openai-agents-python — tool call path: reasoning vs. execution runtime

openai-agents-python — tool call path: reasoning vs. execution runtime A workflow diagram generated by Archify. 01 / Model EX / Plan + approval 03 / Local execution 04 / Runtime Decide Execute Observe Model output · function_call items · Model › Decide Model output function_call items Next turn · function_call_output · Model › Observe Next turn function_call_output classify · tool · handoff · hosted · Plan + approval › Decide classify tool · handoff · hosted approval? · needs_approval · Plan + approval › Decide approval? needs_approval Interruption · ToolApprovalItem · Plan + approval › Execute Interruption ToolApprovalItem on exception · failure_error_function · Plan + approval › Execute on exception failure_error_function invoke · guards · to_thread · Local execution › Execute invoke guards · to_thread out guardrails · then accept · Local execution › Execute out guardrails then accept Sandbox · exec_command · Runtime › Execute Sandbox exec_command MCP server · call_tool · Runtime › Execute MCP server call_tool output item tool calls error text FunctionTool approved MCP tool output pending exception capability tool Legend Agent logic Policy Cloud service External system