Concepts
The handful of terms you need to use Agentiqa, in plain language.
Project. One app under test: its URL plus what the agent learns about it (map, memory, credentials, plans, runs, issues). Pick one with project use, --project or AGENTIQA_PROJECT_ID.
Explore. An open-ended session. You give a goal, and the agent navigates a real browser, proposes scope and a plan, and reports findings. Use it to discover what to test.
Checkpoints. Scope → Plan → Findings. In the app you approve or redirect at each one. Headless runs approve automatically.
Test plan (tp_…). A saved, replayable sequence of steps with expected results. Today's exploration becomes tomorrow's regression check. Group plans with labels to run a subset.
Gate. A test plan (or labeled set) that runs in CI and decides whether a release ships. Agentiqa doesn't treat a gate as a separate object. It's your plans plus the exit code.
Run. One execution of a plan: per-step pass/fail, a verdict, video and screenshots. A deep link (runUrl) opens it in the web app.
Issue. A finding with screenshot and repro steps. Fix it, then retest from the issue.
Exit codes. 0 all passed · 1 a plan failed (real signal, don't retry) · 2 usage, config or plan-limit problem, or skipped plans · 3 infra error (safe to retry). Add --json for a single machine-readable document on stdout.
Service key. A project-scoped key (AGENTIQA_SERVICE_KEY) for unattended CLI and CI runs. Keep it in a secret. Never put it in chat or on the command line.
Credentials, shared accounts, testing mailbox. How the agent signs in. It uses stored logins (secrets stay out of the model's context), a pool of role accounts several runs can lease at once, and a per-run inbox for verification emails and magic links.
Engine. Where the browser runs. Local runs use an in-process engine on your machine. Service-key runs default to the hosted engine. Override with --engine.