Source: Tracing (OpenAI Agents SDK docs)

Built-in tracing in the OpenAI Agents SDK: default spans, disabling, ZDR caveat, custom processors and sensitive-data controls.

Contents

Summary

Tracing is on by default and records each run as a trace with spans for agents, LLM generations, function calls, guardrails and handoffs, exported to OpenAI’s Traces dashboard. Custom processors can add or replace exporters, and many third-party observability tools integrate.

Key claims

  • Tracing is enabled by default; disable with OPENAI_AGENTS_DISABLE_TRACING=1, set_tracing_disabled, or RunConfig.tracing_disabled.
  • Tracing is unavailable to organisations under a Zero Data Retention policy.
  • add_trace_processor() adds exporters; set_trace_processors() replaces the OpenAI default.
  • Integrations listed include Weights & Biases, Arize Phoenix, MLflow, Braintrust, Logfire, AgentOps, LangSmith, Langfuse and Datadog.
  • trace_include_sensitive_data defaults to true, so inputs and outputs are stored unless turned off.

Feeds

Suggest changes →