Skip to main content
Open In ColabOpen on GitHub

desearch

Overview​

AI-Powered Search Engine and API for Advanced Data Discovery Desearch.

Welcome to the comprehensive guide on integrating Desearch with LangChain to enable advanced Retrieval-Augmented Generation (RAG) and agent-style workflows. Desearch is a cutting-edge, privacy-first search API that excels in delivering relevant, real-time web and Twitter content, specifically optimized for use in Large Language Model (LLM) applications.

First, get an Desearch API key and add it as an environment variable. Get $5 free credit (plus more by completing certain actions like making your first search) by signing up here.

Setup​

.env DESEARCH_API_KEY=your_api_key_here

from dotenv import load_dotenv

load_dotenv()
# Set your API key as an environment variable

And install the integration package

%pip install -U langchain-desearch

Instantiation​

🛠️ Tools Overview​

Desearch provides a suite of modular tools compatible with LangChain, each designed for specific use cases:

🔎 Search Tools​

  • DesearchTool: A comprehensive tool for searching across web, AI, and Twitter posts.
  • BasicWebSearchTool: A lightweight tool focused solely on web searches.
  • BasicTwitterSearchTool: A specialized tool for searching tweets with advanced filtering options.

Invocation​

🧪 Quick Examples​

Basic Usage​

Here's a simple example to demonstrate how to use the DesearchTool for web searches

from langchain_desearch.tools import DesearchTool

# Initialize the tool
tool = DesearchTool()

# Run a search query
response = tool._run(
    prompt="Bittensor network activity",
    tool=["web"],
    model="NOVA",
    date_filter="PAST_24_HOURS",
)

# Print the response
print(response)

Search Twitter​

To search Twitter for specific topics, use the BasicTwitterSearchTool

from langchain_desearch.tools import BasicTwitterSearchTool

# Initialize the Twitter search tool
tool = BasicTwitterSearchTool()

# Execute a search query
response = tool._run(query="AI governance", sort="Top", count=5)

# Display the results
print(response)

Use within an agent​

🤖 Building a RAG Chain​

Creating a RAG chain involves integrating Desearch with LangChain's prompt and LLM capabilities

from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate, PromptTemplate
from langchain_core.runnables import RunnableParallel, RunnablePassthrough
from langchain_deepseek import ChatDeepSeek
from langchain_desearch.tools import DesearchTool

# Initialize the Desearch tool
tool = DesearchTool()


# Define a function to fetch context
def fetch_context(query):
    return tool._run(prompt=query, tool="desearch_web", model="NOVA")


# Create a prompt template
prompt = ChatPromptTemplate.from_messages(
    [
        ("system", "You are a research assistant."),
        (
            "human",
            "Answer this using the provided context:\n\n<context>{context}</context>",
        ),
    ]
)

# Set up the LLM
llm = ChatDeepSeek(model="deepseek-chat", temperature=0.7)
parser = StrOutputParser()

# Build the RAG chain
chain = (
    RunnableParallel(
        {
            "query": RunnablePassthrough(),
            "context": lambda q: fetch_context(q["query"]),
        }
    )
    | prompt
    | llm
    | parser
)

# Invoke the chain with a query
result = chain.invoke("What is the latest update on the EU AI Act?")

🧠 Using Desearch in a LangChain Agent​

Integrate Desearch into a LangChain agent to automate complex workflows:

from langchain_deepseek import ChatDeepSeek
from langchain_desearch.agent import create_search_agent

# Initialize the LLM
llm = ChatDeepSeek(model="deepseek-chat", temperature=0.7)

# Create a search agent
agent = create_search_agent(llm=llm)

# Invoke the agent with an input message
response = agent.invoke({"input_message": "Summarize current AI regulation trends"})
print(response["output"])
API Reference:ChatDeepSeek

API reference​

For detailed API reference, please refer to the official documentation or the source code comments for each tool and method.

✅ Testing​

Unit Tests (Mocked)​

To ensure your integration works as expected, run unit tests:

%pytest tests/integration_tests/test_tools.py
%pytest tests/unit_tests/test_imports.py

This structure adds the required headers and organizes the content under them. You can adjust the content under each header as needed to better fit the specific details of your project.

Was this page helpful?