# Greptile: reviews and analysis > Pull request reviewer that indexes the whole repository before commenting on a change. - Canonical: https://serchai.com/en/reviews/greptile/ - Site: Serchai (https://serchai.com) — AI tools comparator - Language: en - Updated: 2026-08-05 --- ## Verdict Greptile indexes the repository before it opines, and that is where its edge comes from: it flags breakages that live not in the diff but in the code around it. It pays for that reach in noise, because it comments more than it lands, and since March 2026 in money, because the $30 per seat carries 50 reviews and the next one costs a dollar. With agents opening pull requests by the dozen, that bill runs away. **Best for:** Teams with large, tangled repositories and a moderate pull request rhythm, who would rather read too much than miss a cross-service breakage. **Rating:** 3.4/5 ## Pros - Indexes the whole repository, so it sees dependencies the diff does not show - Custom review rules written in plain English - Free plan with 50 reviews a month, and free for qualifying open source projects ## Cons - Charges per review beyond the 50 included, which no competitor does - High noise: in the most cited third-party comparison it leaves eleven false positives against CodeRabbit’s two - GitHub and GitLab only, no Bitbucket and no Azure DevOps ## Key facts - Price: Permanent free tier + paid from 30 USD - Free trial: Yes - Platforms: Web - Categories: [Code](https://serchai.com/en/best-ai/code/) - Official website: https://www.greptile.com ## What the internet says (agentic sweep) Greptile is the pull request reviewer that looks deepest, because it indexes the whole repository before commenting, and the one that gets on worst with volume: since March 2026 it charges $30 per seat with 50 reviews included and a dollar for every review beyond that, a model no competitor uses and one that drew public cancellations. The most cited third-party comparison gives it twice CodeRabbit's bug catch rate and also eleven false positives against two. - Sweep date: 2026-08-05 - Derived score: 3.4/5 ### Axes - Results: 3.9/5 - Control: 3.8/5 - Real price: 2.6/5 - Integration: 3.4/5 - Support: 3.2/5 ### Recurring themes in favor - Indexing the whole repository is what sets it apart: it comments with the surrounding code in front of it, not just the diff, and that is where it catches cross-service breakages (strong theme) - Custom review rules written in plain English, with no separate configuration syntax to learn (present theme) - Free plan with 50 reviews a month for one active developer, a 14-day trial and free use for qualifying non-commercial MIT or Apache licensed projects (present theme) - The vendor itself publishes that with version 4 the share of comments addressed by the author went from 30% to 43% after a month of parallel testing (present theme) ### Recurring themes against - The price punishes volume: $30 per seat per month brings 50 reviews, and beyond that each extra credit costs a dollar. With agents opening pull requests, the bill stops resembling the headline (strong theme) - The pricing change left a public protest, though a small one: a pseudonymous page from 30 April 2026 collecting cancellations and open source maintainers billed despite the exemption, plus a Hacker News submission that same day which stalled at 17 points and three comments (strong theme) - Noise: in the most cited third-party comparison over 50 real pull requests it leaves eleven false positives against CodeRabbit's two, and its 82% catch rate comes from an in-house benchmark of 50 pull requests across five repositories, not from an independent measurement (strong theme) - It only plugs into GitHub and GitLab: anyone with code on Bitbucket or Azure DevOps is left out (present theme) ### Sweep sources - [Official pricing] https://www.greptile.com/pricing — Plan Pro a «$30/seat/month» con «50 credits included per seat» y «$1 per additional credit». Plan Starter gratuito con 50 créditos al mes y un desarrollador activo, prueba de 14 días y 50 % de descuento para empresas anteriores a la serie A - [Press] https://www.greptile.com/blog/greptile-v4 — Anuncio del 5 de marzo de 2026: «The pricing is $30/developer/month, which includes 50 reviews/month, after which reviews cost $1 each», y la subida de comentarios atendidos del 30 % al 43 % - [Review sites] https://dev.to/jovan_chan_9500711396d4e6/greptile-review-2026-82-bug-catch-rate-the-1review-trap-and-who-should-pay-30month-4jao — Comparativa del 2 de junio de 2026 sobre 50 pull requests reales de Sentry, Cal.com y Grafana: 82 % de fallos cazados frente al 44 % de CodeRabbit, y once falsos positivos frente a dos - [Communities] https://greptile-fail.vercel.app/ — Página de protesta del 30 de abril de 2026 con el desglose del coste por volumen, cancelaciones citadas con nombre y el caso de un mantenedor de código abierto facturado pese a la exención - [Press] https://techcrunch.com/2025/07/18/benchmark-in-talks-to-lead-series-a-for-greptile-valuing-ai-code-reviewer-at-180m-sources-say — TechCrunch, 18 de julio de 2025: ronda de serie A en negociación por 30 millones de dólares con valoración de 180 millones, liderada por Benchmark - [Communities] https://news.ycombinator.com/item?id=39604961 — Presentación en Hacker News del 5 de marzo de 2024, con 253 puntos y 166 comentarios de desarrolladores evaluando la indexación del repositorio - [Review sites] https://deepsource.com/resources/ai-code-review-tools — Comparativa de un competidor directo, DeepSource: sitúa el 82 % como banco de pruebas interno de Greptile sin validación independiente y le atribuye alucinaciones al no apoyarse en análisis estático > **TLDR:** Greptile reviews every pull request with an index of the entire repository behind it, so it sees dependencies and breakages that never show up in the diff. In exchange it talks too much, and since March 2026 it charges by volume: from $30 per seat per month with 50 reviews included, then a dollar for every extra review. It suits teams with tangled codebases and a contained change rhythm, not workflows where agents open pull requests by the dozen. ## What Greptile is and how it works Most automated code reviewers read the same material a rushed human does: the diff. They see the changed lines and a bit of the file around them, and they opine from there. Greptile starts somewhere else. Before it comments on anything it builds an index of the whole repository, with its functions, its classes and the dependencies between them, and it uses that map as the context for the review. The difference shows up in one specific class of bug, the kind that lives not in the change but around it. You alter the shape of an object in one module and another service still consumes the old shape. You rename a field and a migration written against the old name stays behind. None of that is visible from the diff, because the diff itself is correct. It becomes visible when you have the rest of the codebase in front of you, which is exactly what Greptile is trying to have. The second piece is custom rules. Instead of a configuration file with its own grammar, you write your house conventions in plain English, along the lines of never using a given date library or requiring every new API route to carry a permission check. The tool applies them on each review and speaks up when something drifts. It connects to GitHub and GitLab, and that is the whole list. Anyone whose code sits on Bitbucket or Azure DevOps cannot use it today, which rules out a slice of the corporate market before price even enters the conversation. ## What it is like day to day What people who have run it report and what its documentation describes agree on the opening stretch: setup takes minutes and the first review lands on its own as soon as the next pull request opens. The initial indexing takes as long as the repository size demands, and after that comments appear before any human sits down to look. The day-to-day problem is noise, and this is not a stray impression. The most cited third-party comparison, published on 2 June 2026 over fifty real pull requests from projects including Sentry, Cal.com and Grafana, credits it with an 82% bug catch rate against CodeRabbit's 44%, and in the same batch eleven false positives against two. Both numbers travel together and describe the same temperament: it looks harder, finds more and gets it wrong more often too. A direct competitor, DeepSource, adds a caveat worth carrying while reading that 82%, which is that the original figure comes from the vendor's own benchmark of fifty pull requests across five repositories rather than from an independent measurement. The vendor has moved on this. With version 4, announced on 5 March 2026, it published that the share of comments addressed by the change author rose from 30% to 43% after a month of running both versions side by side. That is its own number, measured by itself, but at least it is dated and says where it comes from. Even so, fewer than half the comments getting addressed is still a lot of reading. Hence the habit that decides whether a team stays or walks: narrow it down. Excluding generated directories, muting comment categories nobody cares about and writing two or three house rules turns an avalanche into a short list. Whoever installs it and leaves it as shipped ends up ignoring its comments within two weeks, which is the worst possible way to spend the budget. ## Pricing and plans There is a free plan, called Starter, with unlimited repositories, 50 credits a month and one active developer. A standard review spends one credit and a deep review spends three. The entry paid plan costs from $30 per seat per month, includes 50 credits per seat and charges a dollar for each additional credit. There is a 14-day trial, custom pricing above that, and 50% off for pre-Series A companies with under two million dollars of revenue in the past twelve months. Non-commercial MIT or Apache licensed projects that meet the requirements pay nothing. That per-review charge is the contentious part, and it is recent. Until March 2026 the rate was flat. The change did draw public protest, though it is worth sizing before leaning on it: there is a critical page published on 30 April 2026, signed under a pseudonym, collecting cancellations and the case of open source maintainers who say they were invoiced despite being approved for free use, plus a Hacker News submission from that same day which stalled at 17 points and three comments. A real, argued complaint rather than a firestorm. The underlying claim is that the included quota was designed for a working rhythm that no longer exists, and that protest cites a developer who opened 571 pull requests in thirty days. The vendor's answer was that fewer than 10% of active users would exceed the quota. Before signing, do one very simple and very local piece of arithmetic: how many pull requests your team actually opens in a month, counting the ones agents open. If that number sits comfortably below the included quota, the headline price is the price. If it goes over, your real cost bears no resemblance to the website and the decision has to be made with the right figure in view. ## Who it is for (and who it is not) Greptile pays off when the codebase is large, old and tangled, with dependencies nobody holds in their head all at once. There the indexing earns back the noise: you would rather read eleven surplus warnings than lose an afternoon chasing a cross-service breakage nobody saw coming. It also suits teams with strong house conventions, because writing those in natural language is cheap in time. It is not for anyone who measures work in pull requests per hour. With agents producing changes in chains, the billing model works against you and there are flat-rate options that do not tax throughput. Nor is it for teams whose code lives outside GitHub or GitLab, or for anyone after a quiet safety net that only speaks when it is confident, because this one speaks a great deal. If your team sits at the other end, with few changes a month and little patience for configuration, [CodeRabbit](https://serchai.com/en/reviews/coderabbit/) reaches usefulness sooner. The segment ranking is in [the best AI code review tools](https://serchai.com/en/best-ai/code-review/) and the wider map is in [the best AI tools for coding](https://serchai.com/en/best-ai/code/). ## Alternatives to Greptile The natural comparison is [CodeRabbit](https://serchai.com/en/reviews/coderabbit/), which covers more platforms, charges a flat per-developer rate and comments less and with fewer false positives, at the cost of catching less according to that same comparison. If the code already lives on GitHub and review is not the priority, the pull request review bundled into [GitHub Copilot](https://serchai.com/en/reviews/github-copilot/) comes free with what you already pay for, though it looks less deep. And for anyone working with agents, something like [Claude Code](https://serchai.com/en/reviews/claude-code/) can review on demand without any per-pull-request automation. ## Frequently Asked Questions ### How is it different from other automated reviewers? It indexes the whole repository before it opines, rather than reading only the change. That lets it flag breakages the diff never shows, and it is also what makes it comment more than it should. ### What does it really cost if my team opens a lot of pull requests? That depends on volume, and that is the catch. From $30 per seat per month with 50 reviews included, and a dollar per review after that. Count a real month of pull requests first, then decide. ### Does it work with Bitbucket or Azure DevOps? No. Today it integrates with GitHub and GitLab only. ### Is the 82% catch rate people quote reliable? Treat it with caution. The figure comes from the vendor's own benchmark of fifty pull requests across five repositories, so it works as an order of magnitude rather than an independent measurement. ### Can I try it without paying? Yes. The free plan gives 50 credits a month for one active developer with unlimited repositories, and the paid plans come with a 14-day trial. ## Alternatives - [CodeRabbit](https://serchai.com/en/reviews/coderabbit/) — Automatic pull request review with context: it comments before a human arrives. (3.7/5)