Agentic review pipeline for GitHub PRs and GitLab MRs

A code review agent that works from real change context.

7review takes an exact PR or MR, enriches it with SCM metadata and diffs, selects repository knowledge, skills, and memory, routes model review through configured providers, validates findings, publishes review output, and waits for human approval before final publication.

Start with quick setup View architecture

SCM enrichmentContext and memory recallModel review with toolsHuman approval gate

7reviewcode review agent

7review agent

Agentic review loop

The operator chooses the PR or MR. The agent handles enrichment, context assembly, model review, validation, publishing state, memory, and final human approval gates.

Ingest

A manual request or policy-gated webhook enters the bounded worker queue.

Understand

The agent loads SCM metadata, diffs, repository knowledge, skills, and approved memory.

Review

Configured model providers review the assembled context with governed read-only tool access.

Approve

Validators classify findings, publish review output, and wait for human approval before final publication.

7review

Operator commands

7reviewOperator commands

setup

$go run ./cmd/7review setup

manual github

$7review review github --repo owner/repo --pr 7 --server http://localhost:8080

manual gitlab

$7review review gitlab --project 25 --mr 19 --server http://localhost:8080

readiness

$7review status --server http://localhost:8080

WEBHOOK_REVIEW_MODE=manual_firstFinal publishing requires human approval

Manual Reviews

Request one GitHub PR or GitLab MR through the same bounded queue used by webhooks.

Webhook Policy

Control automatic review with mode, label, project, repository, and branch rules.

Docker Runtime

Run the agent with Headroom and MemPalace on the private Compose network.

Troubleshooting

Diagnose queue pressure, token failures, sidecar readiness, and publish state.