Skip to content

kaid

kaid logo

AI-powered code review that fits the way you already work. kaid picks the right files, applies your rules, reviews related changes together, and returns findings you can act on - from your terminal, your editor, or CI.

Quickstart Installation View on GitHub

  • Review in minutes

    Install the CLI, connect any OpenAI- or Anthropic-compatible endpoint, and run your first review against a workspace, a branch range, or a single commit.

    Quickstart

  • Context-aware findings

    Related files are grouped into one conversation, so the reviewer sees a handler, its service, and its tests together instead of guessing from a single hunk.

    Architecture

  • Your rules, enforced

    Layer project, global, and system rules. Exclude generated code. Merge house standards with built-in language guidance on a per-path basis.

    Review rules

  • Built for automation

    Emit JSON, suppress progress output, and post findings back to GitHub or GitLab. Run it locally, in a pipeline, or through an agent integration.

    CI/CD recipes

Why kaid?

  • Filters before it spends tokens. Binaries, ignored paths, unsupported extensions, deleted files, and oversized diffs never reach the model.
  • Resolves a rule per file. The four-layer chain gives you predictable scope without a separate configuration service.
  • Keeps the review coherent. Semantic grouping and a shared per-group conversation let the model reason across related changes.
  • Leaves an audit trail. Every run writes a JSONL transcript you can replay in the built-in session viewer.
  • Meets you where you work. Use a hosted model, a local gateway, a coding agent's subscription, or MCP tools.

How it works

diff -> filter & rules -> semantic groups -> review rounds -> comments

The pipeline is deterministic where it matters: file selection and rule resolution are plain Go code, while the model only sees the diffs and context you have allowed. Nothing is reviewed twice by accident, and nothing is sent without a reason.

You want to... Start with
Review local changes before committing kaid review
Review a feature branch in CI kaid review --from main --to feature
Replay what the model saw kaid viewer
Reuse an existing coding agent Delegation mode
Add repository knowledge MCP servers

Get the CLI

curl -fsSL https://raw.githubusercontent.com/CHE3MZ/kaid/master/scripts/install.sh | sh

irm https://raw.githubusercontent.com/CHE3MZ/kaid/master/scripts/install.ps1 | iex

Then verify the binary and configure a model:

kaid version
kaid config provider
kaid llm test

Already have an endpoint?

Skip the interactive setup and write the configuration directly with kaid config set, or point kaid at your existing ANTHROPIC_* / KAID_LLM_* environment variables. See Configuration.

AI review is a second pair of eyes, not a replacement for tests or human judgement.1

  1. Model output can be wrong in both directions - it may miss a real defect or flag code that is correct. Treat findings as leads and keep your normal review process in place. See the FAQ for what leaves your machine. ↩