LaVague — Natural Language Web Automation

What it is

LaVague is an open-source framework for building AI web agents that interact with websites through natural language instructions. You give it a text objective like 'Find the cheapest flight from NYC to London on this travel site,' and LaVague drives a real browser (via Selenium) to accomplish the task. It uses a World Model to understand the page and an Action Engine to generate and execute browser actions.

It targets developers building browser automation, web testing, or data collection workflows who want to replace brittle CSS selectors and XPath with natural language instructions.

How it saves time or tokens

LaVague replaces traditional browser automation scripts that break when a website changes its layout. Instead of maintaining fragile selectors, you describe what you want in plain English. The Action Engine adapts to page changes automatically. For repetitive web tasks -- filling forms, extracting data, navigating multi-step flows -- LaVague reduces the code you write from dozens of lines to a single objective statement.

How to use

1. Install LaVague:

pip install lavague

1. Write a web agent:

from lavague.core import WorldModel, ActionEngine
from lavague.core.agents import WebAgent
from lavague.drivers.selenium import SeleniumDriver

driver = SeleniumDriver(headless=False)
agent = WebAgent(
    WorldModel(),
    ActionEngine(driver),
    driver
)

1. Run the agent with an objective:

agent.get('https://news.ycombinator.com')
agent.run('Click on the top story and summarize the comments')

Example

from lavague.core import WorldModel, ActionEngine
from lavague.core.agents import WebAgent
from lavague.drivers.selenium import SeleniumDriver

driver = SeleniumDriver(headless=False)
agent = WebAgent(WorldModel(), ActionEngine(driver), driver)

# Navigate to a site and perform a multi-step task
agent.get('https://github.com/trending')
agent.run('Find the top trending Python repository and open its README')

# The agent will:
# 1. Analyze the trending page layout
# 2. Identify the top Python repository
# 3. Click through to the repository page
# 4. Scroll to or click on the README section

Related on TokRepo

• AI tools for browser automation -- Compare browser automation frameworks
• AI tools for web scraping -- Data extraction and web scraping tools

Common pitfalls

• LaVague runs a real browser, which requires Selenium and a Chrome/Chromium installation. Headless mode works for servers, but some sites detect and block headless browsers.
• Complex multi-page flows may require breaking the objective into smaller steps. A single vague objective can lead to incorrect actions on complex websites.
• LLM API costs apply for each page interaction. Pages with many elements generate longer prompts. Set token budgets for cost control in production automation.