How to Add Web Search to LangChain Agents (Python)

Why Agents Need Web Search

LLMs have a knowledge cutoff. When a user asks about today's news, recent prices, or current documentation, the model either hallucinates or says "I don't know." Web search tools solve this by letting the agent retrieve real-time information from Google.

Prerequisites

bash
pip install langchain langchain-openai httpx

You'll also need a Searlo API key (free, 3,000 credits included).

Step 1: Create the Search Tool

python
import httpx
from langchain.tools import StructuredTool
from pydantic import BaseModel, Field

SEARLO_API_KEY = "your_api_key_here"

class SearchInput(BaseModel):
    query: str = Field(description="The search query")

def web_search(query: str) -> str:
    """Search the web using Searlo API and return results."""
    response = httpx.get(
        "https://api.searlo.tech/api/v1/search/web",
        params={"q": query, "limit": 5},
        headers={"X-API-Key": SEARLO_API_KEY},
        timeout=10.0,
    )
    data = response.json()
    results = data.get("organic", [])
    return "\n".join(
        f"[{r['position']}] {r['title']}\n{r['snippet']}\nURL: {r['link']}"
        for r in results
    )

search_tool = StructuredTool.from_function(
    func=web_search,
    name="web_search",
    description="Search Google for real-time information. Use this when you need current data, news, or facts you're not sure about.",
    args_schema=SearchInput,
)

Step 2: Create the Agent

python
from langchain_openai import ChatOpenAI
from langchain.agents import AgentExecutor, create_openai_functions_agent
from langchain.prompts import ChatPromptTemplate, MessagesPlaceholder

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

prompt = ChatPromptTemplate.from_messages([
    ("system", "You are a helpful research assistant. Use the web_search tool to find current information when needed. Always cite your sources with URLs."),
    ("human", "{input}"),
    MessagesPlaceholder("agent_scratchpad"),
])

agent = create_openai_functions_agent(llm, [search_tool], prompt)
executor = AgentExecutor(agent=agent, tools=[search_tool], verbose=True)

Step 3: Run It

python
result = executor.invoke({
    "input": "What are the latest developments in quantum computing this week?"
})
print(result["output"])

The agent will:

1. Recognize it needs current information

2. Call web_search with an appropriate query

3. Read the results

4. Synthesize a response with source citations

Tips for Production

• Set a timeout: Searlo's web search has a median of about 1 second and a p90 of about 2.5 seconds (measured September 2026), and the slowest calls take several seconds. Use a 10-second timeout.

• Limit results: limit=5 is usually enough (Searlo ignores num). More results cost the same (1 credit per search) but add more tokens to the LLM context.

• Use TOON format: Add &format=toon to the API call to get token-optimized output that's 60% smaller — saves money on LLM calls.

• Cache frequent queries: If multiple users ask similar questions, cache search results for 5-10 minutes.

Next Steps

• See the full Integrations page for CrewAI, LlamaIndex, and n8n examples

• Learn about MCP protocol for zero-code AI tool integration

• Read about SERP API for AI agents for more architecture patterns