cubic – AI Code Reviews That Actually Get Your Codebase

TL;DR

TL;DR: cubic is an AI code reviewer that integrates with GitHub and your IDE, learns your team’s standards from senior engineer PR comments, and continuously scans your entire codebase overnight to catch bugs you’d otherwise miss.

Source and Accuracy Notes

• Product: cubic.dev (YC X25, 221 points on HN)
• HN Launch: “Launch HN: mrge.io (YC X25) – Cursor for code review”
• Note: The product was initially named mrge.io and rebranded to cubic.dev. Both domains resolve to the same product.

What Is cubic?

cubic positions itself as “the world’s most accurate AI code reviewer.” The core pitch: most AI code review tools give generic feedback or surface trivial issues. cubic is built to find hard-to-find bugs—the kind that live in complex cross-module interactions, subtle race conditions, and context-dependent logic that senior engineers catch but junior reviewers miss.

The product sits at the intersection of a few trends:

• PR-level inline feedback — not just a summary, but active review annotations in your existing workflow
• Learning from your team — rather than applying generic rules, cubic reads what your senior engineers actually said in PR comments and applies those lessons
• Continuous codebase scanning — overnight, thousands of AI agents scan your entire codebase (not just diffs) looking for patterns that indicate bugs

The 2-click GitHub install is notably simple. Most enterprise developer tools require significant configuration or onboarding. cubic skips that friction entirely.

How cubic Works

GitHub and IDE Integrations

cubic reviews code wherever your team works. The GitHub integration posts inline comments directly on PR diffs, matching the format your reviewers already use. No new UI to learn. No separate dashboard to cross-reference.

The IDE plugin (VS Code and JetBrains) brings review feedback into your editor. You see issues in context as you code, not after a round-trip to the PR page. This closes the gap between “write code” and “learn that there’s a problem” — which is where developers actually fix issues fastest.

Learning Team Standards in Plain English

Rather than configuring a rigid ruleset, you write standards in plain English. “We don’t use raw SQL strings because of injection risk.” “All public methods must have XML doc comments.” “Avoid async void except for event handlers.”

cubic encodes these and enforces them across PRs automatically. The rules evolve as your team refines its thinking — no YAML refactoring required.

Senior engineers’ historical comments are another signal. If your team’s senior developers consistently flag certain patterns in certain file types, cubic picks up on that and applies it proactively to new PRs.

Overnight Codebase Scanning

This is where cubic diverges most from conventional PR review bots. Once a night, it runs thousands of AI agents across your full codebase — not just the changes in a given PR — looking for structural bug patterns, security vulnerabilities, and places where the code has drifted from its own standards.

Issues are triaged automatically: cubic determines which developer owns the issue, creates a ticket, and notifies them. One-click fix commits handle straightforward cases. More complex issues get routed to the right person with full context included.

This is analogous to having a senior engineer read through your entire codebase every night and flag anything concerning — except it’s running continuously and at a scale no human team could sustain.

Practical Evaluation Checklist

• Does it integrate with existing GitHub PR workflow? — Yes. Native GitHub App, posts inline comments on diffs.
• Can you customize the rules without writing config files? — Yes. Plain English rule authoring.
• Does it learn from your team’s own code review history? — Yes. Reads senior engineer comments and applies patterns.
• Does it surface issues beyond individual PR diffs? — Yes. Overnight full-codebase scanning with thousands of agents.
• Is the initial setup actually simple? — GitHub App install is 2 clicks. No credit card required for free trial.
• Does it work in the IDE? — VS Code and JetBrains plugins available.

Security Notes

• cubic requires read access to your GitHub repositories to post comments and read code.
• The codebase scanning feature means code is processed by cubic’s infrastructure — verify your data processing agreement if working with sensitive or regulated code.
• Review the cubic security documentation before deploying on enterprise or security-sensitive projects.

FAQ

Q: How does cubic compare to GitHub Copilot’s code review?

A: GitHub Copilot code review gives general suggestions on a diff. cubic is designed specifically for review workflows: it customizes feedback to your team’s standards, learns from your senior engineers’ actual comments, and runs full-codebase scans that Copilot doesn’t perform.

Q: Does cubic replace human code reviewers?

A: No. cubic is positioned as augmenting human reviewers, not replacing them. Teams use it to catch the issues that slip past junior reviewers and to reduce the load of mechanical feedback so humans can focus on design and architecture review.

Q: What languages does cubic support?

A: cubic works across common languages including Python, TypeScript, JavaScript, Go, Rust, and Java. Check the documentation for the full supported list.

Q: How does the overnight codebase scan work at scale?

A: cubic runs thousands of AI agents nightly against your codebase. For large monorepos, this scanning is distributed across the codebase segments. Issues found are automatically triaged and routed.

Conclusion

cubic’s strongest differentiator isn’t any single feature — it’s the combination of instant PR feedback, plain-English rules that require no configuration syntax, team-learned standards from senior engineer comments, and continuous overnight full-codebase scanning. The 2-click GitHub install suggests the product is serious about low friction onboarding.

If your team struggles with code review quality inconsistencies across reviewers, or if you want to catch serious bugs before they ship rather than after, cubic is worth evaluating. The free trial removes the commitment barrier — worth a look if you’re already paying for code review tooling that gives generic feedback.