Skip to main content

ML model monitoring

Monitor inference volume, latency, and drift for each deployed ML model.

Before you start
  • Deployed model with API key generated
  • Some metrics appear only after first predict call (UI or API)
  1. Open Machine Learning and click your model.
  2. Generate API key if you have not already (deployment guide).
  3. Scroll to the Usage section on the model detail page.

You will see:

  • Summary cards — total inferences, requests, average response time, endpoints used
  • Charts — daily inference volume and response time, split by Heimdall UI vs API
  • Request log — sortable table with endpoint, inference count, response time, channel, user agent, and drift % when available
  • Filters — 7 / 30 / 90 day windows and endpoint filter

Use the request log to debug integration issues (wrong features, auth errors showing as zero traffic, latency spikes on specific routes).

Account-level monitoring​

Open Usage (/usage) → Data Intelligence for workspace-wide trends:

  • Daily volume stacked by ML, Forecast, Loop, Read, and Vision
  • Channel mix (in-app tests vs production API)
  • Top assets ranked by call volume across all deployed models

See Production monitoring for the full Usage page walkthrough.

What gets recorded​

FieldMeaning
EndpointREST path called (typically predict)
Inference countNumber of predictions in one request
Response timeMilliseconds to complete
ChannelHeimdall UI or API
Drift %Performance drift indicator when enabled
User agentClient string when present

Next steps​