# Create your first regression test with FlowQA

> Register with your site's address, let the AI agent explore it, get a generated or recorded test, read its assertions and run it again: what needs the desktop app and what does not.

This is the first session with FlowQA: an account, a project pointed at your site, a first exploration by the AI agent, one test that is generated or recorded, a look at its assertions, and a second run to see it hold.
It also says plainly which steps need the desktop app and which work in the browser.

## What runs where

| | FlowQA on the web | FlowQA Desktop (Mac, Windows) |
|---|---|---|
| Sign in, open projects, see results | Yes | Yes |
| Run tests on cloud browsers | Yes | Yes |
| Run tests on your own machine, free, including sites behind a VPN or on localhost | No | Yes |
| Record a test by clicking through your site | No | Yes |
| Let the AI agent map the site and generate tests | Yes, on a cloud browser | Yes, on a cloud browser or on your machine |
| Repairs, schedules, CI triggers, GitHub hooks, site notes | Yes | Yes |

The web workbench is enough to generate, run and maintain tests.
Recording by hand needs the desktop app because it drives a real browser on your machine.

## 1. Create your account

Open the FlowQA page on testfinch.com and choose Start free, or go straight to app.testfinch.com/register.
Registration is one screen: your name, your email, and what you want to do first.
Choose "Test a website" (coming from testfinch.com, it is already chosen) and give its address, such as `app.example.com`; https:// is added if you leave it out.
We email you a link and a six-digit code; open the link, or type the code on the "Check your email" screen.
Nothing is created until you do.
If you came from a plan on the pricing page, its 14-day trial of Team starts when you confirm, with no card, and the Overview says so.

A TestFinch account works for both products: if your team already uses Snag, sign in at app.testfinch.com/login with the same email instead.
FlowQA Desktop, from the downloads page, signs in with your email and the six-digit code.

## 2. Land on your project

Confirming creates your workspace and a first project, and gives the project its Tests half at the address you entered: one environment, Default, allowed to visit that host.
Both are named for you, from your website or your email's domain, and you can rename them in settings.
You land on the project's Overview with "Get your first result" at the top and the Tests track first: add your site (already done), let the AI agent explore it, run your first test.
The checklist is for the workspace's admins.
Each step turns done by itself from what happened; "Hide this" puts the card away for you on this project.
If the address did not answer when we checked it, the first step says so and links to Test settings, where you can correct it.
Add staging or preview environments later in the project's Test settings.
If the site sits behind a preview password or a sign-in, record the steps that pass it as an opening once; the AI agent and every test can run it first.

## 3. Get a first test

Two ways.

**Generate it.**
Choose Explore on the checklist, or Explore site in the Tests tab.
The AI agent maps the site on a cloud browser, proposes test journeys with a goal and a success condition each, and waits.
Keep, rename or drop the journeys, then generate.
Each kept journey is explored, recorded as steps, given assertions, replayed twice on fresh sessions, and checked against a deliberately broken page; only a test that passes clean and fails when it should is accepted.
Accepted tests appear in the project.
Page text the AI agent stores (titles, link labels, excerpts) is always well formed: a cut never splits a character, and any half character a page or a client sends is stored as U+FFFD.
If the results still cannot be saved, the exploration ends as failed with "The results could not be saved:" and the reason, keeps the map when it can, and the next exploration starts from that map.

Everything the workbench shows about an exploration is read from TestFinch, so closing the dialog or reloading the page loses nothing.
While one is mapping or generating, the project's Explore button shows where it is (for example "Mapping 12/60", from the pages and budget the worker reports), and the dialog opens on it with its progress, the page it is on, and its spend.
Otherwise the dialog opens on the project's explorations, newest first, one row each with its status, environment, when, spend, and "Map kept" where a stopped one kept its map.
A failed exploration shows why it stopped, word for word, when it failed and what it spent.
When its map was kept, it shows how many pages the map covers and the first of them, and the map can be read in place or downloaded; the journeys it proposed are listed too.
Try again opens the start form filled in with the failed or cancelled exploration's own options: environment, browser, journeys, pages to map, link depth, steps per journey, openings, what the AI agent should know, and spend cap.
When the map was kept and the environment is the same, the form says the new exploration starts from it.
Explore again starts a new exploration and leaves the old one in the list.

**Record it** (desktop only).
Choose Record a test, click through the flow in the browser that opens, and stop.
The steps appear in the editor.
Add an assertion for what the page should show at the end, so the test fails when the flow breaks rather than only when it crashes.

## 4. Read the test

Open the test.
Each step shows what it does and the element it acts on; assertion steps show what they check.

To change a test by hand, choose Add step under the last step, or Add step below in a step's menu.
Pick the kind of step (Go to URL, Click, Type, Choose option, Press key, Check, Wait for, Reload, Back, Forward or Screenshot) and fill in what it needs, such as the element's role and name for a click.
An empty test can also start with an opening.
Move up and Move down in the same menu put a step in order, or drag a step by the handle beside its number.
Steps that belong to an opening move and change with the opening, so they offer none of these.
The site notes, under the project's Test settings, hold what the AI agent learned about each page; correct anything that is wrong and the next run reads it.

## 5. Run it again

Run tests on the checklist opens the Tests tab; choose Run on the test there.
On the web the run happens on a cloud browser; on the desktop you can pick your own machine.
The result lists every step with its status, and a failing step shows the error, the screenshot and what the page looked like.
Run history is kept per test, so the next run compares against this one.
The checklist's last Tests step turns done when the first run finishes.

## 6. Keep it running

- On a schedule: the project's Test settings, then Schedules and CI, runs the tests nightly or hourly and emails or Slacks the people named when a run fails.
- From your pipeline or on every deployment: the same page gives a CI trigger token, and the GitHub deploy hooks run the regression after every successful deployment with a check on the commit.
- When a page changes on purpose and a test fails, FlowQA proposes a repair, proves it twice, and waits for your approval.
- With Snag in the workspace, a failing run files a bug with the evidence, and a bug can be reproduced as a test from its trail; the Bugs track of the same checklist gets the extension installed.
- Try Team: on Free, the checklist offers a 14-day trial of Team with no card.
