Quorum
Gemini 2.5 Pro · GitLab MCP · Google Cloud

The coordination bugs
your linter can't see.

Quorum is an AI agent that reviews merge requests for distributed-systems anti-patterns — missing fencing tokens, incomplete sagas, retries without jitter, lost updates, and more. It investigates across your whole repo, posts structured findings, and opens draft fix MRs.

"SonarQube doesn't have rules for saga compensation. Semgrep can't reason across service boundaries. CodeRabbit can't search the full repo. Quorum can."

Live review playground

Paste a GitLab MR or GitHub PR URL and watch the agent work — step by step.

Pipeline

Agent trace

Waiting for events…

Findings

No findings yet.
Investigating…

🧭 What happens next

Does Quorum stop when an MR is blocked? No — "blocked" only marks the merge as unsafe. Quorum still posts every finding and opens a draft fix MR for each critical bug. Merge a fix MR and the Cloud Run webhook automatically re-reviews the branch — so it's a loop: detect → fix MR → re-review → unblock.

Built on GitLab's MCP server

The agent's superpower is semantic_code_search over GitLab's MCP server — it reasons across the whole repository, not just the diff. Click any stage below to see what it does.

Chat with the agent — the Agent Engine Playground

A Vertex AI ADK agent calling GitLab's MCP server live. Watch a real session below — then open it yourself with your own Google Cloud credentials.

Agent Engine Playground · session replay
Open it yourself
  1. 1
    Open the deployment in the Cloud Console
    Open the ADK agent in Google Cloud →
  2. 2
    Click the Playground tab on the agent page
  3. 3
    Type a prompt — tap one to copy:
    The agent reads the repo + MR number straight from your message — that's how it knows what to review.
Need access? The Playground is IAM-gated in Google Cloud. Deploy your own in ~5 min, or have an admin grant you roles/aiplatform.user:
gcloud projects add-iam-policy-binding gen-lang-client-0294573094 \
  --member=user:YOU@example.com --role=roles/aiplatform.user
engine: reasoningEngines/1659372503279075328 · us-central1

14 coordination rules

A named taxonomy from Kleppmann, AWS, microservices.io, Confluent and the Google SRE book.

Three ways to run it

Same review pipeline, three entry points. Pick one to see it in action.

What you get
9
open-source repos with real bugs found
3
languages · Java, Go, Python
0
false positives across all runs