What's new

Tests can now sign in to your app. Open an environment in Apps, choose Configure auth profile, save a username and password or custom login steps, and check them with Save & test login. On New test, pick the environment in the Sign-in picker: planning reads the signed-in page and the run signs in once. The password is encrypted and never appears in screenshots, logs or proof pages.

Plans now use credits: Builder $29/mo (300 credits), Team $99/mo (1,000 credits), Scale $299/mo (3,000 credits). 1 credit is 1 passing run on one browser, up to a minute. Only passing runs use credits. A run that catches a bug in your app, or fails because of us or your environment, is free. There is no Free plan; try it first with the free, no-signup check at /check.

The welltested CLI is on npm. Run npx welltested setup and it adds the Well Tested MCP server and a "verify before done" skill to Claude Code, Cursor, Codex, Windsurf, Copilot or Kiro. Your agent reruns your kept tests before it tells you it's done, and failures come back with the reason and a fix prompt. welltested tunnel lets a cloud run reach your localhost.

Keep a test after any run and it joins a suite; Home shows each suite's latest result. Suites rerun on every GitHub push, on every deploy through a hook URL, or on a schedule (daily on Builder, hourly on Team and Scale). You get an email when a suite goes red or recovers, and PR comments on Team and Scale.

Failed runs now keep the browser console and network requests for each step, show what changed since the last green run, point at the likely source file when your app ships source maps, and offer a fix prompt to paste into your coding agent.

When a button is renamed or moves, the run repairs that step from a fresh look at the page and keeps going. Repaired steps are labelled "Passed after repair" and you can keep the repair with one click.

If a step still can't find its target, for example on a canvas or an icon-only button, a vision model finds it in the screenshot. Those steps are labelled "Clicked by vision".

Point Well Tested at a chat widget or AI endpoint on an origin you verify. Twelve scenarios check that it does its job, stays on task, resists prompt injection, doesn't leak data, keeps a safe tone and gets known facts right. Each scenario runs only after you approve the suite.

welltested run --local reruns your kept tests in a browser on your computer, so localhost works without a tunnel. Evidence is masked and uploaded, and the proof page says it ran on your machine.

Endpoints your app calls during browser runs are listed under "Seen during browser runs". Turn one into an API test with one click; it opens for review and never runs on its own.

Download any kept test as a standalone Playwright file that runs without Well Tested.

Failed runs now get an assessment from two models. Jev reads the step trace and Clef reviews the screenshots against your goal, then the run page says whether it looks like a real bug in your app or a test or environment problem. It never changes your pass/fail result, receipt or credits.

Describe the journey out loud, up to 60 seconds in English or Turkish. The transcript becomes an editable goal and nothing runs until you approve it.

Jev now tracks TypeSafe's latest release on OpenRouter, so decision quality improves without waiting on us.

Chromium, Firefox and WebKit scans all passed on the current release. Firefox, WebKit and mobile-web emulation are included on Team and Scale.

Standalone page scans run on their own isolated worker with their own key. If that key is missing, scans stop with a clear error instead of borrowing another worker's credentials.

Browser workers run as a non-root user with outbound network controls, so a test can only reach the site you asked it to test.

Values you type into a test are stored encrypted and masked in screenshots, page snapshots and step logs. Videos that would show them are withheld.

Every failed step now says what went wrong in plain language, in the run view and the replay.

A run that checked nothing is marked inconclusive, never passed, and every view shows the same verdict.

Ask for something like "check every tab in the nav" and the plan is split into parts of up to 12 steps that you review and continue one at a time.

Sign-in uses an emailed link, and you can sign out every other session from Settings.