What is TestDriver?

TestDriver tests apps by seeing the screen and controlling a real machine, without selectors.

TestDriver is an end-to-end testing platform that tests your app the way a person uses it.

• It looks at the screen.
• It finds elements from plain-language descriptions.
• It controls a real machine with the mouse and keyboard.

It doesn’t rely on CSS or XPath selectors. So it can test flows that selector-based tools can’t reach, such as desktop apps, browser extensions, and content that changes every time it renders.

Where selector-based tests fall short

Tools such as Playwright and Cypress drive a single web page through its DOM. They locate elements with selectors.

That works until a test needs something the DOM can’t describe reliably:

Situation	Why selectors break down	Examples
A fast-changing UI	Each redesign changes the markup and breaks CSS and XPath selectors.	Startups, teams shipping AI-generated code
Dynamic content	Generated content has no stable structure to select.	AI chat responses, PDFs, images, video
Software you don’t own	Third-party markup often lacks stable IDs or accessibility attributes.	Other websites, embedded widgets, third-party apps
More than one application	Web-only tools can’t leave the browser tab.	Desktop apps, browser extensions, IDEs, installers
Visual state	Whether something looks right isn’t in the DOM.	Charts, layout, spelling, rendered media

How TestDriver works

A TestDriver test is a Vitest test. It runs against a cloud sandbox, a fresh Linux machine, or Windows on Enterprise plans.

Each step describes what a user would see or do:

1. Provision launches the app under test in the sandbox, such as Chrome at a URL, a Chrome extension, VS Code, an Electron app, or a downloaded installer.
2. Find takes a description like 'Sign in button' and uses AI vision to locate the matching element on screen.
3. Act clicks, types, scrolls, and presses keys on the real machine. The app receives the same input a person would produce.
4. Assert checks a plain-language statement about the screen, such as 'the cart shows one item'.

This test installs the Loom extension from the Chrome Web Store. It checks that Loom appears in Chrome’s extensions menu, a flow that a DOM-only tool can’t drive:

tests/loom-extension.test.mjs

import { describe, expect, it } from 'vitest';
import { TestDriver } from 'testdriverai/vitest/hooks';

describe('Loom extension', () => {
  it('appears in the Chrome extensions menu', async (context) => {
    const testdriver = TestDriver(context);

    // Launch Chrome with Loom installed by its Chrome Web Store ID
    await testdriver.provision.chromeExtension({
      extensionId: 'liecbddmkiiihnedobmlmillhodjkdmb',
    });

    const extensionsButton = await testdriver.find('puzzle-piece extensions icon in the Chrome toolbar');
    await extensionsButton.click();

    const loomExtension = await testdriver.find('Loom in the extensions dropdown');
    expect(loomExtension.found()).toBe(true);
  });
});

To run this yourself, follow the Quickstart.

What’s included

• The SDK, the testdriverai npm package. Use it in Vitest tests to provision apps, find elements, act, and assert.
• Cloud sandboxes that run your tests on fresh Linux machines, or Windows on Enterprise plans. Enterprise customers can also self-host them.
• Debugging tools: screen recordings of every run, screenshots, and element caching for faster, more stable repeat runs.
• The TestDriver agent, which reviews your pull requests and writes and maintains tests from GitHub.

When to use TestDriver

TestDriver is built for flows that are hard to cover any other way:

• Multi-application workflows
• Desktop software
• Browser extensions
• Third-party UIs
• Visual checks

Each step runs on a real machine and uses AI vision. So a TestDriver test takes longer per step than a test that talks to the DOM directly.

If stable, selector-based tests already cover a flow, keep them. Use TestDriver for the parts they can’t reach.

Related

Quickstart

Install TestDriver and run your first test.

Locate elements

Write descriptions that find the right element reliably.

The TestDriver agent

Let the agent review pull requests and build tests from GitHub.

Hosted sandboxes

Run tests on TestDriver’s cloud infrastructure.