Every pull request gets a review. Before a human reads it.

    Code Reviewer posts inline findings with severity, a summary, and a score on every PR. Your rules, your model, your budget.

    Get KimchiBook a demo
    curl -fsSL https://github.com/getkimchi/kimchi/releases/latest/download/install.sh | bash
    Code Reviewer findings and review summary
    AkamaiWeWorkRokuCiscoYotpoSwisscom

    Open a PR. Get a review.

    No workflow file to commit, no bot to babysit.

    01

    PR opened

    A webhook fires the moment a PR opens or gets pushed to.

    02

    Diff goes to the model

    Your instructions and excluded paths applied, routed to the model you picked.

    03

    Findings land inline

    Findings post inline, each with a severity: critical, warning, or info.

    04

    Summary and score

    One summary: score, security concerns, missing coverage.

    Eight priorities, in order.

    Correctness before style. Bugs before nitpicks. Same order, every review.

    01Correctness
    02Error handling
    03Security
    04Concurrency
    05Performance
    06Design
    07Maintainability
    08Testing

    What lands on your pull request.

    A welcome comment tracks the run. One summary closes it out.

    Pull request with kimchi-review findings and review summary

    Your rules, not the bot's defaults.

    Free-form instructions per repository. Exclude generated code. Pick the model. Change any of it when your standards change.

    Review instructions

    "Reject any PR that adds a dependency without a named owner." Your words, in every review.

    Excluded paths

    vendor/, generated/, dist/ - never sent to the model.

    Model per repo

    Most review bots hide which model reads your code. Kimchi puts it in your config.

    repository settings · acme/payments-api
    # Review instructions, in plain language
instructions: "Reject any PR that adds a
  dependency without a named owner."

# Paths the reviewer never sees
exclude:
  - vendor/**
  - generated/**
  - dist/**

# The model that reviews this repo
model: kimi-k2.7

    Reviews run inside the enforcement layer.

    Reviews run on the same governed infrastructure as everything else on Kimchi.

    Attribution

    Every review is metered

    Tokens, duration, and cost per run, attributed to your org.

    Budgets

    Reviews count against caps

    Reviews are model calls. They run under the budget caps you set.

    Data

    Your code is not training data

    Prompts, diffs, and outputs are excluded by architecture. In any mode.

    Audit

    Every event stored

    Webhook events are recorded for audit and replay.

    Installed in minutes. No tokens to babysit.

    Install the GitHub App from the console. Pick repositories. Reviews start on the next PR.

    GitHub App

    1-click install. Installation-scoped tokens rotate automatically. No PAT to manage.

    Per-repo settings

    Instructions, excluded paths, and model choice, all in the console.

    GitLab

    GitLab MR is enabled through a dedicated user access token or through a group access token via the GItLab API.

    Every PR, reviewed. From the first one.

    Install the app, open a PR, get a review. No per-seat bot pricing.

    Get KimchiBook a demo