content: ${doc.pageContent}`);
}

# MMR 平衡相关性与多样性
retriever = vectorstore.as_retriever(
    search_type="mmr",
    search_kwargs={"fetch_k": 20, "lambda_mult": 0.5, "k": 5},
)

# 创建文档时添加元数据
docs = [
    Document(
        page_content="Python programming guide",
        metadata={"language": "python", "topic": "programming"}
    ),
]

# 带过滤条件的检索
results = vectorstore.similarity_search(
    "programming",
    k=5,
    filter={"language": "python"}  # 仅检索 Python 相关文档
)

from langchain.agents import create_agent
from langchain.tools import tool

@tool
def search_docs(query: str) -> str:
    """在文档中检索相关信息。"""
    docs = retriever.invoke(query)
    return "\n\n".join([d.page_content for d in docs])

agent = create_agent(
    model="gpt-4.1",
    tools=[search_docs],
)

result = agent.invoke({
    "messages": [{"role": "user", "content": "How do I create an agent?"}]
})

import { createAgent } from "langchain";
import { tool } from "@langchain/core/tools";
import { z } from "zod";

const searchDocs = tool(
  async (input) => {
    const docs = await retriever.invoke(input.query);
    return docs.map(d => d.pageContent).join("\n\n");
  },
  {
    name: "search_docs",
    description: "Search documentation for relevant information.",
    schema: z.object({ query: z.string() }),
  }
);

const agent = createAgent({
  model: "gpt-4.1",
  tools: [searchDocs],
});

const result = await agent.invoke({
  messages: [{ role: "user", content: "How do I create an agent?" }],
});

• 块大小 / 重叠长度（Chunk size/overlap）
• 嵌入模型
• 检索返回数量（k）
• 元数据过滤器
• 检索算法：Similarity、MMR

不能配置的内容

• 嵌入向量维度（由各模型决定）
• 在同一个向量存储中混用来自不同模型的嵌入向量

# 错误：太小（丢失上下文）或太大（超出限制）
splitter = RecursiveCharacterTextSplitter(chunk_size=50)
splitter = RecursiveCharacterTextSplitter(chunk_size=10000)

# 正确
splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)

// 错误：过小或过大
const splitter = new RecursiveCharacterTextSplitter({ chunkSize: 50 });

// 正确
const splitter = new RecursiveCharacterTextSplitter({ chunkSize: 1000, chunkOverlap: 200 });

# 错误：无重叠 - 边界处的上下文会断裂
splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=0)

# 正确：10-20% 的重叠
splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)

# 错误：InMemory - 重启后数据丢失
vectorstore = InMemoryVectorStore.from_documents(docs, embeddings)

# 正确
vectorstore = Chroma.from_documents(docs, embeddings, persist_directory="./chroma_db")

// 错误：Memory - 重启后数据丢失
const vectorstore = await MemoryVectorStore.fromDocuments(docs, embeddings);

// 正确
const vectorstore = await Chroma.fromDocuments(docs, embeddings, { collectionName: "my-collection" });

# 错误：索引与查询使用不同的嵌入模型 - 互不兼容！
vectorstore = Chroma.from_documents(docs, OpenAIEmbeddings(model="text-embedding-3-small"))
retriever = vectorstore.as_retriever(embeddings=OpenAIEmbeddings(model="text-embedding-3-large"))

# 正确：使用相同的模型
embeddings = OpenAIEmbeddings(model="text-embedding-3-small")
vectorstore = Chroma.from_documents(docs, embeddings)
retriever = vectorstore.as_retriever()  # 使用相同的嵌入模型

const embeddings = new OpenAIEmbeddings({ model: "text-embedding-3-small" });
const vectorstore = await Chroma.fromDocuments(docs, embeddings);
const retriever = vectorstore.asRetriever();  // 使用相同的嵌入模型

# 错误：加载下载的、共享的、云端托管的或由第三方控制的
# 且启用了危险反序列化的 FAISS 索引。
loaded_store = FAISS.load_local(
    "./untrusted_faiss_index",
    embeddings,
    allow_dangerous_deserialization=True,
)

# 正确：仅在该索引目录由您创建且始终在您的控制之下时才启用。
loaded_store = FAISS.load_local(
    "./faiss_index",
    embeddings,
    allow_dangerous_deserialization=True,
)

如果无法保证持久化索引的来源安全性，请勿使用 allow_dangerous_deserialization=True 进行加载。请从受信任的源文档重新构建索引，或者对于不受信任的文件使用不需要 pickle 反序列化的向量存储/后端。

# 错误：索引为 1536 维，但使用了 512 维的嵌入
pc.create_index(name="idx", dimension=1536, metric="cosine")
vectorstore = PineconeVectorStore.from_documents(
    docs, OpenAIEmbeddings(model="text-embedding-3-small", dimensions=512), index=pc.Index("idx")
)  # 错误：维度不匹配！

# 正确：匹配维度
embeddings = OpenAIEmbeddings()  # 默认 1536 维