FAQ
Questions about AI coding agent guardrails, answered
Straight answers about Cotterpin's guardrails for AI coding agents: what leaves your machine, who can approve an agent's actions, what is free and how payment works, how fast it is, what it needs, and how to remove it.
Questions and answers
Does Cotterpin replace my coding agent?
No. You keep using Antigravity,Claude, Codex or Grok under your own account. Cotterpin runs them, checks their work, and decides what is allowed to land.
Does Cotterpin replace CI or code review?
No. It shifts the checking left, onto your machine and into the agent session, before the commit. CI and review still run; what reaches them has already been checked, so they find less and you spend less time on them.
How is Cotterpin different from free tools like Backstage?
Backstage catalogs your software and its docs; pre-commit and linters run the checks you configure; your agent tool's hooks run the commands you write. Cotterpin checks that your agent's “done” is true, with a real build and the tests the change reaches, using rules it ships and maintains, and it works beside those tools: it can feed your own Backstage. See the comparison.
Does my code leave my machine?
Only in the model call you already make to your own provider. Cotterpin itself has no cloud that sees your source, diffs or prompts, and session reports are computed locally.
Does Cotterpin collect telemetry?
Telemetry is off. It can be turned on only by you, explicitly, through cotterpin consent. What a subscription sends to our licence service is listed in the Privacy Policy. Bug reports are not telemetry: you send each one after reading it, or, from the next release, you can turn on automatic sending yourself for one repository and kind of report, and turn it off again with cotterpin bug-autosend off.
Can I use it on my work laptop and my personal machine?
Yes. Pro covers three machines per person — work, home and a spare — and you can move a licence to a new machine yourself, any time, with cotterpin machines. Watch works on as many machines as you like.
Is Cotterpin free?
Yes. Anyone can download and install it at no cost: no card, no sign-up and no time limit. The free version is Watch, plus the free routes: the language packs' upgrade routes (Java, Jackson 3, Spring Boot 4.1, .NET, System.Text.Json, Angular, RxJS, Python, Pydantic v2 and the Terraform AWS provider) and the Snyk vulnerability fixes (cotterpin snyk). The free routes run in full, and their checks enforce as on Pro. You only enter payment details if you choose to subscribe.
How do I pay, and who sees my card?
Payments are processed by Paddle, our merchant of record, in Paddle's own checkout — Paddle also handles sales tax and VAT and issues your invoices. Cotterpin never sees your card details. Pro is $100 a year; Team is $30 per seat per month, billed annually, with a 5-seat minimum. The simplest way to subscribe is cotterpin subscribe, which binds the licence to your machine. If you bought on the website, claim it with cotterpin subscribe --claim <email>. Cancel or update your payment method any time from the Paddle customer portal.
What does Watch do, and what do Pro and Team add?
Watch runs beside your agent on every machine and repository and never blocks. After each agent turn, and in each commit and push report, it tells you what Cotterpin would have caught — a false “done”, a failing gate, a commit or push it would have refused — and why. That report stays on your machine and is never sent. Pro and Team turn it into enforcement: the gates refuse, and the work goes back to your agent and gets fixed. Subscribe whenever you want that; nothing about Watch expires.
Why doesn't the free plan enforce anything?
Because enforcement is the product. Watch shows you exactly what Cotterpin is worth on your own code, session by session — then you decide whether that's worth $100 a year (about $8.33 a month). No crippled features, no nagging: just the evidence.
What's the difference between Watch and Pro gates?
Every gate has two halves. The detector finds the problem and tells you where it is — Watch gets that. The enforcer blocks the commit, sends the work back to your agent with the evidence and a fix, and has the refuter attack it — that's Pro. New language packs reach Pro the day they ship, and Watch 60 days later.
What if a gate is wrong?
Only gates with a zero-false-positive record block from install; the rest advise until they earn it. Every finding names the rule, the file and the fix. You can switch a rule off, but never anonymously: it takes a person's signed approval and a reason, recorded with the change.
Isn't this just linting and tests?
Those are the floor, and Cotterpin runs them. It's built for what they miss: work that passes every check and still isn't wired up, tests that only agree with themselves, and fixes that break under a retry or a race.
Will it slow my agents down?
Each task gets a real build and the tests your change can reach — not the whole suite. Unchanged inputs reuse their proof. The hook every agent tool call passes through takes about 40–60 ms of CPU time per call (the measured median for each kind of call) and a commit's fast tier runs in seconds. Both are held to measured speed budgets in our own test suite: at most 10% slower than the baseline, or the release doesn't ship. The trade is minutes per task against hours of debugging later.
What do I need to run it?
macOS; or Linux, including Windows through WSL, in developer mode (hardened mode is macOS only for now); Python 3.11 or newer; git 2.30 or newer; and one of Antigravity,Claude, Codex or Grok. cotterpin doctor checks your machine and prints the exact fix for anything missing. See the quickstart.
Which languages does it support?
Five, each through a language pack installed beside Cotterpin: Java, .NET, Python, Angular/TypeScript and Terraform. Another 51 language-neutral gates check any repository.
How do I remove it?
cotterpin uninstall removes it from one repository and lists everything first; cotterpin --uninstall removes it from your whole machine. Every change it made to your code was a commit, so each one can be reverted.
Can an agent approve its own actions?
No. An approval is a signature by a key only root can read, made by a root-owned signer after the operating system has checked it's you — your account password, or Touch ID (your fingerprint) on a Mac where Touch ID is enabled for sudo — afresh every time. (To turn Touch ID on for sudo on macOS 14 or later: copy /etc/pam.d/sudo_local.template to /etc/pam.d/sudo_local and remove the # before its auth sufficient pam_tid.so line.) The signer shows the exact action it is about to sign. A record an agent writes itself doesn't verify, and is reported. On a developer's own machine this is a guardrail for agents; for a boundary against any process running as you, use hardened mode — see Security.
What licence is Cotterpin under?
The source is available under the PolyForm Internal Use License 1.0.0: you may use and modify it for your own organisation's internal operations, but not distribute it. Enforcement on Pro and Team comes with a subscription under our Terms. Licences are signed by Cotterpin's licence key and checked on your machine against the key pinned in the build. See the Software Licence page for a summary and the full text (pending solicitor review).
Still have a question?
Ask us directly, or see it on your own code — free, with no time limit, no credit card.