All projects

Warroom

0 comments

What it does

WARROOM is an agentic incident-response system that investigates production outages in real time. A team of specialized AI agents independently analyzes metrics, logs, deployments, and payment-provider telemetry. They share structured evidence, challenge weak hypotheses, identify the most likely root cause, and recommend a mitigation. The system provides a live investigation timeline, agent activity and tool-call indicators, deterministic safety checks, incident cancellation and timeout handling, and downloadable PDF and JSON audit reports.

How the agents run concurrently

WARROOM uses Mozaik’s event-driven agent runtime. When an incident starts, specialist agents subscribe to the same typed incident event and begin investigating simultaneously instead of running through a fixed sequential chain. Each agent uses Mozaik’s runLoop to: 1. Receive structured incident context. 2. Decide which approved tool to call. 3. Execute that tool through a typed registry. 4. Add the tool result back into its context. 5. Publish structured evidence to the shared event stream. Other agents can immediately react to these events. For example, the Skeptic challenges an early deployment hypothesis when Metrics reports contradictory evidence, while the Investigator combines findings from multiple specialists. The Commander acts last and can recommend mitigation only after deterministic safety gates confirm sufficient independent evidence and hypothesis validation.

Comments

No comments yet.

Sign in to like and comment

Newsletter

For developers who want to learn how to build self-organizing agents.

Join our online hackathon