The problem
AI made writing code fast. Reviewing it didn't keep up.
Review is the bottleneck
Agents open pull requests faster than your seniors can read them. Either reviews get thin, or work queues up behind them.
The author shouldn't approve its own work
When one model writes the code and the same model reviews it, the same blind spots get waved through.
Your auditor will ask
"Show me who reviewed this AI-written change, and how it was tested." An approving click on GitHub isn't a good answer.
How Arrgate works
A review crew on every pull request. You only see what needs you.
Install it in your repository once. From then on, every pull request (from an agent or a person) goes through the same checks, inside your own GitHub Actions.
01Independent reviewer
Runs on a different model from whatever wrote the code, and sees only the diff, the ticket and the tests.
02Tester
Writes tests that protect what already works, and shows them failing where behaviour changed, so you know they catch something.
03Security red team
Tries to break the change and reports proven risks, not a list of maybes.
04Reuse check
Spots code that duplicates something you already have and points to it, before your codebase fills with near-copies.
05Manager
Applies your team's review rules. Findings go back to the agent to fix; a person gets one short packet only when a rule says so.
06Cost per change
Every pull request shows what its agent runs cost, and Arrgate stays inside your plan's limits unless you choose otherwise.
Evidence your auditor can use
Every change leaves a signed record.
Change management is where AI-written code gets awkward in an audit. Arrgate writes one record per pull request and exports them for your SOC 2 (CC8.1) or ISO 27001 (A.8.32) evidence.
- Who asked for it: the ticket and the request.
- What checked it: each reviewer, its model, and what it found.
- How it was tested: the tests added, shown failing before the change.
- Who approved it, and what the agent runs cost.
- Signed, so it can't be quietly edited later. Export to CSV in one command.
Example record. Field names may change before release.
Built to fit how you already work
No new platform to trust with your code.
Runs in your GitHub
Everything runs in your own GitHub Actions with your own model keys. Your code never passes through us.
Works with your agents
Claude Code, Cursor or anything else that opens pull requests. Arrgate reviews the result, not the tool.
Your rules decide
You set what needs a person and what doesn't. Change the rules any time; they're a file in your repo.
Founding teams
Help shape it, and pay half.
We're starting with a small group of teams in Australia and New Zealand, from January 2027.
- 50% off your first year.
- Set up with you, on a call, in your repo.
- A direct line to the founder, and a real say in what gets built next.
- Your time zone. Calls during your working day.
Questions
Before you ask
Can I use it today?
Not yet. Founding teams start in January 2027 with a guided setup. Sign up and we'll tell you as soon as your spot is ready.
Does my code leave my account?
No. Arrgate runs in your GitHub Actions with your model keys. We only ever see usage counts, never code, diffs or prompts.
Which AI providers does it use?
The ones you already pay for. Reviews run on a different model from the one that wrote the code, so blind spots don't repeat.
We use CodeRabbit or similar. Why both?
Review comments are one part. Arrgate adds tests that prove they catch something, a red team, your team's rules about what needs a person, and audit evidence.
What will it cost?
Priced per team, not per seat, in US dollars. Founding teams get 50% off the first year.
Who's behind it?
An engineer in Western Australia building the tool they wanted for their own AI-written code. Email hello@arrgate.com any time.