Six stages from alert firehose to drafted runbook. Every stage is inspectable; no stage hides behind an LLM call.
Salvo connects to Prometheus, Datadog, Grafana, or a plain webhook. No agent to install; no sidecar to inject.
Temporal + service + label correlation; 42 alerts from one P0 event become one incident, not forty PagerDuty pages.
Alerts emit from services. Services depend on services. The graph is derived from your service mesh and deploy pipeline — not guessed.
Deploys, schema migrations, traffic changes, upstream partial outages — each is a candidate with an evidence path and a confidence score. Salvo never picks one silently.
For the top candidate, Salvo drafts an ordered runbook. Every step has a confidence score. Every mitigation is copy-to-clipboard; the agent never executes.
The incident closes with a signed timeline (who-did-what, salvo-vs-human) and a pre-populated postmortem doc. You write the narrative; Salvo owns the scaffolding.