Gitar

The trait that decides this day to day is that it is interactive on the pull request. You answer it, you tell it what you actually meant, and it takes the instruction from the comment rather than repeating the same objection on the next push. That turns a bot you learn to scroll past into something closer to a junior reviewer who listens.

You give up any say in where the work happens, because this lives in the cloud on your repository. Pick it when review is the bottleneck. Pick CodeRabbit when you want commentary and nothing more.

The risk is separation of duties. Autofix writes to the branch, and the higher tier documents auto-approve, merge blocking and auto-apply. An organisation that turns all three on has an agent authoring changes and then signing off on them. Every governance model in code review exists to stop exactly that, and no documented human gate sits between the two roles.

What it does right is the noise problem. Flaky-test detection and de-duplication are named features, and a review tool that suppresses its own repeats is rarer than one that finds bugs.

1. The verification signal is external. Rather than judging its own patch, the agent reads the failing build and keeps working until the pipeline reports green, so correctness is decided by something that was already trusted. 2. That is the right dependency direction, and rare on this board. 3. No benchmark is published, so there is no methodology to audit.

The documentation is thin about the loop's limits: no iteration ceiling, no stopping rule, and no description of what happens when a suite is red for reasons the diff did not cause.

The exit happened before you evaluated it. Sonar is listed as the vendor, and that resolves the usual startup question and replaces it with a product-strategy one: an autofix agent absorbed by a static-analysis incumbent gets distribution into an installed base immediately, and gets its roadmap set by that suite's priorities rather than its own.

Pricing power comes from the parent, not from this product line. Expect bundling into a quality tier before you see an independent price rise. Position: buy it as part of the suite you already renew, and do not underwrite a workflow that assumes it stays a separate SKU.

Sixty seats lands at $1,200 a month on Core and $2,400 on Pro, which is legible budget rather than a meter, and that alone puts it ahead of most of this category in a finance review. A 14-day trial is enough to run it against two repositories before the purchase order.

The blockers are above that line. Self-hosted code hosting and API access are both Enterprise features, so anyone on an internal GitLab is quoting custom before they can pilot. Identity integration is not documented anywhere I can cite. Approved with conditions: Core tier, two repos, and a written answer on retention first.

There is nothing to read. The licence is proprietary, execution is somebody else's cloud, and the model is undisclosed, so I cannot tell you what is generating the patch that lands on my branch. The one lever that would matter, pointing it at my own key, is documented as an Enterprise-tier feature, which means the knob exists and I am not allowed to touch it.

No fork survives this vendor because there is no source to fork. Grudging credit: at least the tier gating is written down instead of discovered during a trial.