Alerts that fire when
something is actually wrong.
ML-powered anomaly detection learns your baseline, deduplicates related alerts into single incidents, and surfaces correlated context automatically — so your team sleeps through the noise and wakes up for the signal.
Error rate — api-gateway
Last 60 minutes · anomaly detected at t+28m
Anomaly detected · 4.2× above baseline
Error rate jumped from 0.38% → 5.1% at 14:28. Correlated spans show DB connection pool exhausted.
Active rules
API error rate anomaly
FIRINGerror_rate(service=api-gateway)
> 3× baseline over 5mp99 latency SLA breach
PENDINGp99(latency_ms, service=checkout)
> 300ms over 3mDB heartbeat absent
OKabsent(service=db-primary
message='ping') over 2mFrom alert fatigue to actionable signal.
Median results across obseria.io customers after 30 days of ML baseline training.
Pages / month
900
12
Mean time to detect
18 min
2 min
False positive rate
74%
4%
p99 latency — checkout service
Alert fires before SLA breach — not after
Everything alerting should be.
Built for teams who are done tuning thresholds at 3 AM.
ML baseline detection
obseria.io learns your service's normal behaviour over 7 days and alerts only on genuine anomalies — not noise from regular traffic patterns.
Alert deduplication
Related alerts are grouped into a single incident automatically. No more 900 pages for a single upstream outage.
Automatic correlation
When an alert fires, obseria.io surfaces correlated metrics, traces, and logs so you have context before you even open the terminal.
On-call scheduling
Built-in rotation management with overrides, escalation chains, and follow-the-sun support. No third-party on-call tool required.
Alert preview & backtesting
Before saving a rule, see how it would have fired over the last 7 days. Catch noisy rules before they wake up your team.
Maintenance windows
Suppress alerts during planned deployments or maintenance. One-off or recurring — configure from the UI or API.
Notification channels
Your tools, your way.
Route alerts to any combination of channels based on severity, service, team, or time of day. One rule can notify Slack for warnings and PagerDuty for criticals.
- Severity-based routing (info / warn / critical)
- Team routing by service ownership
- Time-of-day routing (business hours vs. on-call)
- Escalation if unacknowledged after N minutes
Slack
Post to any channel or DM
PagerDuty
Create incidents automatically
OpsGenie
Route by team & schedule
Rich HTML digests
Webhook
POST to any endpoint
SMS / Call
Twilio or your carrier
❝
We went from 900 pages a month to 12. The ML baseline detection is the single best ROI we've had from any tool this year.
PMPriya Mehta
SRE Lead, FinTech startup · 60-person eng team
❝
The deduplication alone saved us. During a cascading DB failure we got one incident, not four hundred Slack messages.
LELars Eriksson
Platform Engineer, e-commerce scale-up
Common questions
How long does ML baseline training take?
obseria.io starts collecting baseline data immediately on rule creation. Anomaly detection becomes active after 24 hours of data, with full accuracy after 7 days. Rules fire statically (threshold-based) during the warm-up period.
Can I alert on logs as well as metrics?
Yes. obseria.io supports metric alerts, log-pattern alerts, trace-error-rate alerts, and composite alerts that combine conditions across signal types. All use the same rule editor and notification channels.
How does deduplication work?
Alerts with the same root cause (correlated service, time window, and error type) are automatically grouped into a single incident. You receive one notification, one timeline, and one place to coordinate the response.
What happens if obseria.io itself is down?
Our alerting pipeline runs in a separate, isolated tier from the ingest and query services. Alerts are evaluated independently and we maintain a 99.95% uptime SLA for the alerting tier specifically.
Sleep through the noise.
Wake up for the signal.
obseria.io Smart Alerting is included in all plans. Set up your first ML-powered alert in under 2 minutes — no threshold configuration needed.