Tracing & Observability
See every step,
token and tool call.
Every run of every agent and workflow is traced end to end: model calls, retrieved context, tool inputs and outputs, latency and cost. Tied to the exact version that produced it.
Status
Succeeded
Duration
3.84 s
Tokens
4,812
Cost
$0.014
- run · support-triage v1.43840 ms
- retrieve · help-center420 ms
- llm · classify710 ms
- tool · zendesk.get_ticket260 ms
- tool · shopify.get_order540 ms
- llm · draft_reply1560 ms
- guardrail · pii_redact120 ms
- tool · zendesk.create_draft280 ms
Nothing is a black box.
What you get with Tracing & Observability.
Span waterfall
A timeline of every step with timing, so slow tools and long generations are obvious.
Full payloads
Inspect prompts, completions, retrieved chunks and tool I/O, with PII redaction you control.
Cost per run
Token usage and spend per step, per version, per user and per workflow.
Version-aware
Filter and compare traces by version to see exactly what a release changed.
Alerts
Get notified in Slack, email or PagerDuty when error rates, latency or cost cross a threshold.
Export
Stream traces to your own warehouse or observability stack over OpenTelemetry.
Questions
How long are traces kept?+
Retention depends on your plan. Enterprise customers can configure custom retention and export.
Can I turn off payload capture?+
Yes, per agent or per workflow. Metadata and timing are still captured.
Start building today.
From $5 per user per month. Bring your own models. Connect everything in one click.