You pay for the people we scored. Everyone else is free.

That's what the Executive view is for. It's a written brief — a headline, a health grade, what's behind it, who needs support, and three to five things you could do about it — deliberately written without engineering jargon. It says "their changes often skip safety checks", not "low error-handling score". If a number appears, it's measured from the work, never invented by the AI.

No. Never. Analysis runs on Amazon Bedrock inside AWS, whose terms prohibit using your inputs to train models. Diffs are processed to produce your scores, stored encrypted for your own audit trail, and used for nothing else.

No — deliberately, and we would not trust any tool that claims to. Detection is guessing, and a number your engineers can dispute is a number they will ignore. Almost everyone directs an agent now. We grade the artifact that shipped on the same 12 dimensions we always have, and we hand the developer the prompt that would have produced a better one. We do not rank people on "AI skill."

Two ways. First, severity comes from reachability and blast radius — could an unauthenticated stranger hit this, and what would they get — rather than from pattern matching, so the critical list is short enough to actually work through. Second, findings have an owner, a deadline, an honest "defer with a reason" option, and a re-check anyone can run to prove the fix landed. It reads your code in context, so it weighs a payments bug like a payments bug. Every paid plan includes protection for 25 repositories; additional blocks of 25 are $100 per month, and during your trial Sentinel protects up to 5 repositories.

The fastest way to make engineers hate measurement is to measure motion — commit counts, lines of code. We deliberately never rank on volume. Engineers get the same evidence leaders see, their own private growth view, and training built from their own code. The rubric is fixed and visible: the same bar for everyone, with receipts. Praise is public; comparison is private.

On GitHub the app reads repository contents, metadata and pull requests on only the repositories you enable, organization membership (to map authors to people) and branch protection settings (to confirm the check is required). It writes pull request comments, the “NRD Sight review” check and merge-queue status, nothing else. On Bitbucket Cloud it reads contents, pull requests and metadata and writes review comments, a build status and PR tasks. We never push code or change branches. Uninstalling the GitHub App, or revoking our access in Bitbucket, severs it immediately.

Connecting takes minutes. About five minutes after you enable a repository, First Look shows real pull request reviews, scores and security findings from your own code. From then on, new pull requests are reviewed as they open, usually within a couple of minutes. Up to 90 days of history fills in behind it, usually within hours; the largest histories finish overnight. Want to see it before connecting anything? Load the sample team on signup and walk the whole product with realistic data in about two minutes.

Only a developer whose code we analyzed in that billing month. Everyone who just views dashboards — executives, managers, stakeholders — is free, and your invoice lists exactly which developers were counted. Each plan’s base price already includes a set number of developers (5 on Starter, 15 on Team, 40 on Scale); the per-developer price applies only beyond that, so a team that fits inside its plan pays one flat number. Starter is for teams of up to 10 analyzed developers; from 11, Team is the plan.

All of them. The rubric measures engineering judgment — correctness, testing, error handling, design — which the AI evaluates in any language or framework, informed by each repository's own conventions.

No. AI review and scoring run on every pull request with no per-PR fee, and trivial changes are filtered before any AI runs. Every plan includes a monthly AI allowance, sized for every developer you pay for (or every analyzed developer, if there are more). If you use it up, pull request reviews continue as usual from a reserve (32% of the allowance), and the automated checks for secrets, risky CI, and infrastructure misconfiguration never stop. Other AI analysis waits until you add prepaid credits or your period renews. Overage is never charged unless you opt in, so an invoice can never surprise you.

Yes, if you want it to. The merge gate is optional and set by an owner or manager per organization, team, repository, or developer. Advisory mode holds nothing and reports what it would have blocked. Acknowledge mode makes a pull request wait until its blocking findings are resolved or acknowledged; strict mode waits until each is fixed or someone records a reason. It works through a required GitHub check or Bitbucket Cloud build status and tasks (Bitbucket only blocks on its Premium plan; otherwise it warns). Developers answer from the PR with commands like /nrd false-positive, every override is recorded, and a category that is dismissed as a false positive too often stops blocking automatically.

Every plan includes our standard Data Processing Agreement — read it at /dpa and accept it on the Billing page. A workspace owner accepts it in the app in about a minute, during a trial too, and can save a signed copy as a PDF. It incorporates the EU Standard Contractual Clauses and the UK Addendum.

Signing up gives you a sample workspace to explore, no card needed. To connect your own code, add a card: $0 today, 14 days free on the plan you choose, cancel anytime. During the trial you get that plan’s features and its included developers on up to 5 repositories, with real Security Protection on those 5 — so a Team or Scale trial includes architecture reviews, and a Starter trial does not. (Checking code against your own policy documents needs an active Scale subscription.) It converts to the plan you chose, and you can switch plans any time from Billing.