Do I need to know how to program?
No. A project is created with a plain form, a screen is captured with a button, and the comment is written in words — “the button is invisible on the dark background”. The agent then turns the comment into a rule: a small check. Programming is only needed if you want to extend a check or an action by hand — and even that is optional. Writing your own checks as scripts (Python code files) is a branch for those who have a programmer: without a single line of code, everything else works. Details are in the Rules section.
Will AI check my interface on every run?
No — and that is the main idea of the product. The model works only at the learning stage: it analyzes your comment and writes a rule. The checks themselves are plain code that runs right in the browser without calling the model. That is why a check is fast, costs nothing and does not depend on the agent being available. The more rules accumulate, the smaller the share of work left for AI. Details are in the Rules section.
What happens if the agent is unavailable?
Everything basic keeps working: projects, screen captures, comments, rules, actions, tests, runs, schedules, issues and the attention summary. Only the things that need the model become unavailable: comment analysis, action generation in words and auto-fix of broken checks. The chat with the agent is hidden too. Written comments do not disappear — they live in the project’s files and wait for the agent to return.
Do runs need my computer?
Yes, and this is worth knowing in advance. Checks are executed by a Chrome with the addon in runner mode — that is, a browser on a powered-on machine. The server does not open pages by itself. If no executor is online, the run is not lost: it waits in the queue until a Chrome appears. For scheduled checks it is convenient to keep a separate computer on all the time — the reliability of the schedule is exactly the reliability of that machine.
What if the computer was off at the scheduled time?
That moment counts as missed: the product does not catch up on it. If the server was down during the scheduled minute, no run appears for that moment at all. If the server was up but no Chrome was free, the run lands in the queue and executes once an executor appears. The next run fires at its own scheduled time.
Where do my screenshots and data go?
Nowhere: everything stays on your server. Nothing is sent to anyone else’s cloud, and you do not need an account with us. One honest detail: the screenshot the model references in a prompt lives at a public address without sign-in — that is how the model can download the image with a plain request. It is a deliberate trade-off for simplicity, which is why the Test Agent server is best kept in a closed network or behind a password. Details are in the Screens & Comments and Live Page & MCP sections.
Can I check pages behind a login?
Yes. For that, record an action once: the product saves the steps — the login and the moves between pages — and from then on those steps repeat on every run, even when the journey crosses several pages. Passwords never get into the files: they are stored encrypted in the project’s credentials and substituted only for the duration of a run. Details are in the Actions and Environments, Credentials & Variables sections.
How is this different from regular interface-testing services?
By three things. First: nothing is sent to someone else’s cloud — the server is yours, the data is yours. Second: rules and actions are plain files in your version store (git), not scenarios locked inside someone else’s interface; you can read them, edit them and take them with you. Third, and the main one: the model is needed only for learning. Regular services run AI on every check, so it costs money and time — here, after learning, code does the checking, and every new rule makes checking cheaper.
Do I need Xedant Agent?
For comment analysis, action generation in words and auto-fix — yes, that is the core of the product. It connects with an address and a key, after which the “Call agent” button appears in the panel. Without an agent the product still runs and executes already-written checks, but it cannot learn new ones. Details are in the Chat with the AI Agent section.
What does visual testing add on top of ordinary checks?
An ordinary rule catches what you described in words: whether the button exists, whether the text is right, where the link leads. But a rule will not notice that a block shifted twenty pixels, that a button became paler, or that a whole section disappeared. Visual checks compare the page build against a baseline — a sample captured and approved by a human — and find exactly those differences. Details are in the Visual Testing and Build Verification sections.
Who approves the differences?
Only a human. The AI suggests what changed and how much it looks like a real breakage, but the decision is always the reviewer’s. Bulk acceptance covers only certainly-safe differences — the ones the system confidently recognized as insignificant; everything disputed is handled one by one. Details are in the Review Board & Tasks section.
What does a visual check cost? Is AI called on every comparison?
No. Fast and precise rules work first: identical snapshots compare by fingerprint and cost nothing, and stationary pages do not disturb the model at all. The AI joins only when a difference is unclear — for example, a font changed across the whole site. The longer the project lives, the rarer the AI is needed. Details are in the Visual Testing section.
Can I show the report to a client?
Yes. There is a printable per-build report with a mandatory boundaries block: it honestly says that visual checks do not replace checking business logic and do not explain why the page rendered differently. The report exports to JUnit or JSON for the build pipeline, and for a client you can publish a share link — read-only, watermarked, with a term from one to ninety days. Details are in the Report & Share Links section.
How do I wire the checks into a build pipeline?
A build has a computable status: pending, failure or success. The pipeline calls the viz tool and gets a clear exit code, and the platform leaves a comment on the merge request. Webhooks (notifications to your address) report when a build needs human review or has failed. One important caveat: “no checks is not a pass” — an empty build does not count as successful. Details are in the CI & Notifications section.
Does the team need separate accounts?
Yes, if more than one person handles the checks. The administrator creates the users personally, there are no email invitations, and roles separate the access: the viewer only looks, the reviewer delivers verdicts, the member queues runs, the admin changes settings, the owner manages the keys. A disabled account cannot sign in, but its trace in the audit log stays. Details are in the Users & Roles section.
What does the license cost and what does it unlock?
The trial period is 30 days. The personal license costs
Can I extend the product with my own checks?
Yes. A script is a Python code file in the project folder, and it can become a test step, a snapshot comparison engine, a difference classifier or a finished-run handler. If every step of a test is a script, no browser is needed at all: such a test runs on the server, while the queue, the schedule and external access work as usual. An edit takes effect at once, without reinstalling the product. One honest caveat: there is no isolated environment — a script runs with the server’s rights, with its access to files and the network, so write your own scripts for yourself and do not run others’. Details are in the Scripts & Extensibility section.