For QA and engineering
Add AI agents to your test matrix
A site can pass every check for labels, named buttons and structured data and still lose agents at a custom size picker or a cookie banner that comes back on the cart page. The only way to know an agent can finish a journey is to send one through. Ghost Agents do that on a schedule, record every step and tell you where it broke.
What gets tested
Journeys, not just checks
Your key journeys
Start from checkout, product search, sign-up or a contact form, or describe your own goal in plain words.
Pass conditions you set
Reached the payment step, the page shows certain text, or the address contains a path. Checked directly, not by asking an AI model.
Several runs per check
Agents vary, so each check runs more than once and reports passing, flaky or failing. Flaky means some customers' assistants fail.
Every step recorded
The page the agent saw, what it did and its reasoning. A failed run's replay opens on the step where it got stuck, next to the last pass.
Alerts when it breaks
Failing tests alert your team by email or Slack, so you hear about a broken journey the day a theme update or new pop-up lands.
Readiness checks too
AgentScore checks bot protection, content without JavaScript, named controls and pop-ups, ranked by score gained for the effort.
How Ghost Agent Lab helps
Safe to run against production
- 1
Stops at payment
Ghost Agents stop at the payment form, never enter card details, and refuse buttons that place orders, pay, or delete or cancel accounts.
- 2
Stays on your site
An agent can only visit pages on your site, and uses the test email and values you give it.
- 3
Identifies itself
Every request says GhostAgent/1.0 in the user agent and is signed with Web Bot Auth, so bot protection and analytics can recognize and filter it.
Further reading
Guides for QA and engineering teams
FAQ
Questions
How is this different from our end-to-end tests?
Your end-to-end tests follow a script you wrote against selectors you know. A Ghost Agent is given a goal and finds its own way, the way a customer's AI assistant would, so it trips over the unlabelled button, the overlay or the account wall your script was written to avoid.
Will tests place real orders or create junk data?
They never pay or place orders. Sign-ups, leads and bookings they submit are real on your side, so use recognizable test values, such as a “test+” email address, and test risky journeys on a staging copy.
How do we keep Ghost Agents out of our analytics and bot rules?
They identify themselves with GhostAgent/1.0 and sign every request with Web Bot Auth. Tell whoever runs bot protection and analytics before you start, and filter on either.
Can an AI agent finish your checkout?
Scan your site free to clear the obvious blockers, or start a 30-day trial and schedule your first Ghost Agent test.