No more writing brittle XPath. No more guessing CSS selectors.
With talk2dom, you describe what you want to find — it returns the right locator.

🧠 What is talk2dom?

talk2dom is a Python tool that lets you locate any DOM element using natural language + an LLM.

from talk2dom import get_locator
by, value = get_locator(driver, "Find the login button")

Internally it sends the page HTML and your instruction to an LLM (e.g., GPT-4o or LLaMA-3), and returns:

("xpath", "//div[@class='header']//button[1]")

You then use it to perform your next action.

💡 Why is this useful?

Because in test automation and AI agent flows, the hard part is not clicking — it’s locating the right thing to click.

  • Pages are complex
  • IDs and classes often change
  • Writing XPath is error-prone

talk2dom solves that by letting you focus on intent, not HTML structure.

✨ New: Automatic Element Highlighting

We recently added built-in highlighting to make debugging and demos easier.

from talk2dom import get_element

elem = get_element(driver, "Click the confirm button")
elem.click()

What it does:

  • Adds a red solid outline and light background to the element
  • Keeps it visible for duration seconds
  • Resets the style after — no manual work needed

This helps you:

  • Visually confirm what the LLM selected
  • Record videos with agent feedback
  • Build no-code workflows with visual trace

📦 Install 

pip install talk2dom

Works with OpenAI, Groq, or any LangChain-compatible model.

🧪 Sample Script 

from selenium import webdriver
from talk2dom import get_element

driver = webdriver.Chrome()
driver.get("http://example.com")

el = get_element(driver, "Click the search icon")
el.click()

You can also pass a WebElement to scope the HTML and save tokens:

sidebar = driver.find_element(By.ID, "sidebar")
el = get_element(driver, "Find the settings button", element=sidebar)

🧠 Philosophy

We don’t control your browser.

We help you find what to control.

GitHub → https://talk2dom.itbanque.com/
PyPI → https://pypi.org/project/talk2dom/

Feedback, stars, contributions — all welcome ✨