OpenAI Agents SDK adapter

Debug OpenAI agent runs without changing your workflow.

AgentDbg ships an optional tracing processor integration for OpenAI Agents SDK so generation spans, function calls, and handoffs map directly into your local run timeline.

Integration docs Install AgentDbg
pip install agentdbg[openai]

Generation spans

Mapped to LLM_CALL with model, prompt, response, and usage fields where available.

Function calls

Mapped to TOOL_CALL with function name, arguments, status, and error context.

Handoffs

Captured as tool-like events with metadata, preserving step order in the run timeline.

Why this integration works well for debugging

  • No new event taxonomy required.
  • Records only within an active AgentDbg run.
  • Keeps local-first storage and redaction behavior.
  • Lets guardrails stop expensive runaway behavior.

Common debugging wins

  • Spot repeated tool-call loops quickly.
  • Inspect malformed function arguments.
  • Find first error point in multi-step handoff chains.
  • Correlate guardrail aborts with preceding span sequence.
AgentDbg timeline UI showing detailed run events

Keep your OpenAI workflow. Add timeline-level evidence.

Use AgentDbg to reduce guesswork and contain runaway behavior while iterating.

Install AgentDbg All integrations