Codacy

You will get the most from the part that costs nothing. The editor plugin scans while you type and fixes issues before they ever become a diff, which is the deciding daily trait, because the cheapest place to fix anything is the moment you wrote it, not two days later in a comment thread. The hosted review then catches what survived.

Pick it if you want both halves of that from one place and one invoice. Pick CodeRabbit when what you actually want is a reviewer that argues with you in the pull request.

Look at what is being compared. One input is code and the other is whatever the author typed into a text box at the end of a long day, and the model has no way to know which of the two is wrong. On a team that writes terse descriptions this produces confident findings about intent that nobody actually got wrong, and every such finding costs a reviewer the time it takes to dismiss.

What it does right: merge gates are policy rather than opinion, so the blocking decision stays deterministic even when the commentary is not.

1. The pipeline is hybrid: rule-based analysers produce findings with known behaviour, and a model produces findings with unknown behaviour, and the published material does not say which comments come from which source. 2. Findings are handed onward to a coding agent for correction, which puts a second model downstream of the first without an intervening check.

No false-positive rate, no evaluation set and no methodology accompany the reviewer. The observation: a vendor that owns a deterministic engine could measure its model against it cheaply, and has not published the result.

The interesting asset is the old one. Rule engines across dozens of languages took years to build and cannot be replicated by a model, which makes them the switching cost while the reviewer is the growth story, and the annual discount at $18 is the mechanism for locking the former in before competitors take the latter. Free for open source is distribution priced at zero, which is the right call.

Likely acquirer: a code host or a development platform that wants quality gates it did not write. Position: a real business with an unfashionable moat, priced accordingly.

One sentence on the demo: it commented usefully and it commented a lot. The arithmetic is straightforward at $1,260 a month on the monthly rate, less if we commit for a year, and the editor plugin can go to every engineer at no cost, which is a rare thing to be able to say. It runs on pull requests as a check, so it fits our existing flow without a new pipeline. The business tier is custom, which is where the controls conversation goes.

Onboarding is half a day. Approved with conditions: standards agreed first, blocking gates second.

The two things I can use are both at the edges. There is a documented interface my own agent can query, and a tool-server endpoint that makes the findings available to whatever I am building, so the output is at least programmable even when the producer is not.

The producer is thoroughly not. I cannot read the analysers, cannot supply a key, cannot run inference on my own machine and cannot substitute a model I trust more. Everything that decides what counts as a defect in my repository happens somewhere I am not allowed to look, which is a strange place to put a definition of quality.