The section reference — from installation to fine tuning: how to capture a screen and turn a comment into a rule, how actions are recorded and tests are assembled, and how schedules, issues and auto-fix work. A separate large part is visual checks: the product compares every build against an approved baseline, shows the differences as a picture, and leaves the decision to a human. There is also review, failure triage, coverage, the report, working with continuous integration, and operations. Start with the installation and the first steps, then open the section you need right now. The product can be extended with your own Python scripts — that branch is optional, and without it everything works as before.
Getting started
Docker Compose, environment variables, where the data lives, first sign-in.
What the product consists of: the path of one comment and comparing builds against a baseline.
The project folder, the base address, the timezone, screen types, knowledge.
Interface checks
One-click install, the panel on the page, runner mode.
A snapshot of the visible area and comments in plain words right on the page.
A comment turns into a check that works without the model.
Recording logins and journeys, replay, passwords in variables.
A sequence of steps: open an address, run an action, check.
The test’s version history, the step page and export.
The run queue, a snapshot on every step and the final verdict.
Days and hours: when checks run by themselves.
Identical errors merge into one issue until it is closed.
The key-authenticated address your coding agent sees issues through.
A broken check is repaired by the agent itself and flagged for review.
The global numbers on the Projects screen and the badges on each project.
Connecting Xedant Agent and the “Call agent” button.
The model works with a real open tab — directly or through MCP.
Your own Python scripts: a test step, a comparison engine, a classifier, a run handler.
Visual checks
What a visual check is: a step snapshot, comparison against a baseline, the comparison cascade.
The Verify tab: builds, modes, review states, the build page.
Viewing differences: side by side, overlay, a slider; change blocks and the explanation.
Baselines: versions, variations, moving between branches, storage.
Review and triage
The review board and todos: verdicts, roles, bulk actions.
The audit log: who approved what, and on what grounds.
Build-failure triage: groups by cause and a plain-language explanation.
Test healing: proposal, approval, verification, rollback.
The AI log: how many tasks and tokens were spent, and why requests were refused.
Coverage and agent campaigns
Coverage: the surface-by-state inventory and progress toward the goal.
The app map and crawling: routes found by the browser, and protecting production.
Agent campaigns: scouting, the plan, generation, verification.
Settings and access
Presets, thresholds, check layers, branches, capture stabilization.
Environments, credentials and variables: different staging servers and secrets.
Users and roles: who can do what, and who approves design.
API keys for CI, the viz tool and MCP.
The visual-testing setup checklist — eight steps.
Report, CI and operations
The build report with the boundaries block, and share links with a watermark.
The build status for CI, the viz tool, webhooks and notifications.
Runners, the queue, budgets, health checks and data export.
