Cyclops Request access
Visual end-to-end testing · Early access

Always keeping an eye on your app.

Connect a repo. Cyclops explores your web app the way a user would while it records, then suggests the end-to-end tests that matter, each with a replay of what it saw. Approve them, and it checks your app on every deploy, with video, screenshots and a trace for every result.

Request early access How it works
  • A fresh sandbox for every job
  • Secrets never reach the AI
  • Alerts in Slack and PagerDuty
acme/storefront · explore Example
Replay localhost:3000/checkout
Sign up Search Cart Checkout
00:0003:12

Cyclops suggests · 4 tests

  1. Guest checkout completes with a test cardcritical
  2. Sign up and land on the dashboardcritical
  3. Search filters products by categoryhigh
  4. Cart keeps its items after a reloadmedium
Waiting for your review

The problem

E2E tests are the ones everyone wants and nobody writes.

They take days to write, break on every redesign, and fail for reasons nobody can see. So most teams ship with a handful of smoke tests and hope. Cyclops writes the suite from your running app, and shows you exactly what each test does before it runs.

How it works

From a repo to a suite that runs on every deploy.

  1. 1

    Connect a repo

    Install the GitHub App and pick a repo. Cyclops works out how to install, build and boot your app, and asks only for the keys it can't fill in itself.

  2. 2

    Watch it explore

    An agent signs in and uses your app in a sandbox while it records. Watch it live, or scrub through the replay afterwards.

  3. 3

    Approve the tests

    Each suggestion comes with its steps, its code and the clip where it was found. Approve, edit or reject them, or turn on auto-approve.

  4. 4

    Run on every deploy

    On a schedule or on each deploy, with video, trace and screenshots for every result. Failures reach you in Slack or PagerDuty.

Built to be trusted with your app

An agent in your app, on a short leash.

A fresh sandbox per job

Every job boots on its own throwaway VM with no cloud identity and no route into private networks. It is deleted when the job ends.

The AI sees names, not secrets

Every prompt passes through a redactor, and the model refers to secrets by name only. Signing in is plain code, never an AI action.

Nothing runs unapproved

You approve each test, or opt in to auto-approve. Production environments only run tests a person has marked safe for production.

Real Playwright specs

The model describes steps. A deterministic compiler turns them into Playwright specs, using locators it saw while exploring.

A replay for every result

Video with a step timeline, screenshots and the full Playwright trace. The sign-in step is never recorded.

Flaky tests get quarantined

A test that only passes on retry is marked flaky. Repeat offenders are quarantined, so they stop paging you while someone fixes them.

  • GitHub
  • Playwright
  • Slack
  • PagerDuty

Every test comes with the replay that found it.

Early access

Bring Cyclops to your team.

We're onboarding teams in small batches. Tell us about your app and how you test it today, and we'll be in touch.

What you'll need

  • A GitHub repo with a web app in it
  • A test account that signs in with a password
  • Test-mode keys for any third party your app can't boot without