AI Insights
Structured insights over real observability evidence. The model never has database access and never remediates production for you.
Pipeline#
Every insight follows the same path:
- Telemetry — stored signals only
- Evidence — deterministic, bounded retrieval
- Privacy filter — scrub before any provider
- Provider — LLM when configured, else evidence-compose
- Structured insight — validated schema
- Validation — reject malformed / dangerous output
- UI — labeled sections on the incident
Epistemic labels#
- Observed — evidence-backed facts
- Hypotheses — likely causes with confidence band
- Actions — recommended investigation steps only
- Unknowns — missing dimensions called out
Missing user-impact evidence is never shown as zero affected users.
Guides#
Plans#
Free includes 2 and Pro 20 AI analyses per UTC month (ai_insights); Pro can add +20 analysis add-ons. Each analysis has a fixed input, output and cost ceiling. When the allowance is used up, no new analyses run until the month resets.
Limitations
- Not autonomous remediation — insights never restart services, delete data, or change auth for you.
- Without a provider API key, insights use evidence-compose (still structured, still labeled).
- Quality equals the telemetry you ingest.