Code Review
Code review is one of the applications built on SourceAnt's knowledge layer. When a pull request opens, SourceAnt reviews the change and posts findings directly on the PR, so problems surface before a human reviewer spends time on them.
How It Works
- Your repository sends PR events to SourceAnt through the GitHub App (cloud) or your own webhook configuration (self-hosted).
- The reviewer analyzes the diff with the configured model and posts review comments on the pull request.
- The review draws on the same context interfaces as the rest of the platform. A local review can use durable knowledge only from that same local instance. Code evidence stays pinned to the commit under review, so uncommitted symbols from a local working tree cannot enter it. SourceAnt Cloud adds managed team context and continuous indexing at scale.
Where reviews show up
- On the pull request. Inline comments where your team already works.
- In SourceAnt Lens. A dedicated surface for reviewing changes with human-in-the-loop oversight. See SourceAnt Lens.
Choose what the review includes
Open Settings → Review and check the selected scope before saving:
- Include nitpicks is off by default. Enable it for style, naming, clarity, documentation, refactoring and optional improvement suggestions.
- Expert review passes accepts
autofor automatic selection, or exact skill identifiers, one per line. An empty list disables expert passes while general review and guidance still apply. - Files excluded from review accepts glob patterns, one per line. Lockfiles are excluded by default; dependency manifests remain included. Clear the list to review every file.
- Show minor suggestions in overview, Show critical findings in overview and Show regressions in overview control those summary sections independently. All three are on by default.
See Skills to manage reusable instructions and Settings for inheritance and reset behaviour.
Running it
- Cloud: sign up, connect your repositories through onboarding, and reviews start on your next PR. See Cloud Setup.
- Self-hosted: run the open-source core with your own model keys. See Self-Hosting.
Related
- Knowledge Management: the layer reviews draw their context from.
- Issue Triage: the other automation that runs on your PRs and issues.