The right choice depends on the job. These are the reasons to choose either product, or keep both.
Choose Iris Code when
You need the answer to be identical every time, because it gates a push.
You cannot grant a third party read access to the repository.
You want feedback while writing the code, not after opening the pull request.
You want the check to run on every save without a per-run cost.
You need a finding you can defend: a named rule at a named line, with published precision and recall.
What CodeRabbit is
CodeRabbit is an AI code review service. It connects to your repository, reviews pull requests, leaves inline comments, offers one-click fixes, and provides chat about the change, with IDE and CLI review surfaces alongside the pull request one. Essentials is $24 per developer per month billed annually, Team is $48, and reviews are free forever on public repositories.
The complete comparison.
Supported, partly supported and unsupported capabilities. Read each note for the actual scope.
Iris Code compared with CodeRabbit
Capability
Iris Code
CodeRabbit
How it decides
Same answer on every run
YesFixed rules. No model, no temperature.
NoAsk twice, get two answers.
Accuracy you can check
YesPrecision and recall published per rule, per language.
No
Safe to gate a push on
YesA verdict that varies cannot be a gate.
PartlyBuilt-in pre-merge checks are included in Essentials; Team adds custom checks.
Judges whether the change is correct
NoIt has no model of what the code is for.
YesThe thing it is genuinely good at.
Where your code goes
Runs entirely on your machine
PartlyThe editor and CLI always do. Cloud scans are opt-in per repository and run on ours.
No
No repository access granted
PartlyNot for the editor or CLI, which need nothing. A cloud scan installs a GitHub App with read access to the repositories you pick, and only those.
NoA provider app with read access.
Works offline
PartlyLocal source analysis runs on your machine. Licensing, live dependency lookups, sync, and cloud scans need the network.
No
When it runs
While you are typing
YesSquiggles and the Problems panel.
PartlyIDE review runs on demand.
At the push, before a pull request exists
YesGit hook, locally.
PartlyThe CLI can run earlier, but it is not a gate.
On the pull request
PartlyThrough configured CI or opt-in cloud pull request scans.
YesWhere it lives.
Cost
Free tier covers private repositories
Yes
PartlyOpen-source PR reviews are free; limited IDE reviews need no subscription.
Costs nothing per run
YesRun it on every save.
NoInference is bundled, with usage-based add-ons.
Paid tier under $10 per developer
YesLoading prices a month, adjusted by country.
No$24 per developer per month, annually.
Free on public repositories
Yes
YesFree forever.
Claims about other products were checked against their own published material on 12 October 2026. Features and pricing change. Tell us about a correction.
Questions before you choose.
Is Iris Code an AI tool?
No. Iris Code contains no model and makes no inference calls. Every finding comes from a deterministic rule, which is why the same file always scores the same. Its purpose is to check code that an AI wrote, not to write or review it with another AI.
Why would I want deterministic checks when AI review is smarter?
Because you cannot gate a push on something that answers differently each time. A rule that fires on Monday and not on Tuesday is not an enforcement mechanism, whatever its average quality. Determinism is not a claim about intelligence, it is a requirement for blocking.
Does Iris Code catch things CodeRabbit misses?
It catches a different class reliably: hardcoded secrets including ones deleted from git history, duplicated blocks across files, known dependency vulnerabilities, and quantified complexity growth. An AI reviewer may catch any of these on a given run. Iris Code catches them on every run.
How accurate is Iris Code?
Every rule is measured for precision and recall per language against a labelled corpus, and the results are published rather than summarised into a single marketing number. A checked-in baseline blocks any release that regresses a rule.
Put it to work.
Free file analysis needs no account or repository connection. Open a supported file and inspect its findings.