review
PR review bot
Post gl findings as inline comments on GitHub or GitLab pull requests, gated on your own severity threshold.
// features
> Six things it does, in order.
Every capability below runs from the same binary and the same config. Read them top to bottom; that is roughly the order you will meet them in a real repo.
// 01 — scan
Repo-wide static analysis
Grepline builds a symbol graph of your whole project, then reasons across files — so it catches the null that flows three functions deep, not just the one on the line you edited. Findings come ranked by severity with file and line.
$ gl scan ./src --severity high
auth/session.ts:88 null deref
store/lru.ts:142 write race
2 high findings · 0.9s
// 02 — fix
Suggested patches you can trust
Each finding carries a minimal diff. Preview it, edit it, or apply it — Grepline then re-runs the scan on the changed file to prove the fix landed and nothing regressed.
$ gl fix auth/session.ts:88
- const u = cache.get(id)
+ const u = cache.get(id) ?? load(id)
apply? [y/N] · re-scan clean
// 03 — the rest
agents
BetaFix agents
Hand a whole finding class to an agent that opens one branch, applies the patches, and stops for your review.
self-host
Bring your own model
Point --endpoint at a local or VPC model server; Grepline never assumes a hosted API and ships no default phone-home.
audit
Signed reports
Export SARIF or a signed Markdown report for compliance, with a stable hash per run for your audit trail.
ci
CI gate
One step in Actions or Buildkite fails the build on new high-severity findings and diffs against the base branch.
rules
Custom checks
Write repo-specific rules in plain language in grepline.toml; the model enforces intent, not just regex.