← Platform
Observability
Every agent action, tool call, decision — auditable.
Debug failures, prove compliance, and evaluate regressions on the same trace pane. No separate audit reconstruction project.
Trace
wf-2847-trace-9f1e
4.8s total
workflow.execute
4800ms
activity.extract_documents
1200ms
tool.pdf_parse
400ms
llm.classify · claude-sonnet-4-6 · 1,284 tok · $0.009
800ms
activity.validate · retry · 1
2000ms
activity.request_approval · pending
—
Structured run evidence Tokens · cost · eval
One run, every span · 4.8s total · includes LLM, tool, and activity spans
What you can see
If it happened in a run, it's in the trace.
Trace the whole run
Nested spans, no gaps
Workflow, activity, tool call, model response, and system update. One trace, every level.
Eval in the loop
Regressions block deploys
Wire a golden test set to every agent. Changes fail CI if evaluation quality falls below threshold. No silent regressions.
Cost + latency
Per run, per agent, per version
Token cost, wall time, retries. Aggregate across policy versions to see what each change did to economics.
Exports to your stack
Your SRE and compliance teams keep their tools.
AI traces
Prompts, tool use, costs, and model responses.
Telemetry export
Send traces to existing APM and SIEM tools.
Operations metrics
Run health alongside agent performance.
Compliance feed
Ship raw spans into audit pipelines.
The observability your auditor wants. The telemetry your engineers need.
See what end-to-end tracing looks like on a real workflow.