Osuite is production intelligence for going from a P0 incident to RCA + fix in minutes. System breaks → runtime graph → AI investigation → root cause → fix → <5 min. Unlike traditional observability, Osuite correlates telemetry, code, deployments, and runtime context into a context mesh, giving both engineers and AI the context needed to investigate. An AI investigation agent works directly in your IDE, helping you understand what broke, why, and how to fix it.

Alerts only catch what you predicted. Osuite's new Anomaly Detection watches every service and deploy, and an agent vets each detection so you only hear about the ones that need you. It investigates on its own, links every claim to evidence, hands the fix to Claude Code, Cursor or Codex, and re-checks the evidence once the release ships. It learns your normal from every call you make, so it gets quieter over time.

At 9:40, a deploy broke checkout. No alert fired.

That's the trouble with alerts: they only catch what you predicted. A brand-new error slips under every threshold until a customer tells you.

So we built Anomaly Detection into Osuite.

→ It watches every service and deploy. An agent vets each detection before it reaches you. In the demo stack, fourteen were fine and one needed a human, so that's the only one you hear about.

→ It investigates on its own. In the demo in the video, it found the cause in 74 seconds: 48 database queries per order instead of 1. Every claim links to the chart, trace or query behind it, and you can question it in chat.

→ One click hands the fix to Claude Code, Cursor or Codex.

→ When the fix ships, it re-checks the evidence before calling it fixed.

→ It learns your normal. Every call you make (Expected, Not worth it, Detector wrong) teaches it what needs you, so it gets quieter over time.

It's live now. Try it on one service: osuite.io

One question for you: what's the last incident your alerts missed? We'd love to know whether this would have caught it.