Coderbyte wins on volume. We win on what a screen actually tells you

They lead on timed screens, question libraries, and webcam/device proctoring — and on huge shared plagiarism databases. DevParser is built for real take-homes: hidden tests + Claude review, paste-blocking editor telemetry, optional public-git process evidence, and same-company peer similarity. We are intentionally not a proctoring clone. Their libraries are also their weak spot: a shared bank of thousands of questions is a bank that leaks, which is why they run leak-monitoring services against it — DevParser generates the task per candidate, so there is nothing to leak.

If you want them to. Each invitation is either AI blocked (the default; paste is refused and logged) or AI allowed. In allowed mode the candidate gets an assistant inside the editor, up to 30 messages, which never sees your tests or reference solution. You read the whole conversation next to the code, and the review adds an AI-fluency note on whether they broke the problem down, checked the answers and fixed the assistant's mistakes. That note sits beside the score and never changes it. AI-allowed mode is on Growth and above, and on the 14-day trial so you can try it first. The assistant's usage comes out of your plan's AI budget; there is no per-message charge.

Yes — the UK is where we launched. Plans are priced and billed in pounds, month to month. Our data processing agreement covers UK GDPR and includes the ICO International Data Transfer Addendum; candidate data is hosted in the EU and never used to train models. For recruitment agencies, your logo goes on the candidate report from Starter, and Bullhorn, Vincere and JobAdder connections are in beta.

Every plan includes the paste-blocking editor and server-scored session telemetry, and the reviewer sees each time the candidate ran the visible tests. Candidates can also submit a public git repo so commit history becomes process evidence for take-homes. Growth, Professional and Enterprise add peer similarity and public repository search. Style alone never decides and we never auto-reject: reviewers confirm or override each flag and the decision is logged. On Growth and above you can also allow an AI assistant per invitation: the candidate works with it inside the editor, you read the full conversation, and the review includes an AI-fluency assessment. Optionally the candidate answers a few written questions about their own submission; the answers and an advisory AI reading of them go to the reviewer and never change the score.

We run the code against your hidden test cases, then Claude analyses the outcome. You get objective results (pass/fail, runtime errors) plus scoring on quality, efficiency and best practices. That beats reading the code alone.

DevParser pairs hidden-test execution with AI-assisted review. You get hard test outcomes plus a summary of correctness, quality, efficiency and risks. Your team can always review and adjust the score — the AI supports your judgement, it doesn't replace it.

Yes. We integrate with 11 ATS platforms including Greenhouse, Lever, Workable, BambooHR, Ashby, Teamtailor, Recruitee, eRecruiter, Traffit, Recruit CRM, and more. On every paid plan, agentic screening auto-syncs candidates, generates personalised coding tasks, and queues AI evaluations. You set daily limits; DevParser handles the rest. Agency CRMs Bullhorn, Vincere and JobAdder are available in beta.

Both. A pair session opens a shared editor and gives you one link to send — the candidate joins with no account and no install, you both type in the same document in real time, and the session is saved when it ends. That is on every plan, including the trial. The take-home is what we screen with; the live round is where you decide.

200 curated tasks across backend, frontend, data and QA, in Python, JavaScript, TypeScript, Java, C#, Go and Rust. 196 of them are graded by hidden tests that a reference solution has been run against, so you can send a take-home on day one without writing a prompt. When the brief is specific, DevParser generates a fresh task per candidate from the role, stack and seniority you specify, so there is nothing to leak: that exact problem has never been posted, sold, or indexed. And you do not have to give up the take-home you already use: paste it in — the README, the brief, whatever you send candidates today — and we pull out the description, requirements, test cases and starter code for you to check before anything is saved. We extract, we never invent: a field the brief did not contain comes back empty and says so.

We execute code in 35+ languages, including Python, TypeScript/JavaScript, Java, C, C++, C#, Go, Rust, Ruby, PHP, Swift, Kotlin, Scala, Elixir and Haskell. Framework code — React, Django, Rails, Spring — can be submitted and goes into the AI review, but hidden tests run against the language runtime rather than a framework test harness. New languages land regularly.

Create an account and your 14-day trial starts immediately — no card and no checkout. You only enter payment details when you choose a paid plan at the end of the trial. For those 14 days you get the full plan you chose, with the same limits and features a paying customer gets. After that, plans are month-to-month with no long-term contract.

Candidate data is encrypted in transit, integration credentials are encrypted at rest, and candidate data is processed with GDPR-aligned and EU AI Act-ready controls (transparency notice, explanations, human oversight tools). Submissions are used only to run your evaluations — they are not used to train foundation AI models, and you can delete a candidate and their data at any time.