A repository check is the full 22, a skill or plugin check its own 16, and repositories already known to be malicious stay blocked even past your limit. Paid plans lift the cap and add the rest — more machines, the stats board, the live ticker, the MCP server, Claude Code plugin and Claude skill for AI agents, and deep scans of the files themselves.
Every developer, to start
1 device · 10 repository, skill or plugin checks in your first 14 days
A developer with more than one machine
3 devices · billed yearly
/repogates:checkup and the CLI score what your AI agents already have installedCheckout opens shortly. Nothing to pay yet.
Personal, plus the files themselves read in a sandbox
3 devices · 20 deep scans a year
/repogates:checkup and the CLI score what your AI agents already have installedDeep scans are live. This tier opens as soon as checkout does.
3–5 developers · coming
3–5 seats on one invoice
Buying for a team today? Personal, one per developer — same protection, no seat to wait for.
A deep scan reads a repository's, an extension package's or a skill's files in an isolated sandbox — OpenGrep with our rules, Gitleaks, Trivy and ClamAV — and its verdict becomes the verdict. If you connect your own AI key, a scan too large to read in full also gets an AI review of what the engines could not read: it lists the unread files worth opening and can raise a concern, and it never clears a scan. A clean scan is not proof of safety; it is every engine finding nothing it recognises, and the result says exactly which engines ran.
Check what you built. Scan a public repository you maintain and its report also says what to fix first: the findings ranked, each with where it is, why it matters, what to change and how to confirm it. A button copies a fix as text for your own coding agent, and a second scan shows what cleared, what is still open and what could not be re-checked. RepoGates never changes your code. A scan reads the code and never runs it, a list with nothing left on it is not an audit, and checking again is another scan at the same price. Private repositories are not scanned. How it works.
From 1 October 2026 a scan costs credits by size, and you see the price before you start it — nothing above the price you accept is ever charged:
Most repositories are one credit. Where only an upper bound of the size is known before the scan, the estimate is charged and lowered once the scan has counted — never raised. Credits bought before 1 October 2026 each pay for one scan of any size. Credits you buy never expire. The Premium tier includes 20 a year, which last for that subscription year, and credits that come with a voucher end with it. Fair use: 100 scans a month per account, 20 of them at 2 or 3 credits; a scan that fails gives its credits back but still counts as an attempt. On a paid plan your AI agents can spend them too — through the MCP server or the Claude Code plugin, one scan per call and only after you say yes to that call, at most five agent-started scans a day per account. A repeat of the same repository within 15 minutes returns the same result and costs nothing.
Deep scans are live — the packs go on sale as soon as checkout opens.
Bought at an earlier price? Your subscription keeps renewing at the price you bought it at, for as long as it runs. Moving it to a different plan is charged at today's price.
Choosing a different plan moves the subscription you already have — it never starts a second one. Moving up takes effect straight away and is charged pro rata: you pay only the difference on the part of the period you have not used, and your renewal date stays where it was. Personal to Premium the day after paying annually is about $23, not $92.
Moving down takes effect at your next renewal: nothing is charged or credited when you choose it, and you keep your current plan until then — the full wording.
Free covers 10 distinct repositories, skills or plugins within 14 days of registering — a repository gets the full 22 checks, an AI-agent skill or a Claude Code plugin its own 16. Two limits, whichever comes first: the ten repositories, or the fourteen days. It does not reset month to month. Checking the same repository twice never counts twice.
Past the limit, RepoGates does not go quiet and it does not pretend. Repositories already known to be malicious are still blocked, because that costs us nothing and letting known malware through to someone who has not paid would be indefensible. For anything else you are told plainly that it was not checked, and you choose: upgrade, or continue at your own risk. That choice is recorded either way.
If a subscription lapses, RepoGates drops back to the free tier and keeps gating downloads. Your trial does not restart — new repositories are no longer assessed — but it never switches protection off for non-payment. Nobody should be less safe because their card expired.
The security column is identical on every row that matters. That is the point.
| Free | Personal | Premium | Team | |
|---|---|---|---|---|
| In your browser | ||||
| Trust board and download gate on GitHub | ✓ | ✓ | ✓ | ✓ |
| The same on Hugging Face and GitLab | ✓ | ✓ | ✓ | ✓ |
| Assessments | 10 in your first 14 days | unlimited | unlimited | unlimited |
| Every published check behind a verdict — 22 on a repository, 18 on a model, 22 on a GitLab project | ✓ | ✓ | ✓ | ✓ |
| Four verdicts for what you are about to do: open it, clone it, install it, hand it to an AI agent | where already assessed | ✓ | ✓ | ✓ |
| Stats board, owner and maintainer scores, and what changed since your last look | where already assessed | ✓ | ✓ | ✓ |
| Known malware blocked for good (campaign blocklist) | ✓ | ✓ | ✓ | ✓ |
| Knows when an AI assistant sent you (AgentBaiting) | ✓ | ✓ | ✓ | ✓ |
| Live security ticker | ✓ | ✓ | ✓ | ✓ |
| Allow once or always — your allowlist follows your account | ✓ | ✓ | ✓ | ✓ |
| Audit log of every block, warning and override, kept on your device | ✓ | ✓ | ✓ | ✓ |
| For your AI agents | ||||
| MCP server — twelve tools, for any MCP client | — | ✓ | ✓ | ✓ |
| Claude Code plugin — refuses a clone or install that names a blocked repository, before it runs | — | ✓ | ✓ | ✓ |
| Claude skill for Claude.ai and Cowork | — | ✓ | ✓ | ✓ |
| Verdicts on Hugging Face models, GitLab projects, VS Code and Docker Desktop extensions, agent skills and Claude Code plugins | — | ✓ | ✓ | ✓ |
| Agent check-up — /repogates:checkup and the CLI | — | ✓ | ✓ | ✓ |
| Personal API tokens | — | up to 5 | up to 5 | up to 5 |
| Reading the code itself | ||||
| Deep-scan credits included | — | — | 20 a year | — |
| Buy scan packs | ✓ | ✓ | ✓ | ✓ |
| Deep scan of a repository, a GitLab project, a VS Code extension package or an agent skill | with a pack | with a pack | ✓ | with a pack |
| Findings with a fix for each, report and SARIF export | with a pack | with a pack | ✓ | with a pack |
| AI review of the files a large scan could not read, on your own AI key | with a pack | with a pack | ✓ | with a pack |
| Agent-started scans, with your yes | — | 5 a day | 5 a day | 5 a day |
| Machines and people | ||||
| Devices per person | 1 | 3 | 3 | 3 |
| Billed together, one invoice | — | — | — | ✓ |
| Seats | 1 | 1 | 1 | 3–5, coming |
Chrome and Edge ship a new major version every four weeks. Each one can break a blocking download listener, and each one has to be tested against. That work is constant, and it is what keeps the free tier working for everyone. Personal pays for it.
New repositories stop being assessed and RepoGates says so — it does not go quiet and let things through as though they were fine. Known malware is still blocked. For anything else you choose: upgrade, or continue unchecked. Either choice is recorded.
Yes — one skill or one plugin is one assessment, exactly as one repository is. A skill hosted on GitHub or a Claude Code plugin is scored by its own 16 checks, and the same ten-in-fourteen-days count covers it; checking it again never counts twice. The agent check-up — an inventory and score of what an agent installation already has — is on the paid plans and not on the trial; it reads your AI-agent configuration on your machine and sends RepoGates identifiers, never contents (how it works).
You go back to the free tier — the gate keeps working and known malware stays blocked; new repositories are no longer assessed. You lose the board, the ticker, the MCP server and plugin, and the extra devices. You do not lose the gate. Switching protection off over a card expiry would be a strange thing for a security product to do.
Each browser profile that installs RepoGates registers a random id — not a hardware fingerprint. You can see your devices and remove one at any time to free a slot.
No, and no browser extension can — it cannot see other processes. RepoGates gates browser downloads, which is how the FakeGit campaign actually delivered its payloads. It does not see git clone, package managers or curl — outside Claude Code with the RepoGates plugin, whose hook refuses a clone or install that names a blocked repository on the command line, before it runs. A registry package named on that line — npm install express, pip install flask — is resolved through npm or PyPI to the source repository it declares; direct dependencies only. The hook sees Bash tool calls in that one client; the MCP server covers AI agents that choose to ask, and the same skill in Claude.ai and Cowork asks before a clone with no hook to refuse. The full coverage table is here.
Sign in, open your account page and enter the code under Subscription. It gives the account Personal or Premium — the code says which — for the number of months it carries. No card is asked for, it does not renew, and it ends by itself; the account then goes back to the free tier. A Premium voucher includes a few deep-scan credits, which last until the voucher ends. The voucher terms are here.