如何在5分鐘內打造自己的巴菲特代理

如果你可以隨時向沃倫-巴菲特請教有關股票、市場趨勢或長期投資的問題，你會怎麼做？有報道稱，巴菲特可能很快卸任伯克希爾-哈撒韋公司執行長一職，現在正是反思他的原則的持久價值的好時機。數十年來，巴菲特一直是投資界的堅定代言人，他以注重價值、耐心和了解自己所擁有的資產而聞名。在本指南中，我將向你展示如何將這些原則轉化為對話式巴菲特代理，透過他的視角評估公司，並使用即時股票資料和新聞進行互動。我們的目標不是再現巴菲特，而是建立一個聊天機器人，幫助你以巴菲特的方式思考。

專案目標和架構

我們的目標很明確：建立一個能像巴菲特一樣進行互動的巴菲特代理。它應該討論投資理念，用巴菲特的核心原則分析股票，並利用即時資料。

主要組成部分包括

1. 語言模型（OpenAI）：提供對話能力和角色遵從性。
2. 語言鏈（LangChain）：作為框架，連線語言模型、工具和記憶體。
3. 股票資料 API（雅虎財經）：獲取當前股票價格和基本面資料。
4. 新聞 API (SerpAPI)：檢索最近的新聞頭條以瞭解上下文。
5. Streamlit：為使用者互動構建基於網路的聊天介面。

第 1 步：準備環境

編碼前，請確保您的計算機已準備就緒。

• 安裝 Python：您需要 Python 3.8 或更新版本。
• 獲取 API 金鑰：從 OpenAI 獲取語言功能的 API 金鑰。從 SerpAPI 獲取另一個金鑰，用於新聞搜尋。妥善保管這些金鑰。
• 安裝庫：開啟電腦終端或命令提示符。執行以下命令安裝必要的
• Python 軟體包：

pip install langchain langchain-openai langchain-community openai yfinance google-search-results streamlit python-dotenv streamlit-chat

• 建立 .env 檔案（可選）：在儲存指令碼的目錄下，建立一個名為 .env 的檔案。像這樣新增鍵值：

OPENAI_API_KEY="sk-YOUR_KEY_HERE"
SERPAPI_API_KEY="YOUR_KEY_HERE"

第 2 步：啟動指令碼並匯入庫

建立一個新的 Python 檔案（如 buffett_chatbot.py）。首先在頂部匯入所需的模組：

import streamlit as st
import os
import json
import yfinance as yf
from dotenv import load_dotenv
# LangChain components
from langchain_openai import ChatOpenAI
from langchain.agents import AgentExecutor, create_openai_functions_agent
from langchain.memory import ConversationBufferMemory
from langchain.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.messages import SystemMessage # No need for HumanMessage/AIMessage here anymore
from langchain.tools import Tool
from langchain_community.utilities import SerpAPIWrapper
# --- Load .env file (as a fallback) ---
load_dotenv()

這些匯入包括用於介面的 Streamlit、用於環境變數的 os、用於資料處理的 json、用於股票的 yfinance、用於金鑰載入的 dotenv 以及用於代理邏輯的各種 LangChain 元件。

第 3 步：設定Streamlit介面

配置基本應用佈局，為 API 金鑰建立側邊欄輸入：

# --- Page Config ---
st.set_page_config(page_title="Warren Buffett Bot", layout="wide")
st.title("Warren Buffett Investment Chatbot 📈")
st.caption("Ask me about investing, stocks, or market wisdom - in the style of Warren Buffett.")
# --- API Key Input in Sidebar ---
st.sidebar.header("API Configuration")
# Initialize session state for keys if they don't exist
if 'openai_api_key' not in st.session_state:
   st.session_state.openai_api_key = ""
if 'serpapi_api_key' not in st.session_state:
   st.session_state.serpapi_api_key = ""
# Create text input fields for keys, storing values in session state
input_openai_key = st.sidebar.text_input(
   "OpenAI API Key", type="password", value=st.session_state.openai_api_key, key="openai_input"
)
input_serpapi_key = st.sidebar.text_input(
   "SerpAPI API Key", type="password", value=st.session_state.serpapi_key, key="serpapi_input"
)
# Update session state with current input values
st.session_state.openai_api_key = input_openai_key
st.session_state.serpapi_key = input_serpapi_key
# Determine which keys are active (user input takes priority)
active_openai_key = st.session_state.openai_api_key or os.getenv("OPENAI_API_KEY")
active_serpapi_key = st.session_state.serpapi_api_key or os.getenv("SERPAPI_API_KEY")
# --- Display API Status ---
st.sidebar.header("API Status")
# (Add the if/else blocks using st.sidebar.success/error/warning as in the provided code)
if active_openai_key: st.sidebar.success(...) else: st.sidebar.error(...)
# Check and display SerpAPI status similarly

這段程式碼用於設定 Streamlit 聊天機器人應用程式的視覺化部分。它使用 st.session_state 來記住使用者在會話期間輸入的 API 金鑰。

第 4 步：定義核心設定和巴菲特角色

為人工智慧模型建立常量，並定義塑造聊天機器人個性的詳細指令（系統提示）：

# --- Constants & Prompt ---
MODEL_NAME = "gpt-4o" # Specify the OpenAI model
TEMPERATURE = 0.5     # Controls AI creativity (lower is more predictable)
MEMORY_KEY = "chat_history" # Key for storing conversation history
BUFFETT_SYSTEM_PROMPT = """
You are a conversational AI assistant modeled after Warren Buffett, the legendary value investor. Embody his persona accurately. 
**Your Core Principles:**
*   **Value Investing:** Focus on finding undervalued companies with solid fundamentals (earnings, low debt, strong management). Judge businesses, not stock tickers.
*   **Long-Term Horizon:** Think in terms of decades, not days or months. Discourage short-term speculation and market timing.
*   **Margin of Safety:** Only invest when the market price is significantly below your estimate of intrinsic value. Be conservative.
*   **Business Moats:** Favor companies with durable competitive advantages (strong brands, network effects, low-cost production, regulatory advantages).
*   **Understand the Business:** Only invest in companies you understand. "Risk comes from not knowing what you're doing."
*   **Management Quality:** Assess the integrity and competence of the company's leadership.
*   **Patience and Discipline:** Wait for the right opportunities ("fat pitches"). Avoid unnecessary activity. Be rational and unemotional.
*   **Circle of Competence:** Stick to industries and businesses you can reasonably understand. Acknowledge what you don't know.
**Your Communication Style:**
*   **Wise and Folksy:** Use simple language, analogies, and occasional humor, much like Buffett does in his letters and interviews.
*   **Patient and Calm:** Respond thoughtfully, avoiding hype or panic.
*   **Educational:** Explain your reasoning clearly, referencing your core principles.
*   **Prudent:** Be cautious about making specific buy/sell recommendations without thorough analysis based on your principles. Often, you might explain *how* you would analyze it rather than giving a direct 'yes' or 'no'.
*   **Quote Yourself:** Occasionally weave in famous Buffett quotes where appropriate (e.g., "Price is what you pay; value is what you get.", "Be fearful when others are greedy and greedy when others are fearful.").
*   **Acknowledge Limitations:** If asked about something outside your expertise (e.g., complex tech you wouldn't invest in, short-term trading), politely state it's not your area.
**Interaction Guidelines:**
*   When asked for stock recommendations, first use your tools to gather fundamental data (P/E, earnings, debt if possible) and recent news.
*   Analyze the gathered information through the lens of your core principles (moat, management, valuation, long-term prospects).
*   Explain your thought process clearly.
*   If a company seems to fit your criteria, express cautious optimism, emphasizing the need for further due diligence by the investor.
*   If a company doesn't fit (e.g., too speculative, high P/E without justification, outside circle of competence), explain why based on your principles.
*   If asked for general advice, draw upon your well-known philosophies.
*   Maintain conversational context using the provided chat history. Refer back to previous points if relevant.
Remember: You are simulating Warren Buffett. Your goal is to provide insights consistent with his philosophy and communication style, leveraging the tools for data when needed. Do not give definitive financial advice, but rather educate and explain the *Buffett way* of thinking about investments.
"""

第 5 步：建立資料獲取工具

實現聊天機器人獲取外部股票和新聞資料的功能。

# --- Tool Definitions ---
# 1. Stock Data Tool (Yahoo Finance) - No changes needed here
@st.cache_data(show_spinner=False) # Add caching for yfinance calls
def get_stock_info(symbol: str) -> str:
   # ... (keep the existing get_stock_info function code) ...
   """
   Fetches key financial data for a given stock symbol using Yahoo Finance...
   """
   try:
       ticker = yf.Ticker(symbol)
       info = ticker.info
       if not info or info.get('regularMarketPrice') is None and info.get('currentPrice') is None and info.get('previousClose') is None:
           hist = ticker.history(period="5d")
           if hist.empty:
                return f"Error: Could not retrieve any data for symbol {symbol}."
           last_close = hist['Close'].iloc[-1] if not hist.empty else 'N/A'
           current_price = info.get("currentPrice") or info.get("regularMarketPrice") or last_close
       else:
           current_price = info.get("currentPrice") or info.get("regularMarketPrice") or info.get("previousClose", "N/A")
       data = {
           "symbol": symbol, "companyName": info.get("longName", "N/A"),
           "currentPrice": current_price, "peRatio": info.get("trailingPE") or info.get("forwardPE", "N/A"),
           "earningsPerShare": info.get("trailingEps", "N/A"), "marketCap": info.get("marketCap", "N/A"),
           "dividendYield": info.get("dividendYield", "N/A"), "priceToBook": info.get("priceToBook", "N/A"),
           "sector": info.get("sector", "N/A"), "industry": info.get("industry", "N/A"),
           "summary": info.get("longBusinessSummary", "N/A")[:500] + ("..." if len(info.get("longBusinessSummary", "")) > 500 else "")
       }
       if data["currentPrice"] == "N/A": return f"Error: Could not retrieve current price for {symbol}."
       return json.dumps(data)
   except Exception as e: return f"Error fetching data for {symbol} using yfinance: {str(e)}."
stock_data_tool = Tool(
   name="get_stock_financial_data",
   func=get_stock_info,
   description="Useful for fetching fundamental financial data for a specific stock symbol (ticker)..." # Keep description
)
# 2. News Search Tool (SerpAPI) - Now uses active_serpapi_key
def create_news_search_tool(api_key):
   if api_key:
       try:
           params = {"engine": "google_news", "gl": "us", "hl": "en", "num": 5}
           search_wrapper = SerpAPIWrapper(params=params, serpapi_api_key=api_key)
           # Test connectivity during creation (optional, can slow down startup)
           # search_wrapper.run("test query")
           return Tool(
               name="search_stock_news",
               func=search_wrapper.run,
               description="Useful for searching recent news articles about a specific company or stock symbol..." # Keep description
           )
       except Exception as e:
           print(f"SerpAPI Tool Creation Warning: {e}")
           # Fallback to a dummy tool if key is provided but invalid/error occurs
           return Tool(
               name="search_stock_news",
               func=lambda x: f"News search unavailable (SerpAPI key configured, but error occurred: {e}).",
               description="News search tool (currently unavailable due to configuration error)."
           )
   else:
       # Dummy tool if no key is available
       return Tool(
           name="search_stock_news",
           func=lambda x: "News search unavailable (SerpAPI key not provided).",
           description="News search tool (unavailable - API key needed)."
       )
news_search_tool = create_news_search_tool(active_serpapi_key)
tools = [stock_data_tool, news_search_tool]

這些函式將成為股票資料分析機器人的“感官”，使其能夠訪問當前資訊。將它們封裝為工具物件後，LangChain 就可以使用它們了。

第 6 步：組裝 LangChain 代理

配置核心人工智慧邏輯：語言模型、提示結構、記憶體管理以及將它們聯絡在一起的代理執行器。這通常發生在指令碼的主要部分，也就是條件檢查中。

# --- Main App Logic ---
# Check if the essential OpenAI key is provided
if not active_openai_key:
   st.warning("Please enter your OpenAI API Key in the sidebar...", icon="🔑")
   st.stop() # Stop if no key
# --- LangChain Agent Setup (conditional on key) ---
try:
   # Initialize the OpenAI LLM
   llm = ChatOpenAI(
       model=MODEL_NAME, temperature=TEMPERATURE, openai_api_key=active_openai_key
   )
   # Create the prompt template
   prompt_template = ChatPromptTemplate.from_messages(
       [
           SystemMessage(content=BUFFETT_SYSTEM_PROMPT),
           MessagesPlaceholder(variable_name=MEMORY_KEY),
           ("human", "{input}"),
           MessagesPlaceholder(variable_name="agent_scratchpad"),
       ]
   )
   # Initialize or re-initialize agent components in session state
   reinitialize_agent = False
   # (Add the logic to check if 'agent_executor' exists or if keys changed)
   # ...
   if reinitialize_agent:
       # Initialize memory
       st.session_state['memory'] = ConversationBufferMemory(memory_key=MEMORY_KEY, return_messages=True)
       # Create the agent
       agent = create_openai_functions_agent(llm, tools, prompt_template)
       # Create the executor
       st.session_state['agent_executor'] = AgentExecutor(
           agent=agent, tools=tools, memory=st.session_state['memory'], verbose=True, # Set verbose=False for production
           handle_parsing_errors=True, max_iterations=5
       )
       # Store keys used for this agent instance
       st.session_state.agent_openai_key = active_openai_key
       st.session_state.agent_serpapi_key = active_serpapi_key
       # st.experimental_rerun() # Rerun to apply changes
   # Continue with chat history initialization and display...

這是 LangChain 聊天機器人開發的核心部分。它使用角色、工具和記憶設定代理，透過整合 OpenAI API 實現智慧對話。在這裡使用 st.session_state 對於在使用者互動過程中保持代理的記憶至關重要。

第 7 步：實現聊天互動迴圈

新增程式碼，以便透過代理顯示對話和處理使用者輸入。

# --- Chat History and Interaction ---
   # Initialize chat history if it doesn't exist
   if "messages" not in st.session_state:
       st.session_state["messages"] = [
           {"role": "assistant", "content": "Greetings! ..."} # Initial message
       ]
   # Display existing chat messages
   for msg in st.session_state.messages:
       st.chat_message(msg["role"]).write(msg["content"])
   # Get new user input
   if prompt := st.chat_input("Ask Buffett Bot..."):
       # Display user message
       st.session_state.messages.append({"role": "user", "content": prompt})
       st.chat_message("user").write(prompt)
       # Prepare input for the agent
       agent_input = {"input": prompt}
       # Invoke the agent executor
       try:
           with st.spinner("Buffett is pondering..."):
               # Get the executor instance from session state
               agent_executor_instance = st.session_state['agent_executor']
               response = agent_executor_instance.invoke(agent_input)
           # Display assistant response
           output = response.get('output', "Sorry, an error occurred.")
           st.session_state.messages.append({"role": "assistant", "content": output})
           st.chat_message("assistant").write(output)
       except Exception as e:
           # Handle errors during agent execution
           error_message = f"An error occurred: {str(e)}"
           st.error(error_message, icon="🔥")
           # Add error to chat display
           st.session_state.messages.append({"role": "assistant", "content": f"Sorry... {e}"})
           st.chat_message("assistant").write(f"Sorry... {e}")
   # Optional: Add the button to clear history
   if st.sidebar.button("Clear Chat History"):
       # (Code to clear st.session_state.messages and st.session_state.memory)
       st.rerun()

這部分使 Streamlit 聊天機器人應用程式具有互動性。它讀取使用者輸入，將其傳送給 LangChain 代理執行器，並顯示使用者的詢問和機器人生成的回覆。

第 8 步：執行巴菲特代理

儲存完整的 Python 指令碼。在指令碼目錄下開啟終端並執行：

streamlit run buffett_chatbot.py

在終端執行該檔案，瀏覽器就會開啟應用程式，您就可以輸入 API 金鑰並與聊天機器人互動了。

分析輸出

讓我們用一些問題來測試巴菲特 AI 代理。您可以從此處獲取同樣的檔案。

我們的 streamlit 應用程式看起來是這樣的，在這裡我們可以選擇填寫自己的 OpenAI 金鑰和 SerpAPI 金鑰。現在讓我們測試一下機器人…

問題 1：“Mr. Buffett, could you explain your core investment philosophy in simple terms?”

問題 2：“Analyze Apple (AAPL) based on its current fundamentals. Would you consider it a good long-term investment based on your principles?”

問題 3：“What are your thoughts on Microsoft (MSFT) considering its recent news and developments?”

根據上述輸出結果，我們可以看出機器人表現良好，並利用其所有功能實現了最終輸出。它正在使用我們之前定義的沃倫-巴菲特角色來回答所有問題。機器人利用 yfinance 獲取最新的股票價格和市盈率。SerpAPI 用於獲取股票的最新訊息。

小結

對於任何希望透過永恆原則的視角探索價值投資的人來說，巴菲特的這本參考書都是一個有用的伴侶。無論你是剛剛起步還是正在完善自己的投資方法，這款代理都能幫助你更清晰、更耐心地思考市場，就像巴菲特一樣。

你可以在這裡試用它：BuffettBot on Hugging Face。

有問題想讓代理回答嗎？請在評論中留言，我很樂意聽聽你的問題和代理是如何回答的。