pip install langchain openai psycopg2-binary pgvector-python tiktoken export OPENAI_API_KEY="sk-..." Docker で Postgres + pgvector を立てておくと楽： docker run --name rag_pg -e POSTGRES_PASSWORD=rag -p 5432:5432 \ ankane/pgvector

# 01_rag_agent.py import os, textract, glob from langchain.embeddings import OpenAIEmbeddings from langchain.vectorstores.pgvector import PGVector from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain.docstore.document import Document from langchain.tools import Tool from langchain.chat_models import ChatOpenAI from langchain.agents import initialize_agent from langchain.chains import RetrievalQA ### 1) ベクトルストア接続 ------------------------ CONN = "postgresql://postgres:rag@localhost:5432/postgres" COLL = "eu_docs" embed = OpenAIEmbeddings() vectordb = PGVector( connection_string=CONN, collection_name=COLL, embedding_function=embed) ### 2) ドキュメント投入 (初回のみ) -------------- if not vectordb.check_if_collection_exists(): splitter = RecursiveCharacterTextSplitter(chunk_size=800, chunk_overlap=80) docs = [] for file in glob.glob("./docs/*.pdf"): text = textract.process(file, method="pdftotext").decode() for chunk in splitter.split_text(text): docs.append(Document(page_content=chunk, metadata={"src": file})) vectordb.add_documents(docs) print(f"💾 {len(docs)} chunks embedded!") ### 3) Retriever + QA チェーン ------------------- retriever = vectordb.as_retriever(search_kwargs={"k":4}) qa_chain = RetrievalQA.from_chain_type( llm=ChatOpenAI(model="gpt-4o-mini", temperature=0), chain_type="stuff", retriever=retriever, return_source_documents=True) def rag_tool(q: str) -> str: res = qa_chain(q) answer = res["result"] sources = {d.metadata['src'].split('/')[-1] for d in res['source_documents']} return f"{answer}\n\n🔖 Sources: {', '.join(sources)}" ### 4) ReAct エージェント ------------------------ tools = [Tool(name="RAG", func=rag_tool, description="Use this tool to answer using internal docs")] agent = initialize_agent(tools, ChatOpenAI(model="gpt-4o-mini"), agent_type="chat-zero-shot-react-description") ### 5) 試す！ ------------------------------------ while True: q = input("\n🗨 質問 > ") print(agent.run(q))

python 01_rag_agent.py