LangChain.js用のBox loaderでRAG対応のアプリを作成

import { BoxLoader, BoxAuth, BoxAuthType } from 'langchainjs-box';

const auth = new BoxAuth({
  authType: BoxAuthType.TOKEN,
  boxDeveloperToken: 'DEVELOPER_TOKEN'
});

import { BoxLoader, BoxAuth, BoxAuthType } from 'langchainjs-box';

const auth = new BoxAuth({
  authType: BoxAuthType.JWT,
  boxJwtPath: './path/to/jwt-config.json',
  boxUserId: 'USER_ID'
});

import { BoxLoader, BoxAuth, BoxAuthType } from 'langchainjs-box';

const auth = new BoxAuth({
  authType: BoxAuthType.CCG,
  boxClientId: 'CLIENT_ID',
  boxClientSecret: 'CLIENT_SECRET',
  boxEnterpriseId: 'ENTERPRISE_ID'
});

import { BoxLoader, BoxAuth, BoxAuthType } from 'langchainjs-box';

const auth = new BoxAuth({
  authType: BoxAuthType.TOKEN,
  boxDeveloperToken: 'DEVELOPER_TOKEN'
});
const loader = new BoxLoader({
  boxAuth: auth,
  boxFileIds: ['FILE_ID_1', 'FILE_ID_2'],
  characterLimit: 10000  // Optional, defaults to no limit
});
const docs = await loader.load();

const loader = new BoxLoader({
  boxAuth: auth,
  boxFolderId: 'FOLDER_ID'
});

const docs = await loader.load();

const loader = new BoxLoader({
  boxAuth: auth,
  boxFolderId: 'FOLDER_ID',
  recursive: true,  // Optional, defaults to false
  characterLimit: 10000  // Optional, defaults to no limit
});

const docs = await loader.load();

const loader = new BoxLoader({
  boxAuth: auth,
  boxFolderId: 'FOLDER_ID'
});

for await (const doc of loader.lazyLoad()) {
  // Process each document
}

import 'dotenv/config';
import { createRequire } from 'module';
const require = createRequire(import.meta.url);
const { BoxLoader, BoxAuth, BoxAuthType } = require('langchainjs-box');
import { RecursiveCharacterTextSplitter } from '@langchain/textsplitters';
import { ChatOpenAI, OpenAIEmbeddings } from '@langchain/openai';
import { MemoryVectorStore } from 'langchain/vectorstores/memory';
import { RunnableSequence } from '@langchain/core/runnables';
import { StringOutputParser } from '@langchain/core/output_parsers';

// You can use other authentication methods
async function loadBoxDocuments() {
  const developerToken = process.env.BOX_DEVELOPER_TOKEN;
  const boxFileIds = (process.env.BOX_FILE_IDS || '').split(',').map(s => s.trim()).filter(Boolean);
  const boxFolderId = process.env.BOX_FOLDER_ID?.trim();
  let auth;
  if (developerToken) {
    auth = new BoxAuth({
      authType: BoxAuthType.TOKEN,
      boxDeveloperToken: developerToken
    });
  }
  const loader = new BoxLoader({
    boxAuth: auth,
    boxFileIds: boxFileIds.length ? boxFileIds : undefined,
    characterLimit: process.env.CHAR_LIMIT ? Number(process.env.CHAR_LIMIT) : undefined
  });
  return loader.load();
}
async function main() {
  const openaiApiKey = process.env.OPENAI_API_KEY;
  if (!openaiApiKey) {
    throw new Error('Missing OPENAI_API_KEY in environment');
  }
  if (!process.env.BOX_DEVELOPER_TOKEN && !process.env.BOX_FILE_IDS && !process.env.BOX_FOLDER_ID) {
    throw new Error('Provide BOX_DEVELOPER_TOKEN and either BOX_FILE_IDS (comma-separated) or BOX_FOLDER_ID');
  }
  console.log('Loading documents from Box...');
  const docs = await loadBoxDocuments();
  console.log(`Loaded ${docs.length} documents`);
  const splitter = new RecursiveCharacterTextSplitter({
    chunkSize: 1000,
    chunkOverlap: 200
  });
  const splits = await splitter.splitDocuments(docs);
  const cleanSplits = splits.filter(d => d.pageContent && d.pageContent.trim().length > 0 && !d.pageContent.trim().startsWith('[Error'));
  console.log(`Created ${cleanSplits.length} clean chunks (from ${splits.length})`);
  const embeddings = new OpenAIEmbeddings({ apiKey: openaiApiKey, model: 'text-embedding-3-small' });
  const vectorstore = await MemoryVectorStore.fromDocuments(cleanSplits, embeddings);
  const retriever = vectorstore.asRetriever({ searchType: 'mmr', searchKwargs: { fetchK: 20 }, k: 8 });
  const question = process.env.QUESTION || 'What are the key points in these documents?';
  const prompt = (docsText, q) => `You are a helpful assistant.
Use the provided context to answer the question.
If the answer is not in the context, say you don't know.
Context:
${docsText}
Question: ${q}
Answer:`;
  const model = new ChatOpenAI({ model: 'gpt-4o-mini', temperature: 0, apiKey: openaiApiKey });
  const chain = RunnableSequence.from([
    async (input) => {
      const retrieved = await retriever.getRelevantDocuments(input.question);
      const context = retrieved.map(d => d.pageContent).join('\n\n');
      console.log('Context:', context);
      return { promptText: prompt(context, input.question) };
    },
    (x) => model.invoke(x.promptText),
    new StringOutputParser()
  ]);
  console.log('Asking question:', question);
  const answer = await chain.invoke({ question });
  console.log('\nAnswer:\n', answer);
}
main().catch(err => {
  console.error(err);
  process.exit(1);
});