AI Smart Verification

What it does

AI Smart Verification is the AI side of the GTS Engine + AI, the engine that finds and compares what is on your screen.

When a visual check finds different pixels, or a text check cannot read its text, AI Smart Verification takes a second look and gives the final verdict.It passes the check when nothing major changed, fails it when something did, and adds a sentence a person would write: what changed, where, and how much it matters.

Enlarge

Enlarge

What the AI ignores

The mouse pointer and the blinking text cursor never count as a difference.For a text check the AI reads the wording alone: colours, fonts and styles do not matter.

Minor differences

A check passes when the AI finds no major change.If it found only small changes a user may not notice, the report says "only minor differences were found".To fail such checks instead, turn on Count minor differences as a failure in the AI settings (for every test on this computer) or in a test's Project Settings (for that test).

Text checks

Every kind of text check gets its own question.For a count, the AI counts how many times the text appears and the test's own count rule decides.For a changing value, the AI is told the kind of text you chose, such as a date, a time or your own pattern, and passes any value of that kind.

Your own instructions for a check

Some differences are expected, and only you know which.Give a check AI Instructions: a few words in your own language that the AI follows when it judges that check, before its usual rules.

1. 1

   Select the check in GUI Test Studio

   A Visual Assert, a group member, or any kind of Text Assert: the checks the AI looks at.

2. 2

   Open Advanced and type the instructions

   The AI Instructions field is under Advanced, on the check's own tab of its Properties.Press Ctrl+Enter or click elsewhere to keep them.Select several checks of one kind to give them the same words at once.

The report shows the instructions a verdict was given under, beside the AI's answer.When you change them, the next report asks the AI again instead of reusing its earlier answer.

Which AI model: your choice

GUI Test Studio does not tie you to one AI model.The GTS Engine + AI asks the same question and returns the same kind of answer whichever model you choose, so you can decide by cost, by privacy or by your company's rules, and change your mind at any time.

Not sure where to start?Contact us: we help you choose.We can also serve a cloud model for you, as an optional service priced separately from your subscription.

Cloud or local

Set it up in the cloud

1. 1

   Open the AI settings

   Click the Enable AI Verification button at the top right of either app, or open Settings, Preferences, AI.One setting serves both apps.

2. 2

   Switch it on and choose In the cloud

   Pick the provider: OpenAI or Anthropic.

3. 3

   Paste your API key

   No key yet?The link under the key field opens your provider's key page in your browser, and the ? beside the field walks you through creating one.

   Create a key in your provider's account and paste it.The dot beside it turns green once the provider accepts it, and the list of models fills in by itself.The key is stored encrypted, for your Windows account only.

   

   Enlarge

Set it up on your computer

A local model runs on your own computer: nothing leaves it, and there is no key or account.You can use a model we provide, or bring your own through LM Studio or any OpenAI-compatible local server.For a model we provide there is nothing to install: everything needed to run it is part of GUI Test Studio, and the model itself is downloaded once, when you ask for it.

1. 1

   Choose On this computer

   In the same AI settings, under Run the AI.

2. 2

   Click Set up for me

   The settings look at your computer's graphics card and memory and recommend the best model it can run well.Set up for me downloads that model and selects it.A progress bar shows how far it is, and a download you cancel resumes where it stopped.

3. 3

   That is all

   The model starts by itself when a run needs it.To use another one, pick it under Download a model; Remove... deletes a model you no longer want and gives its disk space back.

Choosing a local model

Two things decide how well a local model serves you: whether it fits in your graphics card's memory, and how large it is.A larger model reads small text and subtle changes better.A model that fits entirely on the graphics card answers in seconds.

The list marks each model for your computer: it runs on your graphics card, it runs partly on the processor, or it is too large.Choose the largest model that still runs on your graphics card.That is what the recommendation does for you.

Precision

Some models come in several precisions, listed under Precision.Standard is the smallest download and the fastest.A higher precision is a little more exact, takes more of the graphics card's memory and answers more slowly.Keep Standard unless your card has memory to spare.

Under Download a model, the list shows every model you can run, with its size and whether it fits your graphics card.Pick one, click Download, and it appears in the Model list.

Already have a local AI server?

Pick Local server in the Provider list and enter its address.Tools such as LM Studio, Ollama and vLLM work this way.

Rather not run a model yourself?

As an optional service, we can run AI Smart Verification for your team on our own service: nothing to download, nothing to set up.It is a separate service with its own price, not included in the monthly subscription of any plan.Contact us to arrange it.

In the report

The explanations start as soon as a run ends, before you open anything, and the check you open goes first.A failed row carries an AI mark, and the check's panel shows the assessment, a summary and every difference found: how much it matters, what kind of change it is, what was expected and what was found.

The result has the same shape whichever model answers, in the cloud or on your computer.

Enlarge

Explanations are kept with the run, so reopening a report asks nothing again, and History shows what the AI said in each past run so you can compare them.They also appear in bug reports and in the results of command-line runs.

Privacy

AI Smart Verification is off until you switch it on.With a cloud provider, only the two small pictures of a failed check travel, directly from your computer to the provider you chose, with your key: never to us.A passing check is never sent.With a local model, ours or yours, nothing leaves your computer at all.