web search agent

web_search_agent.py

from datetime import date
from time import sleep

from selenium import webdriver
from selenium.webdriver.common.by import By
from selenium.webdriver.support import expected_conditions as EC
from selenium.webdriver.support.ui import WebDriverWait

from agents import (Agent, AsyncOpenAI, OpenAIChatCompletionsModel, Runner,
                    function_tool, set_tracing_disabled)
from agents.model_settings import ModelSettings


# disable tracing 
set_tracing_disabled(disabled=True)

# Qwen3 model running on Ollama
qwen_client = AsyncOpenAI(base_url="http://localhost:11434/v1", api_key="secret_key")
qwen3_model = OpenAIChatCompletionsModel(model="qwen3", openai_client=qwen_client)

# web search tool
@function_tool
def web_search(query: str, number_of_pages_to_search: int = 1) -> dict:
    """Search the web using a query"""
    print(f"Search Query: {query}, num of pages to search: {number_of_pages_to_search}")

    op = webdriver.ChromeOptions()
    # op.add_argument("headless")
    dr = webdriver.Chrome(keep_alive=False, options=op)
    dr.get(f"https://duckduckgo.com/?q={query}&ia=web")

    WebDriverWait(dr, 10).until(
        EC.presence_of_element_located(
            (By.XPATH, "//ol[contains(@class, 'react-results--main')]")
        )
    )

    # get all the organic results, skipping ads results or promos
    results = dr.find_elements(By.XPATH, "//li[contains(@data-layout, 'organic')]")
    # print(results)

    # for the first 5 results
    page_results = {}
    for i in range(number_of_pages_to_search):
        # get the results in page
        sleep(2)
        a_tags = [a for a in results[i].find_elements(By.TAG_NAME, 'a') if a.text]
        url = a_tags[0].get_attribute("href")
        print(f"Getting details from page: {url}")
        a_tags[0].click()

        # get the text in the page
        sleep(2)
        p_tags = dr.find_elements(By.TAG_NAME, "p")
        text_in_page = ''.join([p.text for p in p_tags])

        page_results[url] = text_in_page
        
        dr.back()

    dr.close()

    # print(page_results)
    return page_results
  
  
INSTRUCTIONS = f"""You are an expert at researching topics.
Given a search query, you search the web for that query (atleast 3 pages or less) and produce a concise summary of the results.
The summary must be 2-3 paragraphs and less than 300 words. Capture the main points and ignore any fluff.
Do not include any additional commentary other than the summary itself.

The current date is {date.today().strftime("%m/%d/%Y")}
"""

# research agent with tools
search_agent = Agent(
    name="Search agent",
    instructions=INSTRUCTIONS,
    tools=[web_search],
    model=qwen3_model,
    model_settings=ModelSettings(tool_choice="required"),
)

# run the agent to get the result summary
message = "Latest trends in robotics and its applications on the factory floor"
result = await Runner.run(search_agent, message)
print("Research Summary:\n\n", result.final_output)

Sysout:

Search Query: latest trends in robotics factory floor applications 2026, num of pages to search: 3
Getting details from page: https://apera.ai/learn/articles/2026-vision-checklist/
Getting details from page: https://ifr.org/ifr-press-releases/news/top-5-global-robotics-trends-2026
Getting details from page: https://oxmaint.com/industries/manufacturing-plant/top-manufacturing-plant-automation-trends-watch-2026
Research Summary:

In 2026, robotics on the factory floor is driven by AI integration, collaborative robots (cobots), and digital twins. Vision-guided robotics (VGR) with AI, such as 4D Vision, enables real-time adaptability to varying production conditions, critical for retrofitting legacy systems and handling complex tasks. Natural language interfaces are reducing total cost of ownership (TCO), while pre-engineered solutions simplify deployment. Humanoid robots like Hyundai’s Atlas are being tested in assembly lines, and cobots, now affordable for mid-sized plants, improve efficiency by 15–20% in sectors like food processing. Digital twins, projected to reach $34B, offer predictive maintenance and real-time optimization, cutting downtime by 20% and accelerating development cycles.

Cybersecurity threats loom large as interconnected systems expand attack surfaces, with incidents like Jaguar Land Rover’s $260M outage highlighting vulnerabilities. Predictive maintenance powered by AI sensors and CMMS (Computerized Maintenance Management Systems) is now a baseline, forecasting failures and reducing unplanned downtime by 45%. Autonomous Mobile Robots (AMRs), such as DHL’s 7,500-unit fleet, streamline logistics, while IT/OT convergence enhances versatility but demands strict security protocols.

The shift requires a maintenance-first strategy, pairing automation with centralized CMMS to track assets, automate work orders, and manage multi-site operations. Plants adopting phased automation—starting with digitizing maintenance and layering complexity—outperform those neglecting infrastructure. As labor shortages persist, robotics and AI are not just tools but allies, reshaping workplaces and necessitating upskilling programs. The industry’s success hinges on balancing innovation with robust oversight, ensuring automation’s ROI through seamless integration of technology and maintenance protocols.