01. Tools

Tools (Tool) are interfaces for agents, chains or LLMs to interact with the outside world.

You can easily take advantage of the tools using the tools provided by LangChain, and it is also possible to easily build custom tools.

The tool list integrated into LangChain can be found in the link below.

• LangChain Integrated Tools List

Copy

# Configuration file for managing API keys as environment variables
from dotenv import load_dotenv

# Load API key information
load_dotenv()

Copy

 True 

Copy

# LangSmith set up tracking. https://smith.langchain.com
# !pip install -qU langchain-teddynote
from langchain_teddynote import logging

# Enter a project name.
logging.langsmith("CH15-Tools")

Copy

 Start tracking LangSmith. 
[Project name] 
CH15-Tools 

Copy

import warnings

# Ignore warning message
warnings.filterwarnings("ignore")

built-in tools (built-in tools)

You can use predefined tools (tool) and toolkits provided by Langchain.

tool means a single tool, toolkit can tie multiple tools and use them as one tool.

Related tools can be referenced at the link below.

Reference

• LangChain Tools/Toolkits

Python REPL tool

This tool provides two main classes for running Python code in a Read-Eval-Print Loop (REPL) environment. -PythonREPLTool

Explanation

• Python shell environment.

• Execute valid Python instructions as input.

• Print to see results (...) You need to use functions.

Main features

• sanitize_input: option to purify input (default: True)

• python_repl: PythonREPL instance (default: run across range)

How to use

• PythonREPLTool instance creation

• Run Python code using run or arun, invoke method

Input tablets

• Removes unnecessary spaces, back ticks,'python' keywords, etc. from the input string.

Copy

from langchain_experimental.tools import PythonREPLTool

# Creates a tool that runs Python code.
python_tool = PythonREPLTool()

Copy

# Executes Python code and returns the results.
print(python_tool.invoke("print(100 + 200)"))

Copy

 300 

Below is an example that asks LLM to write a Python code and returns the result.

Flow cleanup

1. Ask the LLM model to write a Python code that does a specific job.

2. Run the written code to get results.

3. Output results.

Copy

from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnableLambda


# A function that executes Python code, outputs intermediate steps, and returns the results of the tool execution.
def print_and_execute(code, debug=True):
    if debug:
        print("CODE:")
        print(code)
    return python_tool.invoke(code)


# A prompt asking you to write Python code.
prompt = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            "You are Raymond Hetting, an expert python programmer, well versed in meta-programming and elegant, concise and short but well documented code. You follow the PEP8 style guide. "
            "Return only the code, no intro, no explanation, no chatty, no markdown, no code block, no nothing. Just the code.",
        ),
        ("human", "{input}"),
    ]
)
# Creating an LLM model
llm = ChatOpenAI(model="gpt-4o", temperature=0)

# Generating chains using prompts and the LLM model
chain = prompt | llm | StrOutputParser() | RunnableLambda(print_and_execute)

Copy

# Output the results
print(chain.invoke("Write a code that outputs a lotto number generator."))

Copy

 CODE: 
import random 

def generate_lotto_numbers(): 
    return sorted (random.sample (range (1, 46), 6)) 

print(generate_lotto_numbers()) 
[14, 17, 18, 22, 23, 37] 

Search API tools

A tool that implements search capabilities by utilizing the Tavily search API. This tool offers two main classes: TavilySearchResults Wow TavilyAnswer .

API key issuance address

• https://app.tavily.com/

Set the API key issued to the environment variable.

.env Set to the file as below.

Copy

TAVILY_API_KEY=tvly-abcdefghijklmnopqrstuvwxyz

TavilySearchResults

Explanation

• Query the Tavily search API and return results in JSON format.

• It is a search engine optimized for comprehensive, accurate and reliable results.

• Useful when answering questions about current events.

Main parameters

• max_results (int): Maximum number of search results to return (default: 5)

• search_depth (str): Search depth ("basic" or "advanced")

• include_domains (List[str]): List of domains to include in search results

• exclude_domains (List[str]): List of domains to exclude from search results

• include_answer (bool): Whether to include a short answer to the original query

• include_raw_content (bool): Whether each site contains refined HTML content

• include_images (bool): Query-related image list included

Return value

• String in JSON format containing search results (url, content) Or uncomment below and enter the API key that was issued.

Copy

# import os

# os.environ["TAVILY_API_KEY"] = "Enter TAVILY API key"

Copy

from langchain_community.tools.tavily_search import TavilySearchResults

# Create a tool
tool = TavilySearchResults(
    max_results=6,
    include_answer=True,
    include_raw_content=True,
    # include_images=True,
    # search_depth="advanced", # or "basic"
    include_domains=["github.io", "wikidocs.net"],
    # exclude_domains = []
)

Copy

# Run the tool
tool.invoke({"query": "LangChain Tools 에 대해서 알려주세요"})

Copy

 [ 
{'url':'https://teddylee777.github.io/langchain/langchain-agent/',  
'content':'Tag:\nAgent,\nAPI KEY,\nFAISS,\nLangChain,\nLangSmith,\nmemory,\nOpenAI,\nPython,\nRetriever,\nTavily Search,\ntools,\n Quotes,\n This agent is out of state. LangChain Korean Tutorials\n Shortcut 👀\n[LangChain] Intelligent Search System Build Guide with Agents and Tools (tools) \n2024 Year 02/09/n41 Minute Duration\n In this article you can take advantage of LangChain's Agent framework to get complex searches and 📍 Full template code\n trace\n Document-based QA system design method - Deepening \n2024 years 02/06 \n22 minutes \nLangChain's RAG system has compiled the process of finding questions-answers for documents (PDF, txt, webpage, etc.).\n'},  
{'url':'https://13akstjq.github.io/TIL/post/2024-07-09-LLMStudyDiaryComprehensiveReviewofLangChainPart4','content':'Memory. The original code works as it is. Links to Langchain's list of various memory types were broken and I looked it up. ... tools=toolkit, verbose=True) agent_executor\xa0... '},  
{'url':'https://wikidocs.net/233351','content':'Also LangChain offers MessagePlaceholder that gives you full control over the messages you want to render during formatting. I'm sure what role to use for the message prompt template\xa0... '},  
{'url':'https://ncsoft.github.io/ncresearch/f4a00ed849299e3c91fb3244e74ea7f9b974ebb7','content':'Jun 23, 2023 · LangChain is a framework that embodies the convenient patterns while using LLM, abstractioning concepts to make it easier to use these LLMs. This article... '},  
{'url':'https://13akstjq.github.io/TIL/post/2024-07-12-GenAIwithPythonLLMvsAgents','content':'Jul 12, 2024 · The main libraries are: LangChain — A mashup framework that includes almost all LLM features. The circa \xa0 where the agent should be coded at a low level... '},  
{'url':'https://wikidocs.net/234282','content':'Mar 19, 2024 · langchain is a library that provides a variety of features related to language models, of which the search tool creation function facilitates data retrieval and processing operations. '}] 

Added TavilySearch tool (custom implementation)

Description of the main parameters used in the Tavily search API

Default search settings

• query (str): keywords or sentences you want to search for

• search_depth (str): detail of the search. Select between "basic" (basic) or "advanced" (advanced)

• topic (str): Search subject area. Select between "general" (general) or "news" (news)

• days (int): The latest in search results. Returns only results within the specified number of days

• max_results (int): Maximum number of search results to be returned

Domain filtering

• include_domains (list): List of domains that must be included in the search results

• exclude_domains (list): List of domains to exclude from search results

Result detail setting

• include_answer (bool): Whether the API generated answers are included

• include_raw_content (bool): Whether the webpage contains the original HTML content

• include_images (bool): Whether relevant image information is included

• format_output (bool): Whether the search results are formatted

Etc

• **kwargs : Additional keyword factors. Can be used for future updates or special features of the API

Copy

# !pip install -qU langchain-teddynote

Copy

from langchain_teddynote.tools.tavily import TavilySearch

# Basic example
tavily_tool = TavilySearch()

# Example of using include_domains
# Search including only specific domains
tavily_tool_with_domains = TavilySearch(include_domains=["github.io", "naver.com"])

# Example of using exclude_domains
# Search excluding specific domains
tavily_tool_exclude = TavilySearch(exclude_domains=["ads.com", "spam.com"])

# Search examples using different parameters
result1 = tavily_tool.search(
    query="Tell me about YouTuber Teddy Note", #search query
    search_depth="advanced",  # Advanced search level
    topic="general",  # General Topics
    days=7,  # Results within the last 7 days
    max_results=10,  # Up to 10 results
    include_answer=True,  # Including answers
    include_raw_content=True,  # I am a girl
    include_images=True,  # Including images
    format_output=True,  # Formatting results
)

# News search example
result2 = tavily_tool.search(
    query="Latest AI Technology Trends", #search query
    search_depth="basic",  # Basic search level
    topic="news",  # News Topics
    days=3,  # Results within the last 3 days
    max_results=5,  # Maximum 5 results
    include_answer=False,  # Answer not included
    include_raw_content=False,  # Original content not included
    include_images=False,  # Image not included
    format_output=True,  # Formatting results
)

# Example of a search that includes a specific domain
result3 = tavily_tool_with_domains.search(
    query=Python Programming Tips", # Search Query
    search_depth="advanced",  # Advanced search level
    max_results=3,  # Maximum 3 results
)

# Example of excluding specific domains from search
result4 = tavily_tool_exclude.search(
    query="healthy diet", #search query
    search_depth="basic",  # Basic search level
    days=30,  # Results within the last 30 days
    max_results=7,  # Up to 7 results

)

# Output the results
print("Basic search results:", result1)
print("News search results:", result2)
print("Search results including specific domains:", result3)
print("Search results excluding specific domains:", result4)

Image creation tool (DALL-E)

• DallEAPIWrapper 클래스 : Wrapper for OpenAI's DALL-E image generator.

With this tool, you can easily integrate the DALL-E API to implement text-based image creation capabilities. Various setup options allow you to leverage it as a flexible and powerful image creation tool.

Main properties

• model : DALL-E model name to use (default: "dall-e-2", "dall-e-3")

• n : Number of images to create (default: 1)

• size : Image size to create

• "dall-e-2": "1024x1024", "512x512", "256x256"

• "dall-e-3": "1024x1024", "1792x1024", "1024x1792"

• style : Style of the image to be created (default: "natural", "vivid")

• quality : Quality of the image to be created (default: "standard", "hd")

• max_retries : Maximum number of retries at creation

Main features

• Create images based on text descriptions using the DALL-E API

Flow cleanup

Here is an example of creating an image using DALL-E Image Generator.

this time DallEAPIWrapper Let's create an image using.

At this time, the input prompt asks the LLM model to create a prompt that creates an image.

Copy

from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import PromptTemplate
from langchain_openai import ChatOpenAI

# Initializing the ChatOpenAI model
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0.9, max_tokens=1000)

# Defining a prompt template for generating DALL-E images
prompt = PromptTemplate.from_template(
    "Generate a detailed IMAGE GENERATION prompt for DALL-E based on the following description. "
    "Return only the prompt, no intro, no explanation, no chatty, no markdown, no code block, no nothing. Just the prompt"
    "Output should be less than 1000 characters. Write in English only."
    "Image Description: \n{image_desc}",
)

# Create a chain that connects the prompt, LLM, and output parsers
chain = prompt | llm | StrOutputParser()

# Chain execution
image_prompt = chain.invoke(
    {"image_desc": "A satirist of people looking at their smartphones neo-classicism painting"}
)

# Output image prompt
print(image_prompt)

Copy

 Create a neo-classical painting that satirically depicts a group of people intently gazing at their smartphones. The scene should be be set in a grand, classic architectural setting with marble columns and intricate details reminiscent of ancient Greece. The figures, dressed in elegant, flowing garments typical of neo-classical art, exhibit exaggerated expressions of fascination and distraction as they interact with their devices. Incorporate elements such as classic scrapsures or frescoes in the background that context with the modernity of the smartphones. Use soft, natural lighting to enhance the scene, and include subtle hints of irony, perhaps by depicting traditional activities or conversations happy just out of focus, symbolizing the disconnect in modern life.

So, the image prompt you created earlier DallEAPIWrapper Let's create an image by typing in.

Copy

# DALL-E API 래퍼 가져오기
from langchain_community.utilities.dalle_image_generator import DallEAPIWrapper
from IPython.display import Image

# Initializing the DALL-E API wrapper
# model: DALL-E model version to use
# size: Image size to create
# quality: image quality
# n: number of images to generate

dalle = DallEAPIWrapper(model="dall-e-3", size="1024x1024", quality="standard", n=1)

# 질문
query = "A satirist of people looking at their smartphones neo-classicism painting"

# Create an image and get its URL
# Convert image descriptions to DALL-E prompts using chain.invoke()
# Generating the actual image using dalle.run()
image_url = dalle.run(chain.invoke({"image_desc": query}))

# Displays the generated image.
Image(url=image_url, width=500)

Custom Tool

In addition to the built-in tools provided by LangChain, users can define and use the tools themselves.

To do this langchain.tools Provided by the module tool Convert functions to tools using decorators.

@tool decorator

This decorator provides the ability to convert functions into tools. Various options allow you to customize the behavior of the tool.

How to use

1. Above function @tool Decorator application

2. Set decorator parameters as needed

With this decorator, you can easily convert ordinary Python functions into powerful tools, and create automated documentation and flexible interfaces.

Copy

from langchain.tools import tool


# Use decorators to transform functions into tools.
@tool
def add_numbers(a: int, b: int) -> int:
    """Add two numbers"""
    return a + b


@tool
def multiply_numbers(a: int, b: int) -> int:
    """Multiply two numbers"""
    return a * b

Copy

# Run the tool
add_numbers.invoke({"a": 3, "b": 4})

Copy

 7 

Copy

# Run the tool
multiply_numbers.invoke({"a": 3, "b": 4})

Copy

 12 

Google News Agency Search Tool

langchain-teddynote Packaged GoogleNews A tool that uses tools to search for Google news agencies.

Reference

• No API key required. (Because I use the RRSS feed)

A tool to search for news articles provided by news.google.com.

Explanation

• Search for the latest news using the Google News Search API.

• You can search for news based on keywords.

• Search for the latest news.

Main parameters

• k (int): Maximum number of search results to return (default: 5)

Please update the package before use.

Copy

# !pip install -qU langchain-teddynote

Copy

from langchain_teddynote.tools import GoogleNews

# Create a tool
news_tool = GoogleNews()

Copy

# Search Latest News
news_tool.search_latest(k=5)

Copy

 [{'url':'https://news.google.com/rss/articles/CBMid0FVX3lxTE82ZXF0N1g0Tmc0SjdGcTN4T0ZJWE9UTXRJdDJxb0VxVUhUb2t5elFjWjlvcnZJMDZFWVpDMVBsdnozUlRKbXJkSGJSTTJITkF5bTBLekJyN0dYV3E1UDhMN0FQSWo0WGsxMW1Hay1GeWZfaHhkTXVZ0gF3QVVfeXFMTkg4WXh0Xy15VEV2YzRKLUljZEwzZzFlajV1bmdmZ2MzcnRvaGJGbmxmM3gtZFp5a2c4a3Q2cmxPUjNNRTFDdmtzN2h4NnNSUHk3ME9zY3NldmRwWlZqd1UzdzVJOHBwbUxQVTJCdnVrMHhpLTYwdW8?oc=5','content': "'conditional leave' approval.. compounding and paid - MBC news" }, {'url':'https://news.google.com/rss/articles/CBMid0FVX3lxTE5pWEhEY0Jxd08zQjZ0N3ZySEVDSThRN0NLd2YzV3BRdTZGdmpxbXdpMmZmeE12VWc4cmdQWlJYWGpHMnpua044Y3FmRVZMcEsyU0NsajgycjVlNWI4cU5IRkpGT1ZZbS02LWdpbFhLRVREUGdfQXBV0gFmQVVfeXFMTmU0aU5WV25HTldZSS1JZHVmQmdYZk9PcmtWbXFzNGtLYVRQUmVtMmVHcXN3RHlTem5maENOVFNsYXhlQTFHb0NJM3RsSzg1a1FqeWMyRVJ1Mmp4MUFLZTBRbllyY0FB?oc=5','  content':'‘ Han Dong-hoon attack Raju suspicion ’ Kim Dae-nam, SGI Seoul-boa-jung-East-bo' }, {'url':'https://news.google.com/rss/articles/CBMikAFBVV95cUxQWFRFelJKZTdwSjVDRFloNUpNbF83dVVLOXpYSnZ3Y25BS25jQUFWdFhuYklic0lQT0VwbWw0WEU5MGs2Wm83Y1lQLTBJakRrQTFFTEp0T09tNzY3NVpyVGhyRjdpQ2FNV3pVOVZZaFpwNEl1Q2Rtb1laa2VDY1o0NFRxZXFPOWFyUmF3X0NLa1jSAaQBQVVfeXFMTWJJX040anAtYl9lbzRidURVSDNmMGFlc0t4eHZDa1ZJWk14Z0tzNDd4dzBRTFVyZm5HTFEtMHFSajhkci1PUmlYTTdoUC11RUdNMDhWaW9JakszTlhDWkdmNFc2Vk9BUXVoZEt4dG1qOWVicUs1dXd6cWdtSHh6T2ZpcU9CQXFvMHp3ZElpQ1dZYUloT1NpUDVNMTRGcFNmeDBPVlQ?oc=5','GTA' Han'Philippines FTA takes effect When it comes into effect, it will expand dramatically.  com/rss/articles/CBMickFVX3lxTE1Pd3RtbzBFOEpERWlqYk5ISnIzaldRQ21XdHcyaW5zMWpQYmNWR2ZveHVLRExncDhNX2oweTBFZlhrUV90OVVodE9xaVhqVFc0S1BFWUs5VFFILUExTVdpWE10QUlKbnBrSEgtSkN6Skl3UQ?oc=5','content':'Mundahye's blood alcohol level 0.149%... Hang on and move to the export station.-Handle' }, {'url':'https://news.google.com/rss/articles/CBMickFVX3lxTE9HTU1JajFMeVRPZkViX2EwZkc3R05ySDdWbEJhdjN0QzNnTmo2V0RMbjZQcFI1YTl6dnVHMlpKU0RsODNuRW5TeWYwczBhNjdkNkhiRzc3cWtzQ0wtaHkwU1MwYjNQc2tLX3RjZjdhbEZpQQ?oc=5','content':'[Single] Even after Kim Gun-hee's marriage ‘Deutsche Stock  com/rss/articles/CBMickFVX3lxTE9HTU1JajFMeVRPZkViX2EwZkc3R05ySDdWbEJhdjN0QzNnTmo2V0RMbjZQcFI1YTl6dnVHMlpKU0RsODNuRW5TeWYwczBhNjdkNkhiRzc3cWtzQ0wtaHkwU1MwYjNQc2tLX3RjZjdhbEZpQQ?oc=5','content':'[Single] Even after Kim's marriage, ‘Deutsche Stock Buyer ’ Jeong Hwang... Shaking Yoon Hae-Korean'}]  com/rss/articles/CBMickFVX3lxTE9HTU1JajFMeVRPZkViX2EwZkc3R05ySDdWbEJhdjN0QzNnTmo2V0RMbjZQcFI1YTl6dnVHMlpKU0RsODNuRW5TeWYwczBhNjdkNkhiRzc3cWtzQ0wtaHkwU1MwYjNQc2tLX3RjZjdhbEZpQQ?oc=5','content':'[Single] Even after Kim's marriage, ‘Deutsche Stock Buyer ’ Jeong Hwang... Shaking Yoon Hae-Korean'}] 

Copy

# Search news by keyword
news_tool.search_by_keyword("AI 투자", k=5)

Copy

 [{'url':'https://news.google.com/rss/articles/CBMiU0FVX3lxTFB1VDRjTlIxeHRjV3NKaERHMFJLOFNpUlNUVnZzQ0JvOUpISVE2cW9rbTZPVU9lQXZuUURJNHF0aDZyaDdQM0F5NFhsc0NrN1dHNEZz?oc=5','content':'Global CEOs "In the next three years, AI investments,·, and more employees will be selected" -Nate News'}, {'url':'https://news.google.com/rss/articles/CBMiakFVX3lxTE9BQUtjd3pDS1dXRkZQTlo4Y2w2d2laZFFQRXdsbGlCMFRTcWJYLTNLaksxbFlnYzk3clVnYlFxZF93U2xubXY1VVhBTDQ2SDhLbm9hcDdXQ3prTlJtV0VCYTFINmNRbjR5N3c?oc=5','content':'MIT Economist "Only 5 years to be affected by AI.Companies will only fly their investment costs” -AI times' }, {'url':'https://news.google.com/rss/articles/CBMiW0FVX3lxTE9oenZObVAzOV9hbUliNGhkLTBXZXZrTGpvT0ljcXlpWE5leld2b3RLbUJFTFRtZnU4c0VFaC1oYkVvTTdESEZCYmZUTDJSMFFYdGVXeXN3OVdRTEE?oc=5','content':'MS,  AI infrastructure 640 billion investment decision in Italy -Union News' }, {'url':'https://news.google.com/rss/articles/CBMiYEFVX3lxTE5XdlFLT29uWXR5SmRpS054Z2J3VldjZlJETHJIeVRPakFUNEdiOHh0N0hOU21BRTlHN1haRTBUYnlKUEdVdXJWUWdTbG9vczFHeDNPRG1hSkc2NzVRaVBLQQ?oc=5','content':'Global CEO 64% “Priority investment in AI” -Asian economy'}, {'url':'https://news.google.com/rss/articles/CBMiZEFVX3lxTE02VVREUUxSRExwUkxkY0NGZEw0Z0x0N3FHenZvTXN4TEVlRHEzWEF0SGo4N1Y2Zm8xTE5TaGZDZzNubzFrTmxoUVRiNG9WN01wY2hyUGVyZ1lCS05ua1FLWXYwRTk?oc=5','content:Early 5 years opportunity ” [Herald Money Festa 2024] -Herald Media' }]  Global CEO 64% “Priority investment in AI ” -Asian economy' }, {'url':'https://news.google.com/rss/articles/CBMiZEFVX3lxTE02VVREUUxSRExwUkxkY0NGZEw0Z0x0N3FHenZvTXN4TEVlRHEzWEF0SGo4N1Y2Zm8xTE5TaGZDZzNubzFrTmxoUVRiNG9WN01wY2hyUGVyZ1lCS05ua1FLWXYwRTk?oc=5','content':'“AI era investment keyword is ‘shortage of supply ’...Early 5 years opportunity ” [Herald Money Festa 2024] -Herald Media' }]  Global CEO 64% “Priority investment in AI ” -Asian economy' }, {'url':'https://news.google.com/rss/articles/CBMiZEFVX3lxTE02VVREUUxSRExwUkxkY0NGZEw0Z0x0N3FHenZvTXN4TEVlRHEzWEF0SGo4N1Y2Zm8xTE5TaGZDZzNubzFrTmxoUVRiNG9WN01wY2hyUGVyZ1lCS05ua1FLWXYwRTk?oc=5','content':'“AI era investment keyword is ‘shortage of supply ’...Early 5 years opportunity ” [Herald Money Festa 2024] -Herald Media' }] 

Copy

from langchain_teddynote.tools import GoogleNews
from langchain.tools import tool
from typing import List, Dict


# Create a tool to search news by keyword
@tool
def search_keyword(query: str) -> List[Dict[str, str]]:
    """Look up news by keyword"""
    print(query)
    news_tool = GoogleNews()
    return news_tool.search_by_keyword(query, k=5)

Copy

# Execution results
search_keyword.invoke({"query": "AI 투자"})

Copy

 AI Investment

Copy

 [{'url':'https://news.google.com/rss/articles/CBMiU0FVX3lxTFB1VDRjTlIxeHRjV3NKaERHMFJLOFNpUlNUVnZzQ0JvOUpISVE2cW9rbTZPVU9lQXZuUURJNHF0aDZyaDdQM0F5NFhsc0NrN1dHNEZz?oc=5','content':'Global CEOs "In the next three years, AI investments,·, and more employees will be selected" -Nate News'}, {'url':'https://news.google.com/rss/articles/CBMiakFVX3lxTE9BQUtjd3pDS1dXRkZQTlo4Y2w2d2laZFFQRXdsbGlCMFRTcWJYLTNLaksxbFlnYzk3clVnYlFxZF93U2xubXY1VVhBTDQ2SDhLbm9hcDdXQ3prTlJtV0VCYTFINmNRbjR5N3c?oc=5','content':'MIT Economist "Only 5 years to be affected by AI.Companies will only fly their investment costs” -AI times' }, {'url':'https://news.google.com/rss/articles/CBMiW0FVX3lxTE9oenZObVAzOV9hbUliNGhkLTBXZXZrTGpvT0ljcXlpWE5leld2b3RLbUJFTFRtZnU4c0VFaC1oYkVvTTdESEZCYmZUTDJSMFFYdGVXeXN3OVdRTEE?oc=5','content':'MS,  AI infrastructure 640 billion investment decision in Italy -Union News' }, {'url':'https://news.google.com/rss/articles/CBMiYEFVX3lxTE5XdlFLT29uWXR5SmRpS054Z2J3VldjZlJETHJIeVRPakFUNEdiOHh0N0hOU21BRTlHN1haRTBUYnlKUEdVdXJWUWdTbG9vczFHeDNPRG1hSkc2NzVRaVBLQQ?oc=5','content':'Global CEO 64% “Priority investment in AI” -Asian economy'}, {'url':'https://news.google.com/rss/articles/CBMiZEFVX3lxTE02VVREUUxSRExwUkxkY0NGZEw0Z0x0N3FHenZvTXN4TEVlRHEzWEF0SGo4N1Y2Zm8xTE5TaGZDZzNubzFrTmxoUVRiNG9WN01wY2hyUGVyZ1lCS05ua1FLWXYwRTk?oc=5','content:Early 5 years opportunity ” [Herald Money Festa 2024] -Herald Media' }]  Global CEO 64% “Priority investment in AI ” -Asian economy' }, {'url':'https://news.google.com/rss/articles/CBMiZEFVX3lxTE02VVREUUxSRExwUkxkY0NGZEw0Z0x0N3FHenZvTXN4TEVlRHEzWEF0SGo4N1Y2Zm8xTE5TaGZDZzNubzFrTmxoUVRiNG9WN01wY2hyUGVyZ1lCS05ua1FLWXYwRTk?oc=5','content':'“AI era investment keyword is ‘shortage of supply ’...Early 5 years opportunity ” [Herald Money Festa 2024] -Herald Media' }]  Global CEO 64% “Priority investment in AI ” -Asian economy' }, {'url':'https://news.google.com/rss/articles/CBMiZEFVX3lxTE02VVREUUxSRExwUkxkY0NGZEw0Z0x0N3FHenZvTXN4TEVlRHEzWEF0SGo4N1Y2Zm8xTE5TaGZDZzNubzFrTmxoUVRiNG9WN01wY2hyUGVyZ1lCS05ua1FLWXYwRTk?oc=5','content':'“AI era investment keyword is ‘shortage of supply ’...Early 5 years opportunity ” [Herald Money Festa 2024] -Herald Media' }]