Stop writing XPath. I built a tool to locate web elements using natural language.

If you've ever written a Selenium script, you know the pain:

driver.find_element(By.XPATH, "//div[@class='form']/button[2]")

It’s fragile. It’s unreadable. And when the DOM changes — it breaks.

That’s why I built talk2dom, a Python tool that lets you locate DOM elements using natural language + an LLM.

🧠 The Problem

The hard part of web automation isn’t clicking — it’s locating.

• Buttons with no ID


• Forms wrapped in modals


• Dynamic pages with dozens of similar <div>s

And writing XPath by hand? That’s not automation — that’s archaeology.

🛠️ The Tool

talk2dom does one thing really well:

🗣️ You describe the element.


🤖 It gives you a selector.

Example:

from talk2dom import get_locator



by, val = get_locator(driver, "Find the login button")

driver.find_element(by, val).click()

Internally, it sends the raw HTML + your instruction to an LLM (OpenAI, Groq or other providers), then returns:

("xpath", "//button[@type='submit']")

You still control the browser — it just tells you where to look.

🔓 Free model support (Groq)

You can use open models like LLaMA-3 via Groq:

export GROQ_API_KEY="your_key"

by, val = get_locator(driver, "Find the search field", model="llama-3.3-70b-versatile", model_provider="groq")

It’s fast — and free.

📦 Install

pip install talk2dom

🤖 Built for LLM-native workflows

• Structured function call output (Pydantic)


• Works with both driver or scoped WebElement
  


• No hallucinations — just selectors

📌 TL;DR

If you:

• Write automation scripts


• Build LLM-based web agents


• Hate writing XPath

Try talk2dom.


You might never inspect the DOM manually again.

Feedback and PRs welcome 🙌