Use this page when you need to understand what happened during a run and you want that story in spans instead of only in logs.
- run start and end
- model calls
- tool execution
- handoffs
- guardrails
- session persistence
- speech and transcription flows
TraceSpanSpanData- tracing processors
- tracing setup helpers such as
add_trace_processor,flush_traces, andset_tracing_disabled
- attach a tracing processor in application startup
- add trace metadata on high-value workflows
- flush on shutdown or in tests
Tracing keeps causality intact across:
- nested tools
- handoffs
- streamed runs
- voice pipelines
- realtime sessions