Phidata代理开发简明教程
你看，即使是强大的 LLM 也并不总是无所不知（尽管他们可能表现得像个万事通！）。就像所有伟大的事物一样，他们有时也需要外界的一点帮助——在推理、数学、编码或任何复杂任务中需要额外的帮助，而不仅仅是“精心设计的文字”。这就是代理发挥作用的地方。
有很多工具可以构建 LLM 代理，但并非所有工具都易于使用。今天，我们将查看一个简单但功能强大的框架 Phidata，看看它如何让使用 LLM 代理变得轻而易举！
1、Web代理：通向外部世界的窗口
LLM 通常无法访问互联网，因此让他们体验一下外部世界会立即让他们与时俱进。
例如，让我们在 Python 中设置我们最喜欢的专有 LLM：
from openai import OpenAI
import os
os.environ["OPENAI_API_KEY"] = "get-your-own-api-key"
prompt = "Is there any plane crash in South Korea recently?"
client = OpenAI()
chat_completion = client.chat.completions.create(
messages=[
{
"role": "user",
"content": prompt,
}
],
model="gpt-4o",
)
## Let's print the output:
chat_completion.choices[0].message.contentGPT 4o 没有 2024 年 5 月之后的任何信息。因此，我们得到了预期的结果：
很抱歉，我无法提供有关最近事件或飞机失事的实时信息。为了找到最新的信息，我建议查看可靠的新闻来源或官方航空当局以获取最新更新。
现在，让我们使用 phidata 构建我们的 Web 代理。
web_agent = Agent(
name="Web Agent",
model=OpenAIChat(id="gpt-4o"),
tools=[DuckDuckGo()],
instructions=["Always include sources"],
show_tool_calls=True,
markdown=True,
)
web_agent.print_response("Is there any plane crash in South Korea recently?", stream=True)
答对了！它使用 Duckduckgo 搜索互联网并找到了最新信息。根据我的经验，这是工具最重要的应用。
2、财务代理：你自己的财务顾问
有一件事你绝对不能依赖 LLM 现有的知识库，那就是财务信息。股票价格时常发生变化，要掌握最新情况，你需要最新的数据。这就是 Phidata 介入的地方——它使用 Yahoo Finance 获取最新的财务信息，包括股票更新，因此你始终了解最新情况。
from phi.tools.yfinance import YFinanceTools
finance_agent = Agent(
name="Finance Agent",
role="Financial Expert who can analyze stock price",
model=OpenAIChat(id="gpt-4o"),
tools=[YFinanceTools(stock_price=True, analyst_recommendations=True, company_info=True)],
instructions=["Use tables to display data"],
show_tool_calls=True,
markdown=True,
)
finance_agent.print_response("Based on current stock price, should I buy, hold or sell tesla stock?", stream=True)这就是我们所得到的：
Running: ┃
┃ ┃
┃ • get_current_stock_price(symbol=TSLA) ┃
┃ • get_analyst_recommendations(symbol=TSLA) ┃
┃ • get_company_info(symbol=TSLA) ┃
┃ ┃
┃ Here's an analysis of Tesla's current stock status and recommendations: ┃
┃ ┃
┃ Current Stock Price ┃
┃ ┃
┃ • Tesla Inc. (TSLA): $431.66 USD ┃
┃ ┃
┃ Company Overview ┃
┃ ┃
┃ • Market Capitalization: $1.39 Trillion USD ┃
┃ • Sector: Consumer Cyclical ┃
┃ • Industry: Auto Manufacturers ┃
┃ • EPS (Earnings Per Share): $3.67 ┃
┃ • P/E Ratio: 117.62 ┃
┃ • 52-Week Range: $138.8 - $488.54 USD ┃
┃ • Average Prices: ┃
┃ • 50 Day: $338.59 USD ┃
┃ • 200 Day: $236.36 USD ┃
┃ ┃
┃ Analyst Recommendations ┃
┃ ┃
┃ The table below summarizes recent analyst recommendations for Tesla: ┃
┃ ┃
┃ ┃
┃ Period Strong Buy Buy Hold Sell Strong Sell ┃
┃ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ┃
┃ Current 6 13 15 8 5 ┃
┃ 1 Month Ago 5 13 17 8 5 ┃
┃ 2 Months Ago 6 12 19 7 4 ┃
┃ 3 Months Ago 5 13 18 7 4 ┃
┃ ┃
┃ ┃
┃ Financials ┃
┃ ┃
┃ • Total Cash: $33.65 Billion USD ┃
┃ • Free Cash Flow: $676.63 Million USD ┃
┃ • Operating Cash Flow: $14.48 Billion USD ┃
┃ • EBITDA: $13.24 Billion USD ┃
┃ • Revenue Growth: 7.8% ┃
┃ • Gross Margins: 18.23% ┃
┃ • EBITDA Margins: 13.63% ┃
┃ ┃
┃ Recommendation ┃
┃ ┃
┃ Considering the current stock price of $431.66 USD, which is not far from the 52-week high of $488.54 USD, and ┃
┃ the analyst recommendation breakdown with a predominant "hold" sentiment alongside a mix of buy and sell ┃
┃ recommendations, it would be prudent at this moment to: ┃
┃ ┃
┃ • Hold: if you are currently holding Tesla stock, monitoring market trends and waiting for clearer indicators ┃
┃ could be beneficial. ┃
┃ • Buy: only if you strongly believe in Tesla's growth potential and your risk tolerance supports a high P/E ┃
┃ ratio stock. ┃
┃ • Sell: only if you are looking to take profits at a nearing high resembling the 52-week peak, and have a low ┃
┃ risk tolerance. ┃
┃ ┃
┃ These recommendations are highly dependent on personal risk tolerance, investment strategy, and market ┃
┃ conditions moving forward. Consulting with a financial advisor for personalized advice 3、代理团队：利用多个代理的力量
在现实生活中，有很多情况需要多个朋友的帮助。想象一下，你正在为寻求财务建议的人建立客户服务热线。网络代理可以帮助你收集一般信息，而财务代理可以获取股票价格。但是，当你遇到需要网络和财务代理共同合作的特定问题时会发生什么？
我有 10 万美元的存款，我想在 2025 年进行多元化投资。根据过去 10 年的数据，帮助我确定我应该在哪里投资
web_agent = Agent(
name="Web Agent",
role="Search the web for stock market prices",
model=OpenAIChat(id="gpt-4o"),
tools=[DuckDuckGo()],
instructions=["Always include sources"],
show_tool_calls=True,
markdown=True,
)
finance_agent = Agent(
name="Finance Agent",
role="Financial Expert who can analyze stock price",
model=OpenAIChat(id="gpt-4o"),
tools=[YFinanceTools(stock_price=True, analyst_recommendations=True, company_info=True)],
instructions=["Use tables to display data"],
show_tool_calls=True,
markdown=True,
)
agent_team = Agent(
team=[web_agent, finance_agent],
show_tool_calls=True,
markdown=True,
)
agent_team.print_response("I have $100k savings, and I want to invest it diversely in 2025. Based on last 10 years data, help me where I should invest exactlty", stream=True)输出如下所示：
To give you a comprehensive answer, I will: ┃
┃ ┃
┃ 1 Research historical performance of various investment options over the last 10 years. ┃
┃ 2 Provide analysis and investment recommendations based on diversification strategies. ┃
┃ ┃
┃ Let me gather and analyze the data for you. I will consult financial data sources and come back with a ┃
┃ comprehensive investment plan. Running: ┃
┃ ┃
┃ • transfer_task_to_web_agent(task_description=..., expected_output=A summary of historical returns for ┃
┃ different investment types for the last 10 years., additional_information=Look for reputable financial ┃
┃ sources and reports.) ┃
┃ • transfer_task_to_finance_agent(task_description=..., expected_output=..., additional_information=Consider ┃
┃ factors like historical performance, risk, and potential growth.) ┃
┃ ┃
┃ Based on the analysis of historical performance data and the development of a diversified investment strategy, ┃
┃ here's where you should consider investing your $100,000 in 2025. ┃
┃ ┃
┃ Summary of Historical Performance (Last 10 Years) ┃
┃ ┃
┃ 1 Stocks: ┃
┃ • Average Annual Returns: 9-10% (S&P 500) ┃
┃ • Highlights: Robust performance, though volatile. ┃
┃ 2 Bonds: ┃
┃ • Average Annual Returns: 4-5% ┃
┃ • Highlights: Stability with lower returns than stocks, affected by interest rates. ┃
┃ 3 Real Estate: ┃
┃ • Average Annual Returns: 3-4% ┃
┃ • Highlights: Moderate returns; acts as an inflation hedge. ┃
┃ 4 Commodities: ┃
┃ • Variability: Performance linked to macroeconomic factors with significant volatility. ┃
┃ ┃
┃ Recommended Diversified Portfolio for 2025 ┃
┃ ┃
┃ Here's the diversification strategy for investing $100,000 in 2025: ┃
┃ ┃
┃ ┃
┃ Asset Type Allocation Percentage Notes ┃
┃ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ┃
┃ Stocks (Equities) 50% Focus on sectors like technology and green energy. ┃
┃ Bonds (Fixed Income) 25% Offers stability, important for risk mitigation. ┃
┃ Real Estate 15% A tangible asset and an inflation hedge. ┃
┃ Alternative Investments 10% Includes commodities and potentially cryptocurrency or ┃
┃ tech startups. ┃
┃ ┃
┃ ┃
┃ Rationale: ┃
┃ ┃
┃ • Stocks provide the highest growth potential but require careful selection due to higher risk. ┃
┃ • Bonds ensure stability and provide a steady income. ┃
┃ • Real Estate serves as a hedge against inflation, with potential for steady growth, especially in areas with ┃
┃ population growth. ┃
┃ • Alternative Investments add diversification and the opportunity to capitalize on emerging and unconventional ┃
┃ markets. ┃
┃ ┃
┃ This strategic allocation aims to balance risk and return, leveraging the strength of different asset classes. ┃
┃ For in-depth analysis of specific stocks or real-time market conditions, you may need to consult with further ┃
┃ financial advisors or update the data closer to your investment date. ┃
┃ ┃
┃ Sources include historical financial studies and insights from Quantified Strategies, BlackRock, and industry ┃
┃ performance reports. ┃
┃ 在输出中，请注意以下部分：transfer_task_to_web_agent… 过去 10 年不同投资类型的历史回报摘要……和 transfer_task_to_finance_agent…考虑历史表现、风险和潜在增长等因素……这表明该程序寻求两个代理来获得全面的答案。
4、Phidata 的其他代理
查看此处提供的众多工具。话虽如此，并非所有工具都能完美运行，有些工具还不完全可靠。Phidata 仍在进行中，但如果你想以更实际的方式构建完全可靠的解决方案，并且不依赖框架，请继续关注——我很快会分享一个关于从头开始创建 LLM 代理的详细教程！
原文链接：Building LLM Agents with Phidata: Let’s begin the fun
汇智网翻译整理，转载请标明出处
