Building a Web Browsing Agent with LangChain

chap9 Project

chap9 is a LangChain learning project covering the implementation of a web browsing agent. It ran from January to February 2025, across 8 commits.

Agent Architecture

graph TD
    A[사용자 질의] --> B[LLM Agent]
    B --> C{도구 선택}
    C --> D[DuckDuckGo 검색]
    C --> E[웹 페이지 스크래핑]
    C --> F[데이터 분석]
    D --> G[결과 종합]
    E --> G
    F --> G
    G --> B
    B --> H[최종 답변]

DuckDuckGo Search Tool

Uses the free DuckDuckGo search instead of the Google API:

from langchain_community.tools import DuckDuckGoSearchRun

search = DuckDuckGoSearchRun()
tools = [search]

agent = initialize_agent(
    tools=tools,
    llm=ChatOpenAI(model="gpt-4"),
    agent=AgentType.OPENAI_FUNCTIONS,
    verbose=True
)

Streamlit UI (2/6)

Implements an interactive UI with Streamlit:

import streamlit as st

st.title("🔍 웹 검색 에이전트")

if prompt := st.chat_input("질문을 입력하세요"):
    with st.chat_message("user"):
        st.write(prompt)
    with st.chat_message("assistant"):
        with st.spinner("검색 중..."):
            result = agent.invoke({"input": prompt})
            st.write(result["output"])

Key Takeaways

1. AgentType selection: OPENAI_FUNCTIONS vs ZERO_SHOT_REACT
2. Tool composition: search + scraping + calculation
3. Agent memory: using ConversationBufferMemory
4. Streamlit integration: building a chat interface

This was a solid starting point for getting hands-on with LangChain.