Streaming

Streaming support helps teams evaluate traces as they arrive instead of waiting for a single large batch.

When streaming is useful

Use streaming workflows when you want to:

  • monitor agent quality continuously
  • reduce delay between trace ingestion and evaluation
  • power operational dashboards or alerts
  • support long-running systems with ongoing traffic

Tradeoffs

Compared with pure batch processing, streaming workflows usually require more care around:

  • ordering and completeness of traces
  • idempotency
  • partial results
  • operational observability

Good use cases

  • production monitoring pipelines
  • continuous staging environments
  • shared internal evaluation services
  • Kubernetes-based deployments where evaluation runs as a long-lived service