AI QA Agent for GitHub

👋 Howdy! I'm the TestDriver agent.

I write and debug computer-use tests: I drive a real browser (or desktop app) in a cloud sandbox using AI vision by clicking, typing, scrolling, and asserting on what's actually on screen, just like a human would. For example, you can ask me to:

• @testdriverai Test that the homepage loads and the nav links work
• @testdriverai Sign in with valid credentials and confirm the dashboard appears
• @testdriverai Add an item to the cart and verify the cart count updates

You describe a flow in plain English, and I turn it into a reliable, repeatable vitest test, run it, and open a PR with the recording attached.

Quickstart

1. Install this GitHub App
2. In a GitHub issue or PR, ask @testdriverai to build or maintain a test
3. Review the created tests, merge, and repeat!

FAQ

How does it work?

TestDriver uses AI vision instead of selectors, we use computer-use agents and a custom caching system.

TestDriver generates computer-use tests with the TestDriver JS SDK. (ex await testdriver.find('the sign in button').click())

How is this different than Playwright?

Playwright works best for single browser tabs. TestDriver drives a high powered virtual machine, which means it can handle multiple tabs, videos, animations, desktop apps, and interactions between apps. The kinds of things real users do.

Additionally, TestDriver is a blackbox system, it uses AI vision and pixel matching instead of selectors, so it works without any knowledge of the underlying code.

What does test code look like?

Here is an example test that opens chrome, and logs in.

await testdriver.provision.chrome({ url: 'http://testdriver-sandbox.vercel.app/login' });

const signInButton = await testdriver.find("Sign In, black button below the password field");
await signInButton.click();

const result = await testdriver.assert("an error shows that fields are required");

expect(result).toBeTruthy();

You can find more examples and the full SDK reference in the docs.

How much does this cost?

Your first 4 hours of testing are free, then our plans start at $20 a month.

Helpful Links

🚀 Quick Start • 📖 Documentation • 💻 Examples • 📖 Pricing • 💬 Discord • 🌐 Website