Outside-in (Uptime Kuma)
Synthetic probes check public reachability, TLS, and expected content. A failure triggers the shared triage pipeline.
Back to HARP
Interactive Simulation
Alert to evidence to human decision.
Incidents can begin through two verified signal paths—an external synthetic failure from Uptime Kuma, or an internal infrastructure alert from Alertmanager. Both join a common normalized flow: deduplication, HARP evidence retrieval, Nero-Camp task creation, operator notification, and human closeout.
Verified implemented flow Presentation simulation Optional / future capability
Inside-out (Alertmanager)
Prometheus rules fire on cluster and infrastructure thresholds—disk pressure, pod crashes, node health. Alerts normalize into the same downstream flow.
What the presenter says
"Notice that every AI-facing step is constrained. HARP retrieves evidence and safe diagnostics. Nero tracks the decision. LiteLLM can summarize in the wider architecture, but it does not replace citations. No mutation occurs automatically—the human records the result, and the RCA becomes future evidence."
What the audience sees
HARP result: runbook-wordpress-pvc-storage-fix.md
Nero task: SY-23 (triage ready)
Source: Synthetic Monitoring gateway
Why This Works
The loop creates reusable operational memory.
01
Observe
Uptime Kuma or Alertmanager produces a signal.
02
Normalize
Synthetic Monitoring gateway deduplicates and maps to a shared contract.
03
Retrieve
HARP finds prior context from runbooks and RCAs.
04
Ground
Evidence pack contains citations, snippets, confidence, and gaps.
05
Approve
Nero-Camp records human decisions. No mutation occurs automatically.
06
Document
Outcome becomes an RCA, runbook update, or session log for future retrieval.