Greptile Reads Your Entire Codebase So Your Engineers Don't Have To

The Macro: Code Review Is Still a Bottleneck and AI Is Coming for It

Code review is one of those engineering processes that everyone agrees is important and nobody enjoys. A senior engineer opens a pull request, tags two reviewers, and waits. Sometimes they wait an hour. Sometimes they wait a day. At some companies, the median time to first review is measured in days, and the median time to merge stretches past a week. This is not a productivity problem. It is a momentum problem. Engineers context-switch while they wait, pick up new work, and then have to re-load the original context when review comments finally arrive.

The existing tooling has not solved this. GitHub’s built-in review system is functional but dumb. It shows you the diff and nothing else. Reviewbot and similar lightweight tools catch formatting issues but miss logic bugs. Codacy and SonarQube do static analysis, which is useful but orthogonal to the question “does this change make sense in the context of our codebase?” That last question is the one that requires a human reviewer, because only a human understands the full codebase.

Or at least, only a human used to understand the full codebase. LLMs changed the equation. If you can index an entire repository and give a model access to the full context, you can build a review system that understands not just the diff but the architecture, the conventions, the related code, and the test coverage. That is the bet that multiple companies are making right now.

Sourcegraph pioneered codebase-scale search and understanding. Cody is their AI coding assistant. CodeRabbit does automated PR reviews. Qodo (formerly CodiumAI) focuses on test generation. The space is competitive, but nobody has locked it down yet. The reason is that codebase understanding is genuinely hard to do well at scale, and the quality bar for code review is high enough that a mediocre AI reviewer is worse than no reviewer at all. Bad review comments waste more time than they save.

The Micro: Georgia Tech Grads Solving Codebase Understanding

Greptile indexes entire codebases and then does two things with that understanding: automated code review on pull requests and natural language Q&A about the codebase. It works with both GitHub and GitLab, drops inline comments on PRs, generates summaries with mermaid diagrams, and supports custom rules that teams can define in plain English.

The numbers are solid. Over 1,000 software teams use the product. The headline stat is that median merge time dropped from 20 hours to 1.8 hours for teams using Greptile. That is an 11x improvement, which is large enough that even if the measurement is generous, the underlying effect is clearly real. The customer list includes Brex, Substack, Scale AI, Klaviyo, PostHog, and Mintlify. These are not tire-kickers. These are engineering organizations with high standards and real code review problems.

The founding team is three Georgia Tech CS grads. Daksh Gupta leads the company. Soohoon Choi is the CTO. Vaishant Kameswaran rounds out the founding team. The team has grown to 20 people, which is lean for the traction they have. They came through Y Combinator’s Winter 2024 batch and are based in San Francisco.

Pricing is $30 per developer per month for code reviews, with fixed monthly subscriptions for chat features and per-request API pricing available. There is a free tier for open-source projects, which is smart for adoption. The 14-day free trial lowers the barrier for teams that want to test it on their own repos before committing.

One feature that stands out is the learning system. Greptile observes how engineers respond to its review comments and infers team coding standards over time. If your senior engineers consistently ignore a certain type of suggestion, the system learns to stop making it. If they consistently agree with another type, the system learns to prioritize it. This is the kind of feedback loop that makes an AI tool get better with use instead of staying static.

The product supports over 30 programming languages, including Python, JavaScript, TypeScript, Go, Java, C++, and Rust. Language breadth matters because most engineering organizations are polyglot. A code review tool that only works for your TypeScript services but not your Go infrastructure is not useful enough to buy.

The Verdict

Greptile is one of the more convincing AI developer tools I have looked at. The traction is real. The customer names are strong. The product solves a problem that every engineering team over 5 people actually has. And the pricing is reasonable enough that an engineering manager can expense it without a procurement process.

The competitive risk is real but manageable. Sourcegraph has broader ambitions and a larger team, but they are also a much bigger company with a more complex product. CodeRabbit is a direct competitor, but Greptile’s learning system and customer base give it a defensibility advantage. The biggest long-term threat is probably the code hosting platforms themselves. If GitHub ships a native AI review feature that is “good enough,” it becomes harder to justify a third-party tool. But GitHub’s track record with Copilot-adjacent features suggests they will move slowly and ship something generic.

Thirty days from now, I want to see the retention curve. Are teams that try Greptile still using it after 90 days? The merge time stat is impressive, but retention is the real test. Sixty days, I want to know the expansion motion. Does Greptile grow seat-by-seat within engineering orgs, or does it plateau after initial adoption? Ninety days, the question is whether the API product line is gaining traction, because that is the path from a code review tool to a platform. The foundation is strong. The market is large. The team is executing. This is the kind of company that either gets very big or gets acquired by someone who wants to be.