- Work out from the first page what the application is for.
- Open each entry of the main navigation and check it leads where its name says.
Find the bugs your scripts were never told to look for.
Stepup sends personas through your web application the way your users go through it: with goals, habits and impatience. Each finding comes back with a recording, the steps to reproduce it and the rule it breaks.
For test and staging environments. Runs use your own model key, with a cost cap you approve first.
- Step 12Presses Enter: the list gains “Confirm the caterer”
- Step 13 · findingThe counter reads “2 items left” with 3 active items
- Step 14 · nextAbout to click “Clear completed”, to see what it removes
From an address to your first findings
No recorder, no scripts to write first. You say where the application is and who uses it.
-
1
Give it an address
A test or staging address is enough to start. If your application signs in, add a test account that holds test data only.
-
2
Pick who explores
Start with three personas that suit almost any application, or describe your own users: their goals, their habits and what they must never do.
-
3
Review what they found
Each finding has its basis, a recording and how far it is confirmed. Confirm it, dismiss it or ask for a second look.
Explored the way people use it
A persona has a context, a working style and goals of its own. One reads before clicking and follows the main path. Another never touches the mouse. A third is in a hurry and changes their mind.
- Goals, not scripts. Nothing to record, and nothing to repair when a button moves.
- Your rules, when you have them. Say what the application should do and findings are judged against it.
- Limits a persona keeps. Tell it what it must never do, and it does not.
- Reach every control with Tab alone, and always see where the focus is.
- Open a dialog with the keyboard, close it with Escape and check where the focus goes.
- Start something, leave it half-way, come back and check what was kept.
- Reload in the middle of a step and check where you are afterwards.
The counter shows the number of active items.
“2 items left” with three rows unchecked.
- ✓ObservedRecorded by the persona
- ✓EvidencedA rule, with the recording
- ✓ReproducedSeen again in a fresh session
Every finding arrives with its proof
A finding is a claim, so it comes with what you need to check it in a minute rather than an afternoon.
- Its basis. The rule it breaks, or marked as a question when there is none.
- Its confidence. Observed, evidenced, or reproduced in a fresh session.
- Its recording. The steps around it and the calls the page made.
People decide. Every finding waits for a reviewer before it counts as a defect.
You hold the brake
A persona clicks, types and submits like a person. So every run has limits, and someone who approved it.
A cost cap you approve
No paid run starts without an OK. It stops at the cap and reports what it spent.
Limits on every run
Minutes, actions and cost. The run ends and writes its report when any one is reached.
Pause, step, stop
Hold the next action, see what it is about to do and why, let one through, or stop the run.
Read-only to begin with
Starter missions change nothing: what a persona tries to send is stopped before your server.
Keep what matters as tests
A confirmed finding becomes a Playwright test you can run on every build. Flows a persona walked through can be kept as tests too.
- Plain Playwright. Readable, in your repository, with no runtime of ours.
- Exploration feeds regression. What a persona found once is checked from then on.
test('the counter shows the active items after adding one', async ({ page, fixture }) => {
await fixture.reset();
await page.getByPlaceholder('What needs to be done?').fill('Confirm the caterer');
await page.getByPlaceholder('What needs to be done?').press('Enter');
await expect(page.getByTestId('todo-count')).toHaveText('3 items left');
});
Made for more than one person
An organisation holds your people, your applications and their runs. Invite colleagues by email and give each a role.
- Viewer
- Sees runs, findings and tests.
- Member
- Also reviews findings, and runs or stops generated tests.
- Admin
- Also starts and controls runs, edits settings and manages members.
- Owner
- Also grants and removes owners.
What people ask first
Does it run against production?
No. Point it at a test or staging environment. Personas click, type and submit like real people, so they belong where test data lives.
What do you need from us to start?
An address the platform can open. If your application signs in, a test account that holds test data only. Rules and documents can be added later, and make findings sharper.
Which model does it use, and who pays for it?
Runs use Claude models through OpenRouter, on your own key. Every run has a cost cap that someone approves before it starts, and it reports what it spent.
How is this different from scripted end-to-end tests?
Scripts check what you thought of. Personas explore with goals and report what does not behave as the application's own rules and pages say it should. What you confirm becomes a script.
Does it replace our testers?
No. It finds and documents; people decide. Every finding waits for a reviewer, with the evidence to make that quick.
Can I use it today?
Stepup is in early access. Creating an account and an organisation works today; running against your own application is switched on for each organisation in turn.
See what your users would have found.
Create an account, name your organisation and invite the people who review with you.