Explore lets the AI use your app on its own for a few minutes and tells you what looks broken. You pick a device, a time budget and, if you like, a goal. You get a list of findings, each with a screenshot, the moment in the run’s video, and what the AI did before it. Nothing is saved as a test unless you choose to: each finding has Save as test.
Explore works on iOS and Android phones and tablets, and on Apple TV, Android TV and Roku, where the AI moves focus with the remote the way a person would.
Start an Explore run
You need the Developer or Admin role.
- On the Tests page, click Create test and open the Explore tab.
- Or, on an app’s page, click Explore this app.
Then choose:
- Device — one device for the app, by its usual name (for example Pixel 9). It starts on a phone for a phone app, the same pick as the Run dialog; choose a tablet if you want one. If your plan’s concurrent sessions are all in use, the run waits for a free one.
- Time budget — 5, 10 or 20 minutes. The clock starts when the device is ready, and the first few seconds go to waiting for your app to open.
- Goal (optional) — what to focus on, for example “Try every way to start playback and report anything broken”. Leave it empty to explore the whole app. Up to 1,000 characters.
Below the form, Explore names the AI model it will use and who pays: your usage credit or your own API key. No price is shown before the run. Click Start exploring: the run’s page opens.
The AI reads your app notes and can type your test data. It sees a secret only as its name, such as {{password}}: the value is typed for it and never shown to it, and a secret typed into the goal is masked before the AI reads it. The AI is told never to invent passwords, emails, card numbers or codes, never to buy anything, delete data or message real people, and to stay in your app.
Free plan: Explore is not part of the free test. It needs usage credit or your own API key.
While it runs
The run’s page shows the device live, the findings as they come in, and What the AI did, a list of its actions with their times. The run ends when:
- its time budget is used,
- it reaches the action limit for its budget or its spending safety limit,
- the AI has covered what it can reach,
- you click Cancel (Developers and Admins), or
- your usage credit runs out or AI is paused for your organization — the findings so far are kept.
If the AI keeps working the same button or field, it is told to move on. On a TV, pressing the same remote button again and again when nothing on screen changes (for example Right at the end of a row) counts the same way; presses that move focus don’t. If it goes on anyway, VibeView stops carrying out that action for a while and tells the AI to pick another screen or control; those turns appear greyed out in What the AI did, marked “Not carried out”. This never ends the run.
The spending safety limit only stops a run that runs away. It is set from the AI model’s price, so a pricier model gets a higher limit, and it never stands in the way of your choice of model. The run’s page shows its cost against this limit. If a run stops there, its page says so and links your AI settings: try a longer budget or a cheaper model. With your own API key the limit is based on an estimate, and a model VibeView has no price for only gets the time and action limits.
When a run stops early, its page says why and what to do about it:
- Stopped by VibeView, not your app — the device session ended, or the AI service didn’t answer.
- Setup failed — the app couldn’t be installed or prepared on the device, for example a build the device can’t run.
- Couldn’t start — the run never got a device (for example, your plan’s streaming minutes ran out while it waited), or the AI model isn’t available.
- Out of usage credit, AI paused or API key rejected — with a link to buy credit, add your own API key or check it.
Developers and Admins also get Explore again, which opens the form with the same device, time budget and goal. A run that finishes or stops sends the person who started it a notification and an email; cancelling it does not.
Findings
Each finding has a kind:
- Crash — the app closed, went blank, or another screen took over.
- Error screen — an error message or error screen.
- Broken element — a control that does nothing, cut-off or overlapping text, an empty placeholder.
- Can’t leave the screen — several actions in a row, Back included, changed nothing.
- Error in the app’s log — your app logged an error after something the AI did. The lines are shown with the finding, with secret values masked, along with the screen at the time. Errors your app already logs while it starts, before the AI’s first action, are not reported, and neither are messages the operating system’s own frameworks log on your app’s behalf.
On a TV, moving focus is how the AI navigates, so it reports “can’t leave the screen” only after Back has been tried too.
The same problem seen again is one finding with a count (“Seen 3 times”), not a new one: a crash or an error in the app’s log counts once wherever it happens, other kinds once per screen. If a run finds more than it keeps, the page says how many more weren’t kept. Show me plays the run’s video from just before the finding, once the video is ready; otherwise the finding’s screenshot shows what the AI saw. Permission prompts and system alerts are not reported.
Screenshots are stored as the device showed them, as in test runs: a secret that was on screen is not masked in the image.
Viewers can read a run and its findings. Cancelling and Save as test need the Developer or Admin role.
Save as test
Save as test opens the finding in Review steps, the same screen Describe and Record end on: one step for each thing the AI did before the problem since its last app restart (a run of remote presses becomes one step), then a Verify step for what a working app would show. Edit anything; nothing is saved until you click Save, and the new test opens.
When the test is saved, the steps you kept as they were replay the route Explore took, so they use no AI on the next run (see learned replay paths). From the first step you changed, the AI works out the steps until a passing run learns a route. The route is saved step by step and stops at the first step that can’t be saved as a route: one that typed a secret, whose screen shows one of your test data values, that rotated the device or changed its posture, or that you edited or moved. The steps before it keep the route, and from that step on the AI works them out. Nothing is attached, and the first passing run learns the paths instead, when:
- the run took more than 80 steps to reach the problem (only the last 80 are kept);
- the device was a foldable;
- the test is set to ask before saving paths, or never to save them.
After Save, a message says how many steps got the route, or that none did.
Cost
Explore uses AI on every turn. With VibeView-provided AI, the run’s page shows the usage credit it actually used; with your own API key, VibeView charges nothing and shows an estimate. Next to the cost is the run’s spending safety limit. See Usage credits. The device time counts toward your plan’s streaming minutes like any session.
Where to find past runs
Explore runs are listed in the Tests page’s Runs section as Explore, with how many findings each found. They never count toward pass rates, flaky tests or Needs attention.
An Explore run, its findings and their screenshots are kept for 30 days, like run videos (for runs that have one) and logs, and then deleted. A test you saved from a finding stays.