npm install langchain @langchain/openai @langchain/textsplitters \
  @langchain/community @langchain/core

// rag-min.ts
import { ChatOpenAI, OpenAIEmbeddings } from "@langchain/openai";
import { MemoryVectorStore } from "langchain/vectorstores/memory";
import { Document } from "@langchain/core/documents";
 
// 1. 假装这是从 loader 加载来的文档
const docs = [
  new Document({ pageContent: "我司支持 7 天无理由退货，需保持商品完好。" }),
  new Document({ pageContent: "VIP 用户的退货时限延长至 14 天。" }),
  new Document({ pageContent: "生鲜类商品一经签收，不支持无理由退货。" }),
];
 
// 2. 向量化并入库
const vectorStore = await MemoryVectorStore.fromDocuments(
  docs,
  new OpenAIEmbeddings({ model: "text-embedding-3-large" })
);
 
// 3. 检索 + 生成
const question = "我买的草莓不喜欢能退吗？";
const retrieved = await vectorStore.similaritySearch(question, 2);
const context = retrieved.map((d) => d.pageContent).join("\n");
 
const model = new ChatOpenAI({ model: "gpt-4o", temperature: 0 });
const answer = await model.invoke(
  `只根据以下资料回答用户问题，资料没说就回答"暂无相关政策"。
 
资料:
${context}
 
用户: ${question}`
);
 
console.log(answer.text);

OPENAI_API_KEY=sk-xxx npx tsx rag-min.ts
# 输出大意：根据资料，生鲜类商品（包括草莓）签收后不支持无理由退货。

import { PDFLoader } from "@langchain/community/document_loaders/fs/pdf";
import { RecursiveCharacterTextSplitter } from "@langchain/textsplitters";
 
const loader = new PDFLoader("./policies/return-policy.pdf");
const rawDocs = await loader.load();
 
const splitter = new RecursiveCharacterTextSplitter({
  chunkSize: 500,
  chunkOverlap: 50,
});
const chunks = await splitter.splitDocuments(rawDocs);

import { Chroma } from "@langchain/community/vectorstores/chroma";
 
const vectorStore = await Chroma.fromDocuments(
  chunks,
  new OpenAIEmbeddings({ model: "text-embedding-3-large" }),
  {
    collectionName: "return-policy",
    url: "http://localhost:8000",
  }
);
 
// 后续 query
const found = await vectorStore.similaritySearch(question, 4);

import type { VectorStore } from "@langchain/core/vectorstores";
import type { BaseChatModel } from "@langchain/core/language_models/chat_models";
 
interface AskOptions {
  vectorStore: VectorStore;
  model: BaseChatModel;
  k?: number;
}
 
async function ask(question: string, opts: AskOptions): Promise<{
  answer: string;
  sources: Array<{ content: string; metadata: Record<string, unknown> }>;
}> {
  const { vectorStore, model, k = 4 } = opts;
 
  // 检索
  const docs = await vectorStore.similaritySearch(question, k);
  const context = docs
    .map((d, i) => `[${i + 1}] ${d.pageContent}`)
    .join("\n\n");
 
  // 生成
  const response = await model.invoke(
    `你是企业知识库助手。只根据下面带编号的资料回答，回答时在相关句末标注引用编号如 [1][2]。
若资料不足以回答，直说"资料中没有相关信息"。
 
资料:
${context}
 
用户问题: ${question}`
  );
 
  const text = response.contentBlocks
    .map((b) => (b.type === "text" ? b.text : ""))
    .join("");
 
  return {
    answer: text,
    sources: docs.map((d) => ({
      content: d.pageContent.slice(0, 120),
      metadata: d.metadata,
    })),
  };
}
 
// 使用
const { answer, sources } = await ask("VIP 用户退货有什么特别政策？", {
  vectorStore,
  model: new ChatOpenAI({ model: "gpt-4o", temperature: 0 }),
  k: 3,
});
 
console.log(answer);
console.log("\n引用来源:");
console.table(sources);