No incident loaded
Pick a scenario and load it into FalkorDB.
Demo scenarios use synthetic telemetry on the OpenTelemetry Demo topology. Approved actions are executed as a simulation against the graph, not against real infrastructure.
Service graph in FalkorDB
Healthy
Degraded
No telemetry
Host
Root cause
Causal path
Breaker (min-cut)
Incident timeline
Run an investigation to see when each service degraded and which changes landed before.
Agent trace
Each tool the agent calls appears here with its duration.
Root cause
No investigation yet.
Who to page
Run an investigation.
Ask the graph
Each question runs one fixed, human-written tool. No query is generated.
The local LLM (Ollama) only picks tools over MCP; its sentence is rejected if it names anything the tools did not return. Slow on CPU.
Evidence from the graph
—
Blast radius and containment
—
Proposed actions
—
Memory
—
Accuracy: 0.95 top-1 on held-out synthetic scenarios (our fault model); on real RCAEval fault-injection data the graph ranking is on par with a strong no-graph baseline. Details in docs/evaluation.md.