Start an exploration

An exploration sends the exploratory agent through one of your environments in a cloud browser. It maps the pages it finds, compares them with your existing test cases and proposes new and updated tests for you to review.

You choose where it starts, how many pages it may visit and whether it may interact with the site. Nothing in your project changes until you apply the proposals you accept.

The agent visits pages one by one, up to the page budget you set. On each page it takes a screenshot and records what the page is for, its main actions, its forms and any errors it shows. Then it compares what it saw with your test suite.

For example, Pellbrook Digital points an exploration at the Toolshop’s Production environment with the focus “Concentrate on search, catalogue filters, the cart and checkout”. The agent maps the pages it reaches from the home page, and proposes tests wherever their suite has gaps. See Review and apply proposals for what to do with the results.

• Role: starting an exploration needs the Editor role or above. Every member of the organization can open explorations and read their results.
• An environment: the project needs at least one environment. The agent opens it in a cloud browser, so the site must be reachable from the internet. See Environments.
• A login, if your app needs one: to explore pages behind a sign-in, store a credential first. See Credentials and automatic login.
• Credits: your organization needs a credit balance above zero to start an exploration. At zero, Testmode refuses with “Your organization has no credits left. Buy credits to start an exploration.” See Buy credits.
• One at a time: a project can have one exploration in progress. Wait for it to finish, or stop it, before you start another.

1. Click Explore in the sidebar, or press G then E.

2. Click New exploration, or press C.

   The New exploration dialog opens with the project’s default environment selected.

You can also open the dialog from the command palette: press Cmd+K on a Mac or Ctrl+K on Windows and Linux, then choose Start exploration. On the Assistant page, the suggestion Explore my site and suggest test cases opens it too.

While an exploration is in progress, New exploration is disabled and its tooltip reads “An exploration is already running”.

Environment is the site the agent explores. The dialog shows its Base URL underneath, and a PROD badge on production environments.

Login Credential is optional. It shows None (no login) until you pick a stored credential, then “Will log in as” and the credential’s username.

• With a credential, the agent explores as that user. It starts from the login Testmode saved for the credential, or signs in first when there is none or it no longer works. If sign-in fails, the exploration ends as Failed with “Login failed: …”.
• Without a credential, the agent still records a page that asks for a login, marked “Requires login” in the site map, but it does not follow the links behind it.

The picker lists every credential in the project, including ones scoped to another environment, so check that the one you pick works on the environment you explore.

While an exploration runs with a credential, it counts toward that credential’s Parallel browser sessions limit, so tests that use the same login may wait until it finishes. See Credentials and automatic login.

Field	What it does	Toolshop example
Start from	The paths where the agent begins, comma separated. It follows links from there. Defaults to /.	/, /contact
Page budget	The most pages the agent visits: 10 pages, 25 pages (the default), 50 pages or 100 pages.	25 pages
Propose updates to existing test cases	On by default. The agent also suggests fixes for test cases whose steps no longer match the app, or that look stale. Switch it off to get only new proposals.	On
Focus (optional)	Plain-English guidance on what matters most. It also appears under the exploration in the list.	Concentrate on search, catalogue filters, the cart and checkout
Skip paths containing (optional)	Paths to leave out, comma separated.	/admin, /export

Good to know:

• Skip paths match any part of the path, in any letter case. /admin also skips /administrators.
• The budget is a maximum. The agent also works within a time budget. If that runs out first, it stops visiting pages and still analyzes everything it found. A large budget on a small site may not be used up either.
• The analysis shares that time budget. If too little time is left after the crawl, the exploration fails with “Analysis failed: …”. Start again with a smaller page budget.
• Similar pages share a shape. Pages that differ only by an id, such as two product pages, count as the same shape, and the agent visits only a few pages of each shape.
• The crawl follows links on the environment’s own address only. It skips links to other domains and subdomains, file downloads and sign-out links, and a redirect to another domain ends that path.

Under How far may the agent go?, choose what the agent may do on each page.

Mode	What the agent does	Use it for
Observe (default)	Navigates between pages and reads them.	Any environment, including production.
Interact	Everything Observe does. On a few of the most interactive pages it also tries interactions: it is instructed to open menus and dialogs, try search, filters and sorting, and submit forms empty or with invalid values to see the validation messages. Its findings appear as Observed behaviour on those pages in the site map.	Test and staging environments, when you want proposals for validation and edge cases.

In Interact mode the agent is instructed never to delete, pay, send or publish anything, and not to sign out or leave the site. These are instructions to the agent, not technical locks.

When you pick an environment marked as production, the dialog switches to Observe, greys out Interact and shows an amber warning that production is explored read-only. Testmode applies Observe to production explorations on the server as well, so a production environment is never explored in Interact mode, whichever way the exploration was started.

In Observe mode the agent only navigates and reads, with two exceptions:

• it submits the login form when you choose a credential
• it dismisses a cookie banner, once per exploration

See Mark an environment as production.

1. Check Environment and Start from. Start exploration stays disabled until both are set, and the footer reads “Pick an environment and at least one start path”.

2. Click Start exploration.

   Testmode shows “Exploration started” and opens the exploration’s page.

The exploration’s page updates by itself as the agent works.

• Heading and progress bar. While it runs, the header shows “X of Y pages visited” against the budget, and how many more pages it has discovered.
• Live browser view. While the agent is exploring or trying interactions, you can watch the cloud browser. The view is watch-only: your clicks do not reach it. Click its header to collapse it.
• Site map. The page opens on the Site map tab, which fills with a card for each page as the agent visits it.
• Proposals. They are written once the agent has compared the site with your suite. The page then switches to the Proposals tab.

The Explore list shows every exploration in the project with Exploration, Status, Pages, Proposals and Started. While an exploration runs, Pages reads visited over budget, such as 12 / 25.

Status	Heading on the page	What is happening
Queued	Waiting for a browser	Waiting for a free cloud browser, or for another sign-in with the same credential to finish. The page shows “Waiting for a browser session…”.
Exploring	Exploring the application	Visiting pages and filling in the site map.
Probing	Trying interactions	Interact mode only: trying menus, filters and form validation.
Analysing	Analysing coverage	Comparing what it found with your test suite and writing proposals.
Ready for review	Ready for review	Proposals are waiting for you.
Applied	Applied	At least one proposal has been applied to the project.
Failed	Exploration failed	The exploration could not finish. The page reads “Stopped after N pages.” and a “What went wrong” box gives the reason, or “The exploration could not finish.” when there is none.
Canceled	Canceled	Someone stopped it. There are no proposals.

An exploration can go back to Queued for a moment after it started, for example when no browser is free or Testmode restarts. It then starts again from the beginning, and the site map fills in again.

A failed exploration explains itself. For example:

• “Login failed: …”: the agent could not sign in with the credential you chose.
• “No pages could be visited. Check the environment URL and start paths.”
• “Analysis failed: The analysis did not finish within the exploration’s time budget (…). Reduce the page budget and try again.”

Both need the Editor role or above.

To stop an exploration in progress, click Stop on its page, or open the row menu in the Explore list and choose Stop exploration. Testmode shows “Exploration will stop after the current page”.

• A Queued exploration is canceled straight away.
• An exploration that is Exploring or Probing stops before its next page and ends as Canceled. No analysis runs, so there are no proposals.
• An exploration that is already Analysing finishes the analysis, then discards it and ends as Canceled.

To delete a finished exploration, open its row menu, choose Delete, then click Delete in the Delete exploration? dialog. This removes the exploration, its site map and its proposals. Test cases, plans and variables you already applied from it stay in the project. Deleting cannot be undone.

Stop an exploration before you delete it. Delete appears in the row menu only once the exploration has ended, whether it is ready for review, applied, failed or canceled.

You can ask the Assistant to explore for you, for example: “Explore the Production environment and tell me which tests we’re missing.” It is instructed to confirm the environment with you first, and the credential if the app needs a login.

• Explorations started from chat always use Observe mode.
• The Assistant can set start paths, the page budget, a focus and whether to propose updates. It cannot set paths to skip.
• Later, ask “How is the exploration going?” or “What did it propose?” to get its progress and proposals in the chat.

Starting an exploration from chat needs the Editor role or above and a credit balance above zero, as in the app. See Create tests with the Assistant.