Describe the test.
A real browser runs it.

A model clicks through your site like a user, looks at every screen and hands back a verdict with the evidence.

3 runs free · no card · laptop, tablet and phone

A real run · llmgateway.io · gpt-5.6-luna

01Inspect the landing page before testing sign-up115 ms
02Identify the primary sign-up call to action156 ms
03Open the primary sign-up page712 ms
04Inspect sign-up fields and constraints120 ms
05Check validation when submitting the empty form636 ms
06Check invalid email and weak password validation666 ms
07Submit the requested throwaway account1,601 ms
08Verify post-sign-up onboarding and collect diagnostics1,289 ms

Pass8 of 8 checks · 1m 13s

Any vision modelGPT-6 AstraGPT-5.6 SolGPT-5.6 LunaClaude Sonnet 4.6Claude Haiku 4.5Gemini 3 ProGemini 3.5 FlashDeepSeek V4

How it works

Start it anywhere. Get everything back.

The appPaste a URL, pick a device
Pull requestsEvery commit, on its preview
SchedulesHourly to weekly
Your coding agentOver MCP

A ZeusQA run

A real browser, in the cloud or your Chrome
Any vision model on LLM Gateway
Jev judges every step

Verdict and checksPass, fail or blocked
FindingsEach with its evidence
Playwright testReplay it, keep it
Comment and emailWhere your team looks

Playwright export

Every run writes its own test.

1Explore
2Write
3Replay

The page the run tested: example.com — https://example.com/

1Capture initial state of example.com
2Inspect structure, links, images and errors
3Verify the “Learn more” link
4Verify CSS, SVG and final diagnostics
5Verify HTML structure and security

zeusqa-run_mvwspsyn_ec5ccaf6.spec.tstrimmed

import { test } from "@playwright/test";

test("Run a smoke test of the site", async ({ page }) => {
  await page.goto("https://example.com/");  await test.step("1. Capture initial state of example.com", async () => {
    const screenshot = await computer.screenshot();
  });  await test.step("2. Inspect page structure, links, images, and check for errors", async () => {
    const links = await page.locator('a[href]').evaluateAll((els) => els.map((el) => el.href));
    const diagnostics = computer.diagnostics();
  });  await test.step("3. Verify the \"Learn more\" link is valid and accessible", async () => {
    const learnMoreLink = page.locator('a:has-text("Learn more")');
    const response = await page.request.head('https://iana.org/help/example-domains');
  });  await test.step("4. Verify CSS loading, SVG rendering, and final diagnostics", async () => {
    const svgElements = await page.locator('svg').count();
    const finalDiagnostics = computer.diagnostics();
  });  await test.step("5. Verify HTML structure and security", async () => {
    const htmlLang = await page.locator('html').getAttribute('lang');
    const hasViewport = await page.locator('meta[name="viewport"]').count();
  });});

Replay

Playwright exportRun run_mvwspsyn_ec5ccaf6 on zeusqa.ai, 31 Oct 2026. Code trimmed to a few real lines per step.

What a script walks past

The flow passed. The run failed.

Searching MDN for “flexbox” worked end to end. The console did not: 22 errors on the result page. A recorded script clicks straight past that. ZeusQA reports it.

Run run_mu4c9jds_c39674a4, 16 Sep 2026. Checks and finding quoted from the report.

Faildeveloper.mozilla.org5/6 checks · 35s

Smoke test: take a screenshot, open the site search, search for 'flexbox', verify results appear and open the first result. Then check computer.diagnostics() and submit the report.

Initial homepage screenshot captured
Site search opened
Search for flexbox
Results appear
First result opened
Diagnostics checked

mediumInvalid sandbox attribute console errors on Flexbox page

After opening the first search result, computer.diagnostics() reported 22 console errors: “Error while parsing the 'sandbox' attribute: 'allow-fullscreen' is an invalid sandbox flag.” The page otherwise loaded successfully and there were no uncaught page errors or failed network requests.

Evidence · Steps 5 and 6

On every pull request

Test it. Fix it. Then merge.

ZeusQA tests what the pull request says it changed, marks each acceptance criterion and reads the traffic for security problems. A failing run fails the ZeusQA check unless all it found is below the severity the repository blocks at, High by default, which reports as neutral. With the check required, the merge waits for the fix.

Recorded in the demo workspacePull request claims: how it works

The title and description become checks, run in a real browser on the preview.
Each item under Acceptance criteria or Test plan, marked met, not met or not checked.
Headers, cookies, mixed content, redirects, source maps and leaked secrets, read from traffic the test already caused.
Copy a prompt, or open it in Cursor or Claude Code, from any finding.
Runs per week, pass rate and findings by severity, with a CSV of each table.

Connect GitHub

Pricing

Start free. Pay when you ship more.

Every plan runs the same real browser and the same models. Plans differ in runs and people.

Free

Try it on one site.

$0/mo

3 test runs (total)
1 project
Jev in every run
Live browser view
Findings in your editor (MCP)
Community support

Start free

Pro

Team

For a team sharing one QA budget.

$149/mo

1,500 test runs / month
Organizations with members
Roles & shared projects
50 schedules
Security review of pull requests (500 commits / month, 20 / day)
Vulnerability watch on 50 repositories
Everything in Pro, Jev included
Priority support

Upgrade to Team

Prices in USD, billed monthly through Stripe. Compare plans

Questions

Your first run takes about a minute.

Start free I have an account

What is ZeusQA?

You give ZeusQA a URL and describe the test in plain English. It opens a real browser, works through the flow like a user, and returns a verdict: what it checked, what broke, and a screenshot of every step. On Pro and Team it can also record a video of the session.

How is it different from Playwright or Cypress?

Those run the script you wrote, and break when the page changes. ZeusQA works out the steps from your description and adapts as it goes, so it catches console errors, failed requests and layout problems a fixed script walks past. And you keep the repeatability: every run is also a plain Playwright test you can replay from the run page with no model, or run in your own CI. Full comparison at zeusqa.ai/vs/playwright.

Which AI models can I use?

Any vision model on LLM Gateway that can drive a browser — GPT-6 Astra, Claude, Gemini and others. Pick speed (Luna, Haiku) or depth (Astra) per run. Image generators and embedding models are filtered out: a QA run cannot use them.

How much does it cost?

Free includes 3 test runs. Pro is $39/month for 300 runs, unlimited projects, 10 schedules, live view and optional video. Team is $149/month for 1,500 runs, 50 schedules and organizations with members.

Do I need to verify my email?

You can sign up and run free tests immediately. Email verification is required before subscribing to a paid plan.

Is it safe to run on my site?

It never follows instructions it finds on the page, and it avoids destructive actions unless you allow them in the prompt. It is a real browser clicking real buttons, so point it at staging and use throwaway accounts for anything that writes data.

Describe the test.A real browser runs it.