Skip to content

Code moves faster than ever.Your QA should too.

Hand the testing to QAgents: it writes and runs your tests in a real browser and files every bug with the recording, while you decide what gets checked.

A QAgents Run: the Test Case opens the shop, searches for mugs and adds Trail Mug to the cart. The cart badge shows 2 where the Test Case expects 1, on all 3 Attempts, so QAgents files a Bug with the recording attached.
Run #128
How it works

From your app’s URL to your first bug

  1. 1

    Connect your app

    Its URL and a test login. Have specs or a repo? Add them and QAgents uses them too.

    Environment
    acme-shop.test
    Test login
    ada@acme.test · ••••••••
  2. 2

    Approve the test plan

    QAgents explores your app and proposes tests in plain English. You approve them; it writes the code.

    • Search finds a product by name
    • Adding a product updates the cart
    • Checkout rejects an expired card
    Approve 3 tests
  3. 3

    Every run, triaged

    Every test runs in a fresh browser and is retried before it counts as failed. Every failure is sorted out before it reaches you.

    19 passed · 3 triaged

    • BugFiled, with the recording
    • Stale StepRepair waiting for you
    • FlakeRecorded, not filed
Evidence

Every failure comes with proof

A screenshot of every step, the full recording, the network calls and exactly what was clicked. Open the bug and watch it happen.

Verdicts

A failure is not a Bug until it’s judged

Every failure gets up to 3 attempts. Then it gets one of four verdicts, and each one does something different.

Bug

Your app is wrong.

Filed with the steps to reproduce, expected vs actual, and the recording. Seen again, it's added to the same bug. Fixed, it closes itself.

Stale Step

Your app changed on purpose.

QAgents updates the test, proves the fix with a re-run, and waits for your OK.

Flake

It failed, then passed on a retry.

Recorded, not filed. A test that flakes in 2 of its last 5 runs is marked flaky, so you can deal with it.

Environment

Your app was down.

Not blamed on your app. If it goes down mid-run, QAgents stops instead of piling up false failures.

RepairCheckout button was renamed
−await act("click Checkout")
+await act("click Continue to payment")
await expect(page).toHaveURL(/pay/)

Verified by a passing re-run

Doesn’t change what the test checks

Accept RepairReject

You decide what “correct” means. QAgents can fix a test’s steps on its own, but a fix that changes what the test checks always waits for your OK.

Tests

Code asserts.
AI acts.

Every test is a short TypeScript file you can read, edit and review. The model only chooses what to click; whether your app is right is plain Playwright.

act

The AI picks the element

A renamed button doesn’t break the test. It reports its choice and a confidence, and tells you when the page changed since last time.

expect

Assertions stay deterministic

Playwright’s own expect. No model decides whether your cart shows 1.

%pw%

Your passwords never reach the AI

Logins are filled in at the last moment; the AI picks the field, never sees the value.

cart/add-to-cart.tsTypeScript
1import { expect } from "@qagents/test"
2
3export default async ({ page, act, target, testAccount }) => {
4 await page.goto("/login")
5 const email = testAccount("ada", "username")
6 await act("type %email% into Email", { variables: { email } })
7 const pw = testAccount("ada", "password")
8 await act("type %pw% into Password", { variables: { pw } })
9 await act("click Sign in")
10 await expect(target("the account menu")).toBeVisible()
11
12 await page.goto("/products")
13 await act("type mug into Search products")
14 await act("click Add to cart on Trail Mug")
15 await expect(target("the cart count"))
16 .toHaveText("1")
17}

Ship it. We already checked.

Add a test login, approve the plan, and get your first results. No test code to write.

Start testing your app
Run #122 tests
  • BugCart badge shows 2, expected 1Filed with the recording
  • Stale StepCheckout button was renamedRepair verified, waiting for you
  • FlakeSearch results loaded latePassed on attempt 2 · not filed

19 passed · 3 triaged