Ask in plain English. Your test project answers.
A conversational agent lives in your dashboard. It searches and edits your test cases, builds cycles, records results, and opens any report — the coverage gaps, the flaky tests, the risk matrix — the moment you ask. Press Ctrl+K to start.
Built into your dashboard · nothing to install · press Ctrl+K
Create test cases for the guest checkout flow
Working
Read the project structure — Checkout folder, allowed fields
Created “Guest checkout — happy path”
Created “Card declined shows an error”
“Blockd” is not an allowed status here
Re-read the allowed values, then retried
Created “Payment timeout is retried”
The report that answers your question already exists
Almost nobody opens it. Finding the right one means knowing which report, where it lives, and which filters to set — so the question that would have changed your week goes unasked. Ask the agent instead, and the answer costs one sentence.
Ask in plain English. It does the work in your project.
Find anything
Search your test cases by keyword, status, priority, label or folder, and open one in full.
Create and edit
Write and update test cases, move and rename folders — one at a time or in bulk.
Cycles and plans
Build test cycles and plans, and record results as pass, fail, blocked or not executed.
Open any report
Coverage, uncovered requirements, defects by folder, tester workload, execution trends — by asking, not navigating.
The insight reports
Flaky tests, release readiness, risk heatmaps, stale tests, project health — the ones most likely to sit unused.
Nothing to install
It runs in the dashboard on your session. No extension, no proxy, no setup — press Ctrl+K.
Attempt1
{
"folder": "Payments",
"title": "Card declined is handled",
"priority": "Highest"
}priority is not an allowed value
Allowed: Blocker, Critical, Major, Minor, Trivial
Re-read the project’s priority field
Attempt2
{
"folder": "Payments",
"title": "Card declined is handled",
"priority": "Major"
}Created TEST-247
It works from your project, not a guess
The agent reads your project’s real structure — the folders that exist and the values each field accepts — so it fills in what is allowed, not what sounds plausible. Reject a write and the error tells it how to fix it, so it re-reads the options, retries, and gets there instead of stopping.
Frequently asked questions
No. It records results — it writes down that a test passed, failed, was blocked or was not executed. It never opens a browser, starts a session or touches a device; running tests on real hardware is what Live and Automate do. The advantage of one product is that the cases it writes and the results it reads come from those real runs in the same workspace.
Test management: your test cases, folders, cycles, plans, executions and the whole report set. It works through a fixed set of tools, not free-form access to your database. It does not reach your accessibility scans, your visual baselines or your device sessions.
Read the guide: What happened when we gave our tool an AI agent →It reads your project's real structure first — the folders that exist and the values each field accepts — so the allowed answers are discovered rather than guessed, and those lookups are cached briefly so asking is cheap. When a write is rejected, the error it gets back carries the recovery step, so it re-reads the options and retries instead of apologising and stopping.
Yes — that is the point of it, and it means a badly phrased instruction can create or modify records. Read what it proposes before you confirm, as you would a colleague's bulk edit.
It uses your organisation's shared AI credits, and the model follows the quality level your owner sets. When the credits run out the agent stops, and usage is visible to owners and admins by month.
Yes. AI is an organisation-level setting an owner can switch off entirely, which matters where a contract forbids sending content to a model provider.
The same tools, a different seat. The agent lives in the dashboard on your session and your organisation's AI credits, with nothing to install. The MCP server puts the same capability inside your own AI tool — Claude, Cursor or another client — on your own model subscription, next to the branch and ticket you are already working on. Most teams end up using both.
Open the panel and ask it something
Press Ctrl+K in your dashboard, ask for a report or a batch of test cases, and watch it work.