BLACKBOX is the on-call agent that reads your service graph,finds the change that caused the outage two hops upstream,and pages the human who actually knows.
Production systems are a graph. Your incident tooling is a listof metrics. That gap is why 3am takes four hours.
BLACKBOX puts topology, deploys, people and prior decisions intoone FalkorDB graph and lets an agent traverse it. Not a dashboardto read — a surface to think on.
The graph isn’t where the agent looks things up.It’s where the agent thinks.
An alert fires
p99 latency on checkout-api crosses threshold. One signal, no context, 3am.
The graph answers
One traversal returns every service, dependency and customer path degrading with it.
The cause surfaces
It walks upstream through deploys and ownership until the signal becomes a causal story.
It writes back
The resolved incident lands in the graph as a node. The next one of this shape starts ahead.
One graph. Nine typed tools. Three agents in the loop — and ashorter path from first alert to confident action.
Blast radius in one traversal
Alerts tell you which service is loud. BLACKBOX returns the whole degrading subgraph — every dependency, every customer-facing path, ranked by hops from the origin — before the incident spreads.
Cause, not correlation
It walks upstream through recent deploys, pull requests, config changes and service ownership. The answer is not “these metrics moved together.” It is “this deploy, two hops up, at 02:14.”
Memory that compounds
Every resolved incident is written back as a (:Postmortem) attached to the services it touched. The second incident of a shape resolves in a fraction of the hops, because the graph remembers the first.
Run the same class of incident twice. The cold run walks thegraph from nothing. The warm run finds the postmortem BLACKBOXwrote the first time, attached to the services involved, andskips straight to the cause.
Open the working console and watch BLACKBOX trace a real incident from alert to answer — then do it again, with memory.