LangChain / RAG / Vector DBs · レッスン

外部APIのツールとしての統合

RAGシステムを外部APIやサービスに接続し、エージェントがリアルタイムデータを取得したりアクションを実行したりできるようにします。

「外部APIのツールとしての統合」はCoddyKit上の無料LangChain / RAG / Vector DBsレッスンです。これはレッスン3/4です。下記で完全なレッスンを無料で読むことができます。その後、ブラウザ内の組み込みコードエディタと24時間対応のAIチューターでハンズオン演習できます。これはLangChain / RAG / Vector DBs学習パスの一部であり、ウェブとCoddyKitアプリ全体で進捗が同期されます。 LangChain / RAG / Vector DBsコースには全4レッスンが含まれています。

このレッスンの一部はまだ翻訳されておらず、英語で表示されています。

Agents & External APIs

Agents can do more than just generate text! They can interact with the real world by using special functions called "tools." External APIs are perfect for this, allowing agents to fetch real-time data or perform actions.

Tools for Real-World Interaction

In LangChain, a tool is essentially a Python function that an agent can call. These functions can wrap almost anything:

Searching the web
Querying a database
Sending emails
Interacting with an external API

Tools give agents superpowers, extending their capabilities beyond their internal knowledge.

Crafting an API Tool

To integrate an external API, you first define a standard Python function that makes the API call. Then, you wrap this function using LangChain's Tool class.

The Tool needs:

A name: How the agent refers to it.
A description: What the tool does and its expected input.
The actual function (func) that performs the action.

Simple Tool Function Example

Let's start with a basic Python function that simulates an API call. This function will "get current time" for a given city.

Notice how the function takes a single string argument, which the agent will provide based on its reasoning.

from langchain.tools import Tool

def get_current_time(city: str) -> str:
    """Gets the current time for a specified city."""
    # This would typically make an actual API call
    if city.lower() == "london":
        return "10:30 AM GMT"
    elif city.lower() == "new york":
        return "05:30 AM EST"
    else:
        return "Time not available for that city."

# This function will be wrapped into a Tool later.

Agent Using a Simple Tool

Here's a complete example where a LangChain agent uses our get_current_time tool. The agent decides when to call the tool based on the user's query and the tool's description.

Note: You'll need to set your OPENAI_API_KEY environment variable for this to run.

import os
from langchain.tools import Tool
from langchain_openai import OpenAI
from langchain.agents import AgentExecutor, create_react_agent
from langchain import hub

# Mock function for getting time (simulates API)
def get_current_time(city: str) -> str:
    """Gets the current time for a specified city."""
    if city.lower() == "london":
        return "10:30 AM GMT"
    elif city.lower() == "new york":
        return "05:30 AM EST"
    else:
        return "Time not available for that city."

# Create the LangChain Tool
time_tool = Tool(
    name="get_time",
    func=get_current_time,
    description="Useful for getting the current time for a given city. Input should be a city name (e.g., 'London')."
)

# List of tools available to the agent
tools = [time_tool]

# Initialize the LLM (ensure OPENAI_API_KEY is set)
llm = OpenAI(temperature=0)

# Pull the ReAct prompt template from LangChain Hub
prompt = hub.pull("hwchase17/react")

# Create the agent
agent = create_react_agent(llm, tools, prompt)

# Create the AgentExecutor to run the agent
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)

# Invoke the agent with a query
response = agent_executor.invoke({"input": "What time is it in London?"})
print(response["output"])

Connecting to Real APIs with `requests`

For real-world API interactions, Python's requests library is your go-to. It simplifies making HTTP requests to external services.

Within your tool function, you'll typically send a request, receive an HTTP response, and then parse its content (often JSON) to extract the relevant data.

import requests

def fetch_data_from_api(query: str) -> str:
    """Fetches data from a hypothetical external API."""
    try:
        # Example: calling a public API (e.g., a mock weather API)
        # Replace with your actual API endpoint and parameters
        api_url = f"https://api.example.com/data?q={query}" # Placeholder URL
        response = requests.get(api_url)
        response.raise_for_status() # Raise HTTPError for bad responses (4xx or 5xx)
        data = response.json()
        return str(data) # Return relevant part of the data
    except requests.exceptions.RequestException as e:
        return f"Error fetching data: {e}"

# This function would then be wrapped in a LangChain Tool.

Building a Weather Tool

Let's create a more practical example: a tool that fetches current weather for a city using a mock weather API. In a real scenario, you'd integrate with a service like OpenWeatherMap.

Notice the detailed description, guiding the LLM on the input format.

import requests
from langchain.tools import Tool

# Mock weather API function
def get_weather_for_city(city: str) -> str:
    """
    Fetches the current weather for a specified city.
    Uses a mock API for demonstration.
    """
    # In a real app, replace with an actual weather API call
    mock_weather_data = {
        "london": {"temperature": "15C", "conditions": "Cloudy"},
        "new york": {"temperature": "22C", "conditions": "Sunny"},
        "paris": {"temperature": "18C", "conditions": "Partly Cloudy"}
    }
    weather = mock_weather_data.get(city.lower())
    if weather:
        return f"The current weather in {city} is {weather['temperature']} and {weather['conditions']}."
    else:
        return f"Weather data not available for {city}."

# Create the LangChain Tool for weather
weather_tool = Tool(
    name="get_weather",
    func=get_weather_for_city,
    description="Useful for getting the current weather conditions (temperature, conditions) for a specific city. Input should be a city name (e.g., 'London')."
)

# This tool can now be added to an agent's tool list.

Agent Calling Weather API

Now, let's see an agent use our get_weather tool. The agent's reasoning process (if verbose=True) will show it deciding to use the tool, calling it, and then using the result to answer the query.

import os
from langchain.tools import Tool
from langchain_openai import OpenAI
from langchain.agents import AgentExecutor, create_react_agent
from langchain import hub

# Mock weather API function (from previous scene)
def get_weather_for_city(city: str) -> str:
    """
    Fetches the current weather for a specified city.
    Uses a mock API for demonstration.
    """
    mock_weather_data = {
        "london": {"temperature": "15C", "conditions": "Cloudy"},
        "new york": {"temperature": "22C", "conditions": "Sunny"},
        "paris": {"temperature": "18C", "conditions": "Partly Cloudy"}
    }
    weather = mock_weather_data.get(city.lower())
    if weather:
        return f"The current weather in {city} is {weather['temperature']} and {weather['conditions']}."
    else:
        return f"Weather data not available for {city}."

# Create the LangChain Tool for weather
weather_tool = Tool(
    name="get_weather",
    func=get_weather_for_city,
    description="Useful for getting the current weather conditions (temperature, conditions) for a specific city. Input should be a city name (e.g., 'London')."
)

# List of tools available to the agent
tools = [weather_tool]

# Initialize the LLM (ensure OPENAI_API_KEY is set)
llm = OpenAI(temperature=0)

# Pull the ReAct prompt template from LangChain Hub
prompt = hub.pull("hwchase17/react")

# Create the agent
agent = create_react_agent(llm, tools, prompt)

# Create the AgentExecutor to run the agent
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)

# Invoke the agent with a query
response = agent_executor.invoke({"input": "What's the weather like in Paris?"})
print(response["output"])

Effective Tool Descriptions

The LLM relies heavily on your tool's description to decide if and how to use it. Make your descriptions:

Clear and concise: State exactly what the tool does.
Specific about input: Clearly explain what kind of argument it expects.
Purpose-driven: Explain *why* the tool is useful or in what scenarios.
Provide examples: If the input is complex, offer an example format.

A well-crafted description is crucial for an agent's success.

Tool Integration Check

You're building an agent to help users find real-time information. You need to create a tool that fetches current stock prices for a given company ticker (e.g., "AAPL") by calling an external API.

Which of the following describes the most important aspect of the Tool definition for the LLM to understand and use it correctly?

Recap: APIs as Agent Tools

You've learned how to empower LangChain agents by integrating external APIs as tools. This allows your agents to:

Access real-time data from the internet.
Perform actions in the real world (like sending emails or updating databases).
Extend their capabilities far beyond their initial training data.

By wrapping API calls in well-described Tool objects, you significantly enhance your agent's intelligence and utility!

よくある質問

「外部APIのツールとしての統合」レッスンは無料ですか？

はい。「外部APIのツールとしての統合」の完全なテキストはこのウェブで無料で読めます。インタラクティブに演習し（組み込みコードエディタと24時間対応のAIチューター）、LangChain / RAG / Vector DBsコースの残りをアンロックするには、CoddyKit PROにアップグレードしてください。 LangChain / RAG / Vector DBsコースには全4レッスンが含まれています。

「外部APIのツールとしての統合」で何を学びますか？

RAGシステムを外部APIやサービスに接続し、エージェントがリアルタイムデータを取得したりアクションを実行したりできるようにします。ブラウザで直接実行するハンズオンコードでLangChain / RAG / Vector DBsを演習し、24時間対応のAIチューターがレッスンを進める中での質問に答えます。

LangChain / RAG / Vector DBsを始めるのに経験は必要ですか？

事前経験は必要ありません。CoddyKitのLangChain / RAG / Vector DBsは初級者から上級者向けに構成されているため、ここから始めるか最初から始めて、自分のペースで進むことができます。これはレッスン3/4です。

「外部APIのツールとしての統合」レッスンにはどのくらい時間がかかりますか？

ほとんどのCoddyKitレッスンは約5～10分かかります。各レッスンはコンパクトでインタラクティブなので、着実に進歩し、ウェブとアプリ全体で正確に前回の場所から再開できます。

このLangChain / RAG / Vector DBsレッスンでコードを書いて実行できますか？

はい。すべてのLangChain / RAG / Vector DBsレッスンに組み込みコードエディタが含まれているため、ブラウザでリアルコードを書いて実行し、即座のAIフィードバックを取得できます。ローカル設定は不要です。

このコースのすべてのレッスン

← LangChain / RAG / Vector DBsに戻る