Drop-in UI testing for any GitHub repository.

Mention @testdriverai anywhere in your repository. It writes UI tests and catches regressions before they merge.

Get Started Free · No CC Required

· Platform: Desktop· OS: Windows Analyzing UI…

it("should trigger the upload modal", async (context) => {

  const testdriver = TestDriver(context);

  await testdriver.provision.chrome({
    url: 'https://example.com',
  });

  let button = await testdriver.find("upload button")
  await button.click()

  await testdriver.assert(
    "upload dialog appears"
  );

});

Elements Tested with AI: 0M+
Hours Automated: 0K+
Test Cases Created: 0K+
Test Runs Executed: 0K+

See it in action

Watch how TestDriver automates complex user flows across different platforms with AI-powered precision.

Trusted by leading engineering teams

The next generation of pixel-based testing

How it compares

TestDriver combines the coverage of manual testing with the speed and reliability of automation.

	TestDriver	Playwright	Screenshot Automation	Manual Testing
Blackbox Testing Tests the app without access to the source code, DOM, or internal APIs.Tests the app without access to the source code, DOM, or internal APIs.	Yes	No	Yes	Yes
Runs in CI/CD Tests run automatically in your pipeline on every commit or pull request.Tests run automatically in your pipeline on every commit or pull request.	Yes	Yes	Yes	No
Self-Healing Tests keep passing when colors, layout, or styling change, instead of breaking on every pixel difference.Tests keep passing when colors, layout, or styling change, instead of breaking on every pixel difference.	Yes	Yes	No	Yes

Product Tour

How it works

Explore

Generates tests with AI Vision, then learns your app

Describe a flow in plain English and the vision agent clicks, types, and reads the screen to figure it out, then writes the test for you — no selectors, no DOM, no setup. Every element it finds is cached with a vision fingerprint, so the next run matches the cache instead of calling the AI. Passing tests replay instantly and deterministically, with little to no token cost.

· Platform: Desktop· OS: Windows Analyzing UI…

it("should trigger the upload modal", async (context) => {

  const testdriver = TestDriver(context);

  await testdriver.provision.chrome({
    url: 'https://example.com',
  });

  let button = await testdriver.find("upload button")
  await button.click()

  await testdriver.assert(
    "upload dialog appears"
  );

});

Third-party web apps.: Test apps you don't own or control without access to source code.
Chrome extensions.: Browser extensions with popups, content scripts, and background pages.
Desktop apps.: Native Windows and macOS applications with full interaction support.
VS Code extensions.: Test your editor extensions in real development environments.
Rich media & embedded content.: Canvas, video players, images, iFrames, and OAuth flows that break selector-based tools.
Documents & AI content.: File uploads, PDFs, LLM chatbots, and spelling & grammar validation.

Run

Runs tests within CI

Tests are plain Vitest, so they run the same on your machine and in CI, across web, desktop, and extensions on real VMs. Wire them into the pipeline you already have.

Validate

Validates outcomes with assertions

Run a visual assertions on the machine state. TestDriver uses AI as a "judge" and returns a Boolean response and reasoning. Did we mention assertions are cached as well?

Validate assertion Preparing assertion…

Heal

Keeps tests passing when elements move or change.

When the UI shifts and a cached element no longer matches, TestDriver re-invokes the AI to find it, updates the cache, and keeps going. Tests survive redesigns instead of breaking on every pixel, with no rewrites.

Adapt to UI changes Reading cache…

Debug

Makes it easy to debug failing tests.

Jump from a red build straight to a video replay of the run that broke it, with network logs and every click in order. See what happened instead of guessing.

Prevent

Catches regressions and monitors software quality

Track pass rates, flaky tests, and failure trends across every run, so a regression shows up the moment it lands. Export to JUnit XML to feed any reporter you already use.