How to Intermediate · 3 min read

How to give AI agent web search capability

Quick answer

To give an AI agent web search capability, integrate a web search API (like Bing Search or Google Custom Search) to fetch real-time results, then feed those results as context to a large language model (LLM) such as gpt-4o or claude-3-5-sonnet-20241022. This enables the agent to answer queries with fresh, relevant information beyond its training data.

PREREQUISITES

Python 3.8+
OpenAI API key (free tier works)
pip install openai>=1.0
Access to a web search API (e.g., Bing Search API or Google Custom Search API)
pip install requests

Setup

Install necessary Python packages and set environment variables for your OpenAI API key and web search API key.

Use pip install openai requests to install dependencies.
Set OPENAI_API_KEY and SEARCH_API_KEY in your environment.

bash

pip install openai requests

Step by step

This example shows how to query a web search API, then pass the search results as context to an LLM to generate an informed answer.

python

import os
import requests
from openai import OpenAI

# Initialize OpenAI client
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

# Function to perform web search using Bing Search API
# Replace endpoint and headers as per your search API provider

def web_search(query):
    endpoint = "https://api.bing.microsoft.com/v7.0/search"
    headers = {"Ocp-Apim-Subscription-Key": os.environ["SEARCH_API_KEY"]}
    params = {"q": query, "count": 3}
    response = requests.get(endpoint, headers=headers, params=params)
    response.raise_for_status()
    results = response.json()
    snippets = []
    for item in results.get("webPages", {}).get("value", []):
        snippets.append(item["snippet"])
    return "\n".join(snippets)

# Compose prompt with search results

def create_prompt(query, search_results):
    return f"You are an AI assistant with access to recent web search results.\nSearch results:\n{search_results}\n\nAnswer the question based on the above information:\n{query}"

# Main function

def main():
    user_query = "What are the latest advancements in AI for 2026?"
    search_results = web_search(user_query)
    prompt = create_prompt(user_query, search_results)

    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": prompt}]
    )
    print("AI agent answer:")
    print(response.choices[0].message.content)

if __name__ == "__main__":
    main()

output

AI agent answer:
The latest advancements in AI for 2026 include breakthroughs in large language models, improved multimodal understanding, and more efficient training techniques...

Common variations

You can adapt this pattern by:

Using claude-3-5-sonnet-20241022 or gemini-1.5-pro instead of gpt-4o.
Implementing async calls with httpx or aiohttp for faster search and LLM requests.
Streaming LLM responses for real-time output.
Using different search APIs like Google Custom Search or SerpAPI depending on your needs.

Troubleshooting

If you get HTTP 401 errors, verify your API keys and environment variables.
If search results are empty, check your query parameters and API usage limits.
For slow responses, consider caching search results or using async requests.
If the LLM ignores search context, ensure the prompt clearly instructs it to use the provided information.

✅

Key Takeaways

Integrate a web search API to fetch real-time data for your AI agent.
Feed search results as context in the prompt to an LLM like gpt-4o for up-to-date answers.
Use environment variables for all API keys to keep credentials secure.
Consider async and streaming for better performance and user experience.
Clear prompt instructions ensure the LLM uses the search data effectively.

Verified 2026-04 · gpt-4o, claude-3-5-sonnet-20241022, gemini-1.5-pro

Verify ↗