Demo dashboard. Read-only, seeded with sample data. Sign in to run your own.

Investigator

The AI that reads it first.

When a raw monitor alert arrives, the Investigator reads your repo and recent deploys, ranks likely root causes, and folds them into the page. Read-only: it cannot touch code, ack incidents, or change how you get paged. If it fails for any reason, the alert pages exactly as before.

Enrichment

Enrich thin alerts before the first page goes out.

Connected repo

GitHub · read-only

Connected project

Vercel · recent deploys

Daily budget

7 of 40 calls today

38,412 of 150,000 tokens today

Over budget, alerts page exactly as before, with an audit note instead of an analysis.

Reproducer

After it ranks the causes, prove the top one: clone your repo into an isolated sandbox, write a failing test, run it. It never touches your code, opens PRs, or changes how you get paged. Runs after the page.

2 of 10 runs today

Only the top hypothesis at or above this confidence is worth a sandbox. Needs a connected repo.

Fixer

When a reproduction is confirmed, draft a fix, re-run the test to verify it, and open a PR — gated by your risk profile: it never touches a protected path and never auto-merges. v1 opens PRs only. Needs a write credential.

0 of 5 runs today

Separate from the read-only token above and never given to the sandbox. A GitHub App is preferred over a PAT.