AI Agent Operations

Logs, metrics, traces, deploys,
one root-cause graph per incident

Cross-Signal Correlation joins the four pillars of observability into a single root-cause view. When an incident fires, Nova builds a graph: which traces are slow, which logs spike, which metrics broke, which deploy or change shipped, and how they connect. The graph answers "why" without you running five queries.

Get Started Talk to Sales
app.novaaiops.com / cross-signal-correlation
● LIVE
< 5s
Build time per incident
4
Signal types correlated
Auto
No queries to write
Graph
Visual + tabular view
The Correlation Graph

Every signal becomes a node, every link becomes an edge

Nova builds a graph of signals around each incident. Nodes are concrete things: a trace span, a log spike, a metric breach, a deploy event. Edges are statistically-validated relationships: this log spike happened in the same trace as that slow span, this deploy preceded that metric breach, this metric broke during that change window. The graph is the explanation.

  • Nodes are observed signals: one node per slow span, one per log error spike, one per breached metric, one per recorded change
  • Edges are validated links: edges only appear when correlation passes a 95% confidence test, not just temporal coincidence
  • Pruned to the relevant cone: the graph filters to the service blast radius so it stays readable, not a hairball
app.novaaiops.com / cross-signal-correlation · graph
Suspect Ranking

Three candidates, scored, explained

The graph is too dense to read directly. Nova surfaces the top three suspect nodes, each with a confidence score and a one-sentence explanation: "this deploy preceded the regression by 4 minutes, on the same service, with no other suspect changes in the window." If the top suspect is wrong, you have the next two ready, and a way to drill into the full graph.

  • Top 3, always: three suspect nodes ranked by confidence, there is always a backup if the first is wrong
  • Confidence + explanation: each suspect ships a 0–100 score and a sentence explaining why we think it is the cause
  • Click for the full graph: never trust a black box, every suspect has a "show evidence" link to the full graph view
app.novaaiops.com / cross-signal-correlation · ranking
Cross-Tenant Privacy

The graph never leaks across tenants

Multi-tenant deployments must not let one tenant's correlation graph see another tenant's signals. Nova tags every signal with org_id and the correlation engine filters by tenant before it builds the graph. The graph builder is unit-tested against tenant-confusion attacks weekly. Audit-mode runs cross-tenant probes and reports zero leakage.

  • org_id everywhere: every signal carries org_id; the graph builder rejects any candidate edge that crosses tenants
  • Weekly cross-tenant probe: Safety Auditor runs synthetic cross-tenant signals and confirms 0 graph contamination
  • Platform-admin visible only: cross-tenant graphs exist for support purposes but are visible only to founder-level platform-admin
app.novaaiops.com / cross-signal-correlation · privacy
Use During & After

Live during incidents, archived for postmortems

During an incident, the graph is live: nodes appear as new signals arrive, edges form as the correlation engine confirms relationships. After the incident, the graph is frozen and archived. The postmortem builder consumes the frozen graph directly so the timeline you write up is the timeline that actually happened, not a reconstruction.

  • Live mode: graph updates every 5s during an active incident; new nodes shimmer in green, edges in white
  • Frozen archive: when the incident closes, the graph is hashed, signed, and immutable, perfect for postmortems
  • Timeline export: the postmortem builder pulls the timeline straight from the graph, no copy-paste from logs
app.novaaiops.com / cross-signal-correlation · timeline
Video walkthrough coming soon

Subscribe to Nova AI Ops on YouTube for demos, tutorials, and feature deep-dives.

Stop running five queries to find one cause

Correlation is the work of root-cause analysis. We do the work; you get the answer.

Get Started Request a Demo