Warning: The langchain_community.tools.tavily_search.tool is deprecated. While it remains functional for now, we strongly recommend migrating to the new langchain-tavily Python package which supports both Search and Extract functionality and receives continuous updates with the latest features.

The langchain-tavily Python package is the offical LangChain integration of Tavily, inlcuding both TavilySearch and TavilyExtract

​
Installation

pip install -U langchain-tavily

​
Credentials

We also need to set our Tavily API key. You can get an API key by visiting this site and creating an account.

import getpass
import os

if not os.environ.get("TAVILY_API_KEY"):
    os.environ["TAVILY_API_KEY"] = getpass.getpass("Tavily API key:\n")

Here we show how to instantiate an instance of the Tavily search tool. The tool accepts various parameters to customize the search. After instantiation we invoke the tool with a simple query. This tool allows you to complete search queries using Tavily’s Search API endpoint.

​
Instantiation

The tool accepts various parameters during instantiation:

  • max_results (optional, int): Maximum number of search results to return. Default is 5.
  • topic (optional, str): Category of the search. Can be “general”, “news”, or “finance”. Default is “general”.
  • include_answer (optional, bool): Include an answer to original query in results. Default is False.
  • include_raw_content (optional, bool): Include cleaned and parsed HTML of each search result. Default is False.
  • include_images (optional, bool): Include a list of query related images in the response. Default is False.
  • include_image_descriptions (optional, bool): Include descriptive text for each image. Default is False.
  • search_depth (optional, str): Depth of the search, either “basic” or “advanced”. Default is “basic”.
  • time_range (optional, str): The time range back from the current date to filter results - “day”, “week”, “month”, or “year”. Default is None.
  • include_domains (optional, List[str]): List of domains to specifically include. Default is None.
  • exclude_domains (optional, List[str]): List of domains to specifically exclude. Default is None.

For a comprehensive overview of the available parameters, refer to the Tavily Search API documentation

from langchain_tavily import TavilySearch

tool = TavilySearch(
    max_results=5,
    topic="general",
    # include_answer=False,
    # include_raw_content=False,
    # include_images=False,
    # include_image_descriptions=False,
    # search_depth="basic",
    # time_range="day",
    # include_domains=None,
    # exclude_domains=None
)

​
Invoke directly with args

The Tavily search tool accepts the following arguments during invocation:

  • query (required): A natural language search query
  • The following arguments can also be set during invokation : include_images, search_depth , time_range, include_domains, exclude_domains, include_images
  • For reliability and performance reasons, certain parameters that affect response size cannot be modified during invocation: include_answer and include_raw_content. These limitations prevent unexpected context window issues and ensure consistent results.

NOTE: The optional arguments are available for agents to dynamically set, if you set a argument during instantiation and then invoke the tool with a different value, the tool will use the value you passed during invokation.

# Basic query
tool.invoke({"query": "What happened at the last wimbledon"})

output:

{
 'query': 'What happened at the last wimbledon',
 'follow_up_questions': None,
 'answer': None,
 'images': [],
 'results':
 [{'url': 'https://en.wikipedia.org/wiki/Wimbledon_Championships',
   'title': 'Wimbledon Championships - Wikipedia',
   'content': 'Due to the COVID-19 pandemic, Wimbledon 2020 was cancelled ...',
   'score': 0.62365627198,
   'raw_content': None},
    ......................................................................
    {'url': 'https://www.cbsnews.com/news/wimbledon-men-final-carlos-alcaraz-novak-djokovic/',
    'title': "Carlos Alcaraz beats Novak Djokovic at Wimbledon men's final to ...",
    'content': 'In attendance on Sunday was Catherine, the Princess of Wales ...',
    'score': 0.5154731446,
    'raw_content': None}],
  'response_time': 2.3
}

​
Agent Tool Calling

We can use our tools directly with an agent executor by binding the tool to the agent. This gives the agent the ability to dynamically set the available arguments to the Tavily search tool.

In the below example when we ask the agent to find “What is the most popular sport in the world? include only wikipedia sources” the agent will dynamically set the argments and invoke Tavily search tool : Invoking tavily_search with {'query': 'most popular sport in the world', 'include_domains': ['wikipedia.org'], 'search_depth': 'basic'}

# !pip install -qU langchain langchain-openai langchain-tavily
from typing import Any, Dict, Optional
import datetime

from langchain.agents import create_openai_tools_agent, AgentExecutor
from langchain.chat_models import init_chat_model
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_openai import ChatOpenAI
from langchain_tavily import TavilySearch
from langchain.schema import HumanMessage, SystemMessage

# Initialize LLM
llm = init_chat_model(model="gpt-4o", model_provider="openai", temperature=0)

# Initialize Tavily Search Tool
tavily_search_tool = TavilySearch(
    max_results=5,
    topic="general",
)

# Set up Prompt with 'agent_scratchpad'
today = datetime.datetime.today().strftime("%D")
prompt = ChatPromptTemplate.from_messages([
    ("system", f"""You are a helpful reaserch assistant, you will be given a query and you will need to
    search the web for the most relevant information. The date today is {today}."""),
    MessagesPlaceholder(variable_name="messages"),
    MessagesPlaceholder(variable_name="agent_scratchpad"),  # Required for tool calls
])

# Create an agent that can use tools
agent = create_openai_tools_agent(
    llm=llm,
    tools=[tavily_search_tool],
    prompt=prompt
)

# Create an Agent Executor to handle tool execution
agent_executor = AgentExecutor(agent=agent, tools=[tavily_search_tool], verbose=True)

user_input =  "What is the most popular sport in the world? include only wikipedia sources"

# Construct input properly as a dictionary
response = agent_executor.invoke({"messages": [HumanMessage(content=user_input)]})

​
Tavily Extract

Here we show how to instantiate an instance of the Tavily extract tool. After instantiation we invoke the tool with a list of URLs. This tool allows you to extract content from URLs using Tavily’s Extract API endpoint.

​
Instantiation

The tool accepts various parameters during instantiation:

  • extract_depth (optional, str): The depth of the extraction, either “basic” or “advanced”. Default is “basic ”.
  • include_images (optional, bool): Whether to include images in the extraction. Default is False.

For a comprehensive overview of the available parameters, refer to the Tavily Extract API documentation

from langchain_tavily import TavilyExtract

tool = TavilyExtract(
    extract_depth="advanced",
    include_images=False,
)

​
Invoke directly with args

The Tavily extract tool accepts the following arguments during invocation:

  • urls (required): A list of URLs to extract content from.
  • Both extract_depth and include_images can also be set during invokation

NOTE: The optional arguments are available for agents to dynamically set, if you set a argument during instantiation and then invoke the tool with a different value, the tool will use the value you passed during invokation.

# Extract content from a URL
result = tool.invoke({
    "urls": ["https://en.wikipedia.org/wiki/Lionel_Messi"]
})

output:

{
    'results': [{
        'url': 'https://en.wikipedia.org/wiki/Lionel_Messi',
        'raw_content': 'Lionel Messi\nLionel Andrés "Leo" Messi...',
        'images': []
    }],
    'failed_results': [],
    'response_time': 0.79
}

​
Tavily Research Agent

This example demonstrates how to build a powerful web research agent using Tavily’s search and extract Langchain tools.

​
Features

  • Internet Search: Query the web for up-to-date information using Tavily’s search API
  • Content Extraction: Extract and analyze specific content from web pages
  • Seamless Integration: Works with OpenAI’s function calling capability for reliable tool use
# !pip install -qU langchain langchain-openai langchain-tavily
from typing import Any, Dict, Optional
import datetime

from langchain.agents import create_openai_tools_agent, AgentExecutor
from langchain.chat_models import init_chat_model
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_openai import ChatOpenAI
from langchain_tavily import TavilySearch, TavilyExtract
from langchain.schema import HumanMessage, SystemMessage

# Initialize LLM
llm = ChatOpenAI(temperature=0, model="gpt-4o")

# Initialize Tavily Search Tool
tavily_search_tool = TavilySearch(
    max_results=5,
    topic="general",
)
# Initialize Tavily Extract Tool
tavily_extract_tool = TavilyExtract()

tools = [tavily_search_tool, tavily_extract_tool]

# Set up Prompt with 'agent_scratchpad'
today = datetime.datetime.today().strftime("%D")
prompt = ChatPromptTemplate.from_messages([
    ("system", f"""You are a helpful reaserch assistant, you will be given a query and you will need to
    search the web for the most relevant information then extract content to gain more insights. The date today is {today}."""),
    MessagesPlaceholder(variable_name="messages"),
    MessagesPlaceholder(variable_name="agent_scratchpad"),  # Required for tool calls
])
# Create an agent that can use tools
agent = create_openai_tools_agent(
    llm=llm,
    tools=tools,
    prompt=prompt
)

# Create an Agent Executor to handle tool execution
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True)

user_input =  "Research the latest developments in quantum computing and provide a detailed summary of how it might impact cybersecurity in the next decade."

# Construct input properly as a dictionary
response = agent_executor.invoke({"messages": [HumanMessage(content=user_input)]})