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