reviewd
A local code reviewer that needs no CI: it watches your PRs and reviews them in your own repos using AI subscriptions you already pay for.
reviewd is a code reviewer that lives on your own machine and watches pull requests on GitHub and Bitbucket. No CI setup, no new cloud service: it checks your locally cloned repos on a schedule, and when a new PR shows up it reviews it and posts its comments back to the PR. Reviews run on AI subscriptions you already have, so you do not pay a second time just for reviewing. The repo has 70 stars. The review itself is solid: instead of looking at a shallow CI clone, it reviews right in your local repo, following imports to get full context. Before reviewing it builds an isolated copy in a git worktree, leaving your working tree untouched. It can also run the lint and test commands you configured first, and hand the failures to the AI as reference. Findings are graded by severity — only the serious ones become inline comments, the rest go in a summary comment — and new commits trigger a re-review that clears stale comments. Draft PRs are skipped by default. Safety is thought through too: it only reviews repos you already have locally; Claude runs in read-only mode, so the AI cannot write files or open MCP; each project’s config is read only from the repo owner’s copy, never from configs carried inside a PR, so a PR author cannot slip instructions into the config. You can give different repos different AI backends, models and review rules, or just run a one-off review without the background daemon. One thing to know: this is a local command-line tool, not a website you open and use. You need Python 3.12 or newer, the claude, gemini or codex CLI installed and signed in on your machine, plus a GitHub token with repo access or a Bitbucket app password. And review comments are only the AI’s input — a human still decides what gets merged.
@naomi.w