Usage

Alumnium wraps your existing browser driver with three natural-language methods: do(), check(), and get(). You write what the browser should do; Alumnium calls the LLM to figure out the exact clicks, types, and reads.

Python (pytest + Selenium)

from alumnium import Alumni
from selenium.webdriver import Chrome
from pytest import fixture

@fixture
def driver():
    driver = Chrome()
    yield driver
    driver.quit()

@fixture
def al(driver: Chrome):
    al = Alumni(driver)
    yield al
    al.quit()

def test_search(al: Alumni, driver: Chrome):
    driver.get("https://duckduckgo.com")
    al.do("search for 'Mercury element' and press Enter")
    al.check("page title contains Mercury word")
    al.check("search results contain Wikipedia articles")
    symbol = al.get("chemical symbol")
    assert symbol == "Hg"

TypeScript (Mocha + Selenium)

import { strict as assert } from "assert";
import { Alumni } from "alumnium";
import { Builder, Browser, type WebDriver } from "selenium-webdriver";

describe("TestSearch", () => {
  let driver: WebDriver;
  let al: Alumni;

  before(async () => {
    driver = await new Builder().forBrowser(Browser.CHROME).build();
    al = new Alumni(driver);
  });

  after(async () => {
    await driver.quit();
    await al.quit();
  });

  it("should search", async () => {
    await driver.get("https://duckduckgo.com");
    await al.do("search for 'Mercury element' and press Enter");
    await al.check("page title contains Mercury word");
    await al.check("search results contain Wikipedia articles");
    const symbol = await al.get("chemical symbol");
    assert.equal(symbol, "Hg");
  });
});

Java (JUnit 5 + Selenium)

import ai.alumnium.Alumni;
import org.junit.jupiter.api.*;
import org.openqa.selenium.chrome.ChromeDriver;

class SearchTest {
  private Alumni al;
  private ChromeDriver driver;

  @BeforeEach
  void setUp() {
    driver = new ChromeDriver();
    al = new Alumni(driver);
  }

  @AfterEach
  void tearDown() {
    al.quit();
  }

  @Test
  void shouldSearch() {
    driver.get("https://duckduckgo.com");
    al.act("search for 'Mercury element' and press Enter");
    al.check("page title contains Mercury word");
    al.check("search results contain Wikipedia articles");
    String symbol = (String) al.get("chemical symbol");
    Assertions.assertEquals("Hg", symbol);
  }
}

Core method reference

Method	Description
al.do("instruction")	Perform an action (click, type, navigate, etc.) described in plain English
al.check("assertion")	Assert something about the page state; raises on failure
al.get("question")	Return a value extracted from the page (string, number, etc.)
al.quit()	End the Alumnium session and release LLM resources

Minimal inline example (Python, no test framework)

from alumnium import Alumni
from selenium.webdriver import Chrome

driver = Chrome()
driver.get("https://search.brave.com")
al = Alumni(driver)

al.do("type 'selenium' into the search field, then press Enter")
al.check("page title contains selenium")

driver.quit()
al.quit()

Notes

• Each do(), check(), and get() call makes one or more LLM API requests. OpenAI costs apply per call.
• Tests run slower than pure Selenium because of the LLM round-trip (expect 5-30 seconds per step).
• ChromeDriver must match your installed Chrome version. Use webdriver-manager or the browser's bundled driver.