Frequently asked questions
Quick answers to the things people ask most often. If your question isn't here, head to the contact page or open the chat in the bottom-right corner.
How are agent hours calculated?
Agent hours measure the wall-clock time our AI agents spend running flows for you — navigating, clicking, taking screenshots, and evaluating assertions. Idle queue time doesn't count. Discovery, recovery, and assertion verification all count, since they're what produce the value.
Can I bring my own AI API key?
No. Qassandra uses a combination of in-house, purpose-built models and frontier third-party models, switched and tuned per task. Bringing your own key would lock you out of the in-house models and the smart routing layer, which is where most of the quality gains live.
Does Qassandra need access to my source code?
No. The agent only ever talks to your app over HTTP, exactly like a real user would. The GitHub App reads the list of changed files in pull requests for smarter flow selection, but it doesn't read or store the file contents.
How does Qassandra handle apps behind a login?
Add reusable test credentials to Agent notes in app settings. The discovery agent and the runner read the agent notes before each session. For more complex auth flows (SSO, magic links, MFA), describe the steps the agent should take in the same notes — natural language is fine.
What happens when my app changes?
On every check, individual steps that fail because of a moved or renamed element are recovered automatically by the AI runner. The new selector replaces the old one in the flow, so the next check is fast again. Bigger changes (new pages, removed sections, redesigns) show up either as new flow suggestions or as visual regressions — both surfaced on the check report.
How do I avoid flaky tests on dynamic content?
For pages that show timestamps, recent activity, or randomised previews, the visual regression engine recognises dynamic content and resolves it as expected noise rather than a regression. For pure text assertions you can add an instruction in the flow (e.g. “ignore the relative time labels”) and the AI verifier will respect it.
Can I run Qassandra against localhost?
Not directly — the cloud runner needs to reach your URL. The usual workarounds work: expose your local server with a tunnel like ngrok, point Qassandra at the tunnel URL, and you're set. Most teams instead point Qassandra at preview deployments (Vercel, Netlify, Render), which give you a stable per-PR URL.
How fast are checks?
Individual flow steps run at native Playwright speed (sub-second per step). A typical flow of 10–15 steps finishes in a handful of seconds. A check that fans out across many flows runs them in parallel, so wall-clock time is closer to the longest flow than the sum. AI recovery only kicks in on failure, so it doesn't slow down the happy path.
Can I upgrade or downgrade my plan?
Yes — at any time. Upgrades take effect immediately. Downgrades take effect at the start of the next billing cycle so you keep the higher limits for the period you've already paid for.
Do you offer enterprise / VPC deployments?
Yes. The Enterprise plan supports VPC deployment, custom agent rules, SSO, and dedicated support. Get in touch and we'll set up a call.