![[The AI Show Episode 144]: ChatGPT’s New Memory, Shopify CEO’s Leaked “AI First” Memo, Google Cloud Next Releases, o3 and o4-mini Coming Soon & Llama 4’s Rocky Launch](https://www.marketingaiinstitute.com/hubfs/ep%20144%20cover.png)
![[DEALS] The All-in-One Microsoft Office Pro 2019 for Windows: Lifetime License + Windows 11 Pro Bundle (89% off) & Other Deals Up To 98% Off](https://www.javacodegeeks.com/wp-content/uploads/2012/12/jcg-logo.jpg){kind=logo}
_Andreas_Prott_Alamy.jpg?width=1280&auto=webp&quality=80&disable=upscale#)
![What features do you get with Gemini Advanced? [April 2025]](https://i0.wp.com/9to5google.com/wp-content/uploads/sites/4/2024/02/gemini-advanced-cover.jpg?resize=1200%2C628&quality=82&strip=all&ssl=1)
![Apple Shares Official Trailer for 'Long Way Home' Starring Ewan McGregor and Charley Boorman [Video]](https://www.iclarified.com/images/news/97069/97069/97069-640.jpg)
![Apple Watch Series 10 Back On Sale for $299! [Lowest Price Ever]](https://www.iclarified.com/images/news/96657/96657/96657-640.jpg)
![EU Postpones Apple App Store Fines Amid Tariff Negotiations [Report]](https://www.iclarified.com/images/news/97068/97068/97068-640.jpg)
An Advanced Coding Implementation: Mastering Browser‑Driven AI in Google Colab with Playwright, browser_use Agent & BrowserContext, LangChain, and Gemini
In this tutorial, we will learn how to harness the power of a browser‑driven AI agent entirely within Google Colab. We will utilize Playwright’s headless Chromium engine, along with the browser_use library’s high-level Agent and BrowserContext abstractions, to programmatically navigate websites, extract data, and automate complex workflows. We will wrap Google’s Gemini model via the […] The post An Advanced Coding Implementation: Mastering Browser‑Driven AI in Google Colab with Playwright, browser_use Agent & BrowserContext, LangChain, and Gemini appeared first on MarkTechPost.
In this tutorial, we will learn how to harness the power of a browser‑driven AI agent entirely within Google Colab. We will utilize Playwright’s headless Chromium engine, along with the browser_use library’s high-level Agent and BrowserContext abstractions, to programmatically navigate websites, extract data, and automate complex workflows. We will wrap Google’s Gemini model via the langchain_google_genai connector to provide natural‑language reasoning and decision‑making, secured by pydantic’s SecretStr for safe API‑key handling. With getpass managing credentials, asyncio orchestrating non‑blocking execution, and optional .env support via python-dotenv, this setup will give you an end‑to‑end, interactive agent platform without ever leaving your notebook environment.
!apt-get update -qq
!apt-get install -y -qq chromium-browser chromium-chromedriver fonts-liberation
!pip install -qq playwright python-dotenv langchain-google-generative-ai browser-use
!playwright installWe first refresh the system package lists and install headless Chromium, its WebDriver, and the Liberation fonts to enable browser automation. It then installs Playwright along with python-dotenv, the LangChain GoogleGenerativeAI connector, and browser-use, and finally downloads the necessary browser binaries via playwright install.
import os
import asyncio
from getpass import getpass
from pydantic import SecretStr
from langchain_google_genai import ChatGoogleGenerativeAI
from browser_use import Agent, Browser, BrowserContextConfig, BrowserConfig
from browser_use.browser.browser import BrowserContextWe bring in the core Python utilities, os for environment management and asyncio for asynchronous execution, plus getpass and pydantic’s SecretStr for secure API‑key input and storage. It then loads LangChain’s Gemini wrapper (ChatGoogleGenerativeAI) and the browser_use toolkit (Agent, Browser, BrowserContextConfig, BrowserConfig, and BrowserContext) to configure and drive a headless browser agent.
os.environ["ANONYMIZED_TELEMETRY"] = "false"We disable anonymous usage reporting by setting the ANONYMIZED_TELEMETRY environment variable to “false”, ensuring that neither Playwright nor the browser_use library sends any telemetry data back to its maintainers.
async def setup_browser(headless: bool = True):
browser = Browser(config=BrowserConfig(headless=headless))
context = BrowserContext(
browser=browser,
config=BrowserContextConfig(
wait_for_network_idle_page_load_time=5.0,
highlight_elements=True,
save_recording_path="./recordings",
)
)
return browser, contextThis asynchronous helper initializes a headless (or headed) Browser instance and wraps it in a BrowserContext configured to wait for network‑idle page loads, visually highlight elements during interactions, and save a recording of each session under ./recordings. It then returns both the browser and its ready‑to‑use context for your agent’s tasks.
async def agent_loop(llm, browser_context, query, initial_url=None):
initial_actions = [{"open_tab": {"url": initial_url}}] if initial_url else None
agent = Agent(
task=query,
llm=llm,
browser_context=browser_context,
use_vision=True,
generate_gif=False,
initial_actions=initial_actions,
)
result = await agent.run()
return result.final_result() if result else NoneThis async helper encapsulates one “think‐and‐browse” cycle: it spins up an Agent configured with your LLM, the browser context, and optional initial URL tab, leverages vision when available, and disables GIF recording. Once you call agent_loop, it runs the agent through its steps and returns the agent’s final result (or None if nothing is produced).
async def main():
raw_key = getpass("Enter your GEMINI_API_KEY: ")
os.environ["GEMINI_API_KEY"] = raw_key
api_key = SecretStr(raw_key)
model_name = "gemini-2.5-flash-preview-04-17"
llm = ChatGoogleGenerativeAI(model=model_name, api_key=api_key)
browser, context = await setup_browser(headless=True)
try:
while True:
query = input("\nEnter prompt (or leave blank to exit): ").strip()
if not query:
break
url = input("Optional URL to open first (or blank to skip): ").strip() or None
print("\n
Read More
