Signal detection
Streaming models learn each metric’s normal rhythm and flag the break, not the noise. Seasonality and launches included.
Now with streaming anomaly detection
Every metric your team argues about, settled in one place
Nova reads your warehouse, watches every pipeline and tells you what changed before the dashboard does. No cron jobs, no guesswork, no 2am pages.
- Faster root cause
- 0x
- Alerts that matter
- 0%
- Rows scanned daily
- 0B
SnowflakeBigQueryDatabricksdbtKafkaPostgresLookerSegment
About us
We built the tool our own data team kept asking for
Nova started inside a fintech data platform, where seven analysts spent their mornings reconciling numbers that should have agreed. The dashboards were fine. The pipelines were fine. Nobody could say which one had moved.
So we wrote a service that watched the warehouse itself - every table, every freshness window, every silent schema change - and explained the delta in plain language. Four years later that service is Nova.
- Founded by data engineers, not dashboard vendors
- SOC 2 Type II, audited annually
- Deploys in your own VPC in under an hour
The platform
Four surfaces, one source of truth
Each piece works on its own. Together they close the loop from a raw table to the decision someone makes on Monday.
Lineage graph
Trace any number back through every join, model and source. Click a node to see exactly which commit changed it.
Narrative reports
Nova writes the weekly summary your stakeholders actually read, with the caveats an analyst would have added.
Warehouse-native
Runs where your data already lives. Nothing is copied out, nothing is cached in someone else’s cloud.
Why Nova
Built for the team that gets paged
Four decisions we made early that you will feel on your first bad morning.
01
Answers in seconds
Incident context loads in under two seconds against a 30-billion-row warehouse, because detection runs beside the data.
02
Your perimeter, your keys
Self-hosted or in-VPC by default. Nova never needs egress to read a row, and every query is logged for audit.
03
Fewer, better alerts
Teams that switch to Nova cut alert volume by 92% in the first month, without losing a single real incident.
04
Adopted, not installed
A guided first week walks each analyst through their own tables. Median time to first caught incident: four days.
FAQ
The questions procurement always asks
No. Detection runs as scheduled queries inside your own warehouse, and only aggregate statistics leave it. In self-hosted mode nothing leaves at all.
Most teams connect a warehouse in about forty minutes and see their first real anomaly inside a week. We do not gate the trial on a sales call.
They stay exactly where they are. Nova annotates them - it is the layer that explains why a chart moved, not a replacement for the chart.
Yes. Every detector is a Python class with a documented interface, so you can register your own alongside the built-in ones.
Per monitored table, billed monthly, with a flat cap once you pass a thousand. No per-seat pricing - analysts should not have to ration access.
Contact
Talk to an engineer, not a rep
Tell us what broke last quarter and we will show you what Nova would have caught. Replies come from the people who build it, usually the same day.
- hello@nova.example
- Sales
- +1 (415) 555-0134
- Office
- 312 Bryant St, San Francisco