cubic

Visit Website

Introduction

cubic is an AI code review platform that launched on Hacker News in April 2025 under its earlier name, mrge, as part of Y Combinator’s X25 batch. It describes itself as the #1 AI code reviewer on independent benchmarks and publishes a report to back that claim.

Every pull request is checked against the whole codebase and your team’s rules, with inline feedback in seconds. It flags changes that break code elsewhere, even in another repository, and turns reviewers’ past comments into rules that run on every PR, such as using design tokens, avoiding process.env, requiring tests for payment flows or making migrations use lock_timeout.

cubic can auto-approve safe changes while routing risky ones to a human, and its comments are designed for coding agents: Claude Code, Codex and Cursor can pick them up and fix the issues. A terminal mode applies the same rules before you push, so problems are caught before a PR exists. A handy shortcut is to change github.com to cubic.dev in any PR link to open it in cubic.

Beyond PR review, cubic runs thousands of agents each night across your codebase to find bugs and security issues. AI triage notifies owners and creates tickets, and background agents can fix an issue in one click and resolve the ticket when the fix merges. For privacy, the site says AI reviews code in real time and then wipes everything clean.

Testimonials on the site include the Better Auth project. Pricing is per developer and measured in reviewed lines, counting only added and deleted lines; generated files, vendored code, binaries and ignored files do not count, and incremental reviews count only new changes. cubic is free for open-source teams.

Measure cubic on its auto-approval behaviour as much as its comments: decide which file paths or change types may be auto-approved, start conservatively, and expand as trust grows. A sensible rollout for any AI reviewer is to enable it on two or three active repositories, ask the developers who receive its comments to rate them for a fortnight, and only then widen it. Track how many comments lead to a code change: that ratio says more about value than any benchmark chart.

Because cubic charges by reviewed lines, large generated diffs are excluded and incremental reviews count only new changes, so estimate usage from your typical PR size rather than total repository size.

Nightly background scans are a different job from PR review: they look at code that already shipped. Route their findings to owners with a clear severity, or they risk becoming another ignored queue.

Its terminal mode is a good first step for sceptical developers, because they can try the rules privately before any comments appear on their pull requests.

Launch and background: Y Combinator’s company directory lists cubic in its Spring 2025 batch, with the one-liner “AI-powered code review platform”. cubic was launched on Hacker News on 15 April 2025 as “mrge.io (YC X25) – Cursor for code review” and has a Product Hunt listing.

Features

1. Whole-codebase review: Context beyond the diff.
2. Cross-repo breakage: Flags downstream impact.
3. Learned team rules: From past review comments.
4. Auto-approval: Safe changes merged faster.
5. Agent-ready comments: Claude Code, Codex and Cursor can fix them.
6. Terminal reviews: Same rules before push.
7. Nightly bug hunting: Thousands of background agents.
8. AI triage and tickets: Owners notified automatically.
9. One-click fixes: Background agents resolve tickets.

Pros

1. Free for open source: Open-source teams pay nothing.
2. Usage that excludes noise: Generated and vendored files are not counted.

Cons

1. Line-based allowances: Large teams may hit limits.
2. Auto-approval needs policy: Requires careful configuration.

Usage and Links (Example)

1. Install the GitHub app and select repositories.
2. Import or write team rules; let cubic learn from past comments.
3. Try terminal reviews before pushing.
4. Swap github.com for cubic.dev in a PR URL to open it in cubic.

Resources:

• cubic website

Pricing

cubic charges per developer, billed yearly from $30 per developer per month, with plans sized by reviewed lines per developer (40,000, 80,000 or 160,000 a month on the largest Max plan, listed at $160). It is free for open-source teams.

Recommended for?

1. High-velocity teams using coding agents.
2. Open-source maintainers with many incoming PRs.

Conclusion

cubic combines a fast reviewer with background bug-hunting agents.

Start with reviews, then add auto-approval and nightly scans once you trust it.

On CheckOps, cubic is filed under the Release & CI gates and Unit testing stages. Related tools listed here include CodeRabbit, Greptile, Cursor Bugbot, GitHub Copilot code review, PR-Agent and Sourcery. Which one fits best depends mostly on your stack, your team’s size and skills, and whether you prefer to run tools yourself or pay for a managed service. This page reflects the official website, documentation and repository as they stood on 11 November 2026.