Time-series anomaly detection API

Ship anomaly detection without building the detector.

Send timestamps and values. Get expected values, upper and lower bands, anomaly flags, seasonality, and root-cause slices back as JSON. Use it in your dashboard, alerting job, or customer-facing product.

Try the live demo Read the API docs
Watch a metric leave its expected range. The API returns the band, the anomaly flag, and the values you need to alert or explain the change.
isAnomaly: true
80 120 160 isPositiveAnomaly: true expected 124, observed 188 streaming metric detector response
expected bandExpected values plus upper and lower margins.
anomaly flagThe point is marked positive, negative, or normal.
response JSONReady to chart, store, alert on, or explain.
/timeseries/entire/detect

Score the full history

Run detection across an entire series and return the band, flags, and inferred period.

/timeseries/last/detect

Check the latest point

Use the last-point endpoint when a monitor or scheduled job only needs today’s verdict.

/detect/attributed

Ask why it moved

Send additive dimensions and get the slices that contributed most to the change.

OpenAPI

Drop it into your stack

JSON over HTTP, API key auth, per-key limits, and OpenAPI docs.

Try it out.

Pick a dataset, adjust sensitivity, and inspect the response. The demo uses the same fields your application would receive from the API.

Request

Preparing
95

Response

actual expected expected band anomaly

JSON inspector

{}
JSON inspector Inspect the request sent by the demo and the response returned by the detector. 
{}
Product and ops metricsSignups, API errors, orders, CPU, and usage-style series.
Hard casesLevel shifts, seasonal spikes, square waves, low-volume movement, and noisy traffic.
Validation samplesMarketing, ecommerce, search, ads, and benchmark-style examples.

Seasonality is part of the baseline.

Monday traffic is not Saturday traffic. Lunch-hour load is not midnight load. Month-end revenue is not a random spike just because it sits above last Tuesday.

The detector learns the repeating pattern in the series and scores each point against the right expected range. Weekly, monthly, daily, and hourly cycles are handled automatically, so alerts fire on the break, not the rhythm.

Pattern learned before scoring Expected range follows the cadence of the metric.
period: 7
high low week 1 week 2 week 3 week 4
hourlyIntraday load
dailyWorking patterns
weeklyWeekday cycles
monthlyCalendar effects

Simple API pricing, with rollover.

Start free. Paid plans include monthly detections that roll over, with overage from $0.20 per 1,000.

Free

$0/mo

For testing real payloads and validating the response shape before you commit.

  • Included detections10k
  • Card requiredNo
  • OverageNone

Starter

$25/mo

For prototypes, internal monitors, and the first production jobs.

  • Included detections125k
  • RolloverYes
  • Overage$0.20 / 1k
  • Azure comparison36% cheaper

Production

$99/mo

For customer-facing usage, scheduled checks, and higher-volume workflows.

  • Included detections600k
  • RolloverYes
  • Overage$0.16 / 1k
  • Azure comparison49% cheaper

Scale

$299/mo

For large metric fleets, multi-tenant products, and heavier API traffic.

  • Included detections2M
  • RolloverYes
  • Overage$0.12 / 1k
  • Azure comparison62% cheaper

Azure comparison uses the public Anomaly Detector Standard Univariate price of $0.314 per 1,000 transactions as of June 22, 2026. Prices shown in USD; EUR pricing can use the same headline amounts. Replacing another detector? Compare Azure or see all alternatives.

Bring your own time-series.

Send us a metric you care about. We’ll run it through the API, show the response, and help you decide whether batch, last-point, or attributed detection fits the job.

Get your free API key

1,000 calls/month free — no card required.