Greptile on X

Fans credit it with catching real problems and with fit. One who switched from a rival says the reviews and the vulnerabilities it identifies are impressive; another names the ability to configure settings, rules and instructions as the main reason; a team member describes Greptile as feeling like an "extension of our team", responsive to feedback and focused on code review. Others point to the free open-source program, to fast reviews, to an eval where the tool won, and to chasing a 5/5 score on every PR. One ranks Greptile, alongside CodeRabbit, "generations ahead" of other tools in quality.

The undecided (14 people) weigh the idea more than the experience: whether a separate review layer earns its keep when other agents can run adversarial reviews, whether a $750M valuation proves sustainable revenue and impartiality, and what share of GitHub's daily PRs it really covers. The 5 impressed from outside cite that valuation and the 1% of GitHub PRs, and the new trick of driving an app to catch a button that vanishes on an iPhone screen. Those who left name the cost.

The bill is the one people name most. The worry is that agents open exponentially more PRs every month while per-review charges and $30 seats climb with them. One reports a $2k bill and a move to a local agent; another hit $300 a month and left for gh-aw. A per-PR price is called a worse version of per-seat pricing because a one-line config change costs the same as a 10K+ line feature. People describe their own fixes: a cheap pre-check on file names and diff size so the expensive review runs only on real logic, or flat monthly subscriptions they already pay for.

5 people say the review judges only the diff: it will score 5/5 on a fix without asking whether the approach was right, and a green score is no proof the behavior holds. 4 describe long write-ups, flow diagrams and style nits that bury the two or three comments that would block a merge. 3 report false positives — one critic rated only 68% of its findings as real — and 2 found setup for open source fiddly. 2 mention rate limits or the service being down.

CodeRabbit is the named rival. People rank Greptile ahead of it — one posts Greptile beats CodeRabbit even though CodeRabbit sent a free T-shirt; another calls the two "generations ahead" of other reviewers. Some run both: one team splits PRs between the two tools so that when a rate limit hits one, the other picks up the review.

The other side of the ledger is building or renting a reviewer yourself. One says building your own Greptile is one of the best ways to spend engineering time now, because verification and boundaries are the job; another argues the opposite, that you either ship the best product or waste time and effort rebuilding Greptile in-house. People who left over cost went specific: a local agent through Codex CLI after a $2k bill, a custom reviewer that checks prod DB, logs and Linear issues for about $300 a month, or a coding agent doing the same job for $1 where a seat costs $30.

The rest name one or two people each: Claude and Codex subscriptions, GitHub Copilot, Shadowclone, OpenQodex and gh-aw.

The verdict question comes in a few forms. One asks flatly "Greptile or CodeRabbit??"; another says they need a review tool, that everyone tells them to try CodeRabbit, and then asks whether Greptile is good. People who already run both get asked how they split the work — whether one catches a class of bugs the other misses or the two overlap — and why anyone would run two at once, review limits being the guess.

3 people ask about wiring it up: one is setting up Greptile and CodeRabbit for open source and asks for a contact after hitting trouble; a reply moves to connect them with the founder. Another asks whether a browser-automation tool can be plugged into Greptile so their web apps get reviewed like a human integration test.

The rest are mechanics. 2 people ask whether the 5/5 score matches what a careful human would flag; one answers that the score usually lines up with their own Claude Code and Codex checks, and that they trust it less when they skip those. 2 ask whether each push re-reviews the whole PR, and how the evals decide a comment was worth reading.

The request is consistent: the score is the part people read, and everything under it should be shorter. One says 4/5 or 5/5 is all they care about and calls the diagrams "a complete waste of tokens"; another wants the two or three comments that would actually block a merge, with the rest behind a click; a third asks for a short list of top issues under the score, and a joke reply turns it into a flow-chart request. One raises it directly with the founders to have the diagrams removed.

Price is the other wish. The hope is for a cheaper base review as agents open more PRs every month, phrased as bringing the price down before the bill climbs with the PR count. One describes the workaround they settled on: a cheap pre-check on file names, diff size and lockfile bumps, and the paid review only on PRs that touch real logic.

The rest are single requests: a tighter GitHub setup for Greptile alongside Codex, CodeRabbit and Claude; a CI check that holds docs to the ASD-STE100 writing standard; browser automation so a change is exercised like a human integration test; and, from one person, that PR review simply run itself.