Find out who can actually write the code.

Python. The candidate writes and runs it in the browser, and the submitted answer is graded on our servers against hidden tests, inside execution limits of 12 seconds and 1MB with an import denylist.

For any other stack the coverage comes from single-choice and multiple-answer questions, and the builder drafts them in the language you are hiring for. A senior Go role gets Go questions. What it does not get is Go running in the sandbox.

Counting languages means maintaining a sandbox for each one. We maintain a single execution path and put the effort into the question instead. Whether that trade works for you depends on whether running the code is the part of the screen you actually need.

They can, and we do not claim to stop it. No browser-based assessment can, and any page that says otherwise is selling you something it cannot deliver.

What the product does is record the behaviour that tends to come with it and put it in front of you: paste events from the clipboard and from drag-and-drop, tab switches with the time spent away, fullscreen exits, the cursor leaving the window, blocked shortcuts, network drops, and second-screen presence where the browser reports it.

Coverage is browser-dependent, and that matters here. Second-screen presence is reported by Chromium browsers and is unavailable in Firefox and Safari, and on its own it never marks an attempt. A person reads the flags in context and decides.

The pattern most teams settle on: use the assessment to filter, then spend fifteen minutes talking to the candidate about the code they submitted.

No. There is no recorder in the product — no capture, no upload path, nothing stored. Our candidate privacy notice says so in writing.

With the candidate’s explicit consent, a small face-detection model loads from our own origin and runs on their own device. It counts faces. No frame, no image and no biometric template leaves the browser; what reaches us is counts and durations, of the form "no face detected for 40 seconds".

It is also not identity verification. The model does not match anyone against anything, and there is no ID-document step.

No, and it is prevented from trying. If a model response comes back containing a score, an integrity judgement or a hiring recommendation, it is discarded on our servers before a recruiter ever sees it. That refusal lives in the code, and it is how we stay outside the automated-decision rules in GDPR Article 22.

Grading is deterministic: fixed answer keys for the choice questions, hidden tests for the Python task. An optional pass mark sorts graded attempts into pass and fail — you set the threshold, and candidates never see it.

Telling a candidate the outcome is always a human click. There is no automatic rejection anywhere in the product, by design.

All persisted candidate data is stored in Frankfurt — the application and the database — and error tracking sits in the EU, in Vienna.

Inside the product we publish a sub-processor register naming every vendor, what it processes and where it sits, including the ones outside the EU. Read that instead of taking a badge from us.

One caveat, because it is the one people miss: the optional AI test-creation feature sends the text you paste in — a job description — to a model provider outside the EU. That step runs before any candidate is invited, and it is optional. Build the test manually and it never runs.

A desktop or laptop browser, and nothing installed. No plugin, no desktop app, no local Python — the editor and the runtime are served from our own origin, and the assessment makes zero cross-origin requests.

A compatibility check runs before the assessment starts, so an unsupported browser is caught up front instead of halfway through a timed test. Phones and tablets are out: on a touch-only device fullscreen enforcement, cursor tracking and second-screen reporting are structurally inert.

There is an unscored practice question first, so the editor and the Run button are familiar before anything counts. It is not a rehearsal of the real conditions: practice has no timer and records no integrity signals.

Once your workspace is approved, you choose a plan inside the product and pay by card. Payments are processed by Paddle.com, our online reseller and Merchant of Record: Paddle charges the card, adds VAT where it applies (reverse-charged for EU businesses with a valid VAT number) and issues the invoice. Plans are monthly, you can cancel any time from the Billing page, and your first payment comes with a 14-day money-back guarantee — the refund policy linked below has the details.

Results leave the product as CSV, Excel or PDF from Core up, and through signed webhooks with a delivery ledger and automatic retries. Per-company API keys and the public REST API come with Pro.

Traffit is the only ATS connector we have, so it is the only one we name: we set it up on request for Pro customers, and then moving a candidate into the test stage sends the assessment and the result is written back. From Core up, Slack works by pasting a webhook URL, and Zapier and Make work too — the Zapier app is still private, so we share it with you directly rather than through the directory.

Yes, and that is what the demo is. Bring a live job description. We build the assessment for that role with you on the call, and you preview it exactly as the candidate will see it — answer key stripped, no attempt written.

Or try it on your own: start a free trial — sign up, we approve your workspace, then you get 7 days with everything unlocked before you pick a plan.