Skip to main content

Settings

The Settings page is reserved for pipeline-level configuration of AI Observability — how long telemetry is kept, how traces are sampled, which embedding service powers analysis, and how data is exported downstream.

These controls are not available yet. The page currently shows a placeholder marking them as planned, and there is nothing on it to configure.

Planned controls​

The placeholder names four areas, and they remain the intended scope:

  • Data retention — How long raw traces, spans and evaluator scores are kept.
  • Trace sampling — Per-agent or per-model sampling rates for high-volume workloads that do not need full-fidelity retention.
  • Embedding configuration — The embedding service that powers Drift Detection.
  • Export controls — Scheduled or ad-hoc export of traces and evaluator scores to a downstream warehouse or object store.

How these behave today​

Until the page is implemented, each area follows a platform default rather than a tenant setting:

  • Retention follows the platform default window. It cannot currently be varied per tenant from this page.
  • Sampling is not applied — every trace is retained.
  • Embedding is configured at the environment level rather than here. Drift Detection becomes available once an embedding endpoint is configured for the environment; if drift analysis is unavailable, this is the reason.
  • Export is available on demand from the Export action on the Traces list. Scheduled export is not yet available.

Note: If you need a retention window or a sampling rate different from the platform default, raise it with your Beak contact — it is an environment-level change today, not a self-service one.

  • Traces — Trace list and viewer, including on-demand export.
  • Drift Detection — Embedding-powered drift analysis.
  • Overview — AI Observability overview.