@mneme-ai/correlator

Phase 3 — error/incident correlation engine for Mneme.

The differentiator. Joins commits + incidents + time into one graph, so you can ask:

"Every time PaymentService.charge is touched, a Stripe webhook 500 spikes within 48h."

"This PR touches code that has caused 3 of the last 5 incidents in OrderQueue."

"Incident INC-1287 was likely introduced by commit a1b2c3d (87% confidence — same file, 14h before the spike)."

What's in here

• temporal.ts — TemporalCorrelationEngine (temporal proximity + file overlap)
• adapters/sentry.ts — first pager adapter (REST API; live)
• adapters/datadog.ts — Datadog adapter (stub, in progress)
• adapters/manual.ts — JSON file input (works today, useful for testing)

Usage

import { TemporalCorrelationEngine, SentryAdapter } from "@mneme-ai/correlator";

const sentry = new SentryAdapter({
  orgSlug: "my-org",
  projectSlug: "web",
  apiToken: process.env.SENTRY_TOKEN!,
});

const incidents = await sentry.fetch({ since: "2025-01-01T00:00:00Z" });
const engine = new TemporalCorrelationEngine();
const correlations = await engine.correlate({
  commits,
  incidents,
  windowMs: 7 * 24 * 60 * 60 * 1000,
});

License

MIT.