Skip to main content

Observability

The Observability section (formerly called Tracing) is the visibility layer for everything that runs on SimplAI. It answers: what happened, how long it took, what it cost, and who did what.

This section covers:

  • Live tracing — watch runs as they happen, with tree-view progress over in-progress steps and status filters (Completed / Failed / In-Progress).
  • Run history — the historical run list per application, with status tabs, version + user filters, and CSV download.
  • Analytics (KPI cards + Metric Details) — KPI cards on the Run History page that open a tabbed Metric Details page with per-metric deep-dive charts (runs, reliability, latency, cost, tokens, steps, tool calls, tool-call latency, context utilization).
  • Audit logs — account-level CRUD log with resource type, actor, side-by-side JSON diff, filters, and noise-reduction grouping for draft edits.
  • Usages & Insights — the workspace-level Observability Suite that aggregates volume, quality, and cost signals across the platform.

How to think about Observability vs Evaluation​

  • Observability answers: what did happen? (descriptive — runs, traces, latencies, costs).
  • Evaluation answers: was it good? (judgmental — scores against criteria).

Most workflows pair them: use Observability to filter to a population of runs (e.g., "all failed runs last 24h"), then send that population to Evaluation to score quality. Observability also feeds evaluation directly for scheduled evals — see Scheduled evaluations.

Voice Agents​

Voice agents have additional observability concerns — call recordings synced to the trace, voice-specific spans (STT, TTS, queue wait), and TTFS-focused latency metrics. See Voice Agent run history and Voice latency for voice-specific views; the rest of this section applies equally to voice agents.

What's been renamed​

The section was previously called Tracing. Anywhere you see "Tracing" in legacy material, read "Observability" — it's the same surface.