Code Terrier

Code in your coding agent,
Review in Code Terrier.

Agents write more code than anyone can read. Code Terrier finds verified issues on the exact changed line and, in auto mode, re-checks later pushes.

Install on GitHubOpens GitHub. Auto mode reviews new PRs and later pushes by default; change the trigger any time.

Code Terrier never approves or auto-merges. It only posts COMMENT reviews with read-only tools.

Model families in the review architecture

Claude·GPT·Gemini·DeepSeek·Kimi·Qwen·GLM·Grok

The architecture supports up to three configured model families. The shared pool currently uses one model; BYOK controls the reviewer setup.

tokens spent on reviews · review runs.Loading live aggregate…Review runs are executions, not unique pull requests.

Review a GitHub pull request

Need an on-demand pass? Comment on any pull request.

Bring an Anthropic key or an OpenAI-compatible endpoint, or run on the shared pool with no key at all.

Review local code in your terminal

Have local or uncommitted code? Run a direct review from the terminal; findings return there and are never posted to GitHub.

Paste this into your coding agent.

Frequently asked questions

Does Code Terrier approve or auto-merge pull requests?

No. Code Terrier only posts COMMENT reviews. It never approves, never auto-merges, and never pushes or commits code.

How does a new commit change the review?

In auto mode, a new commit starts the next review. Prior human comments are carried forward as bounded, untrusted context, and findings that no longer apply close quietly.

Can I bring my own model key?

Yes. Bring an Anthropic key or an OpenAI-compatible endpoint so reviews run on your own account, or use the shared model pool with no key at all.

What trigger modes does Code Terrier support?

Three: auto (reviews new PRs and later pushes), on-creation (reviews only the first look), and manual (reviews only when @code-terrier review is commented). Later pushes are automatic only in auto mode.

Continue exploring

Start with your question

Parallel, then checked

Configured reviewers can run together, up to the three-reviewer limit. Near-duplicates are consolidated, then the verification stage checks line legality.

Inline, and categorized

Security, Bug, Flag and Slop findings anchored to the exact line — not a wall of prose at the bottom of the PR.

Fixes you can commit

Findings arrive with a GitHub suggestion you can apply; a non-blocking check run follows every head commit.

Push and it re-checks

In auto mode, a new commit starts another review. Prior human comments are carried forward as bounded, untrusted context.

Bring your own key

Use Anthropic or an OpenAI-compatible endpoint so review runs on your own account. No key? The shared pool covers you.