Add Option to Configure Guardrail Execution Mode (Sequential vs Parallel)

[Hamza123545]

Confirm this is a feature request for the Python library and not the underlying OpenAI API.

• This is a feature request for the Python library

Describe the feature or improvement you're requesting

Summary:

Currently, the OpenAI Agents SDK runs guardrails and agent logic in parallel, prioritizing low latency. While this is a smart design for speed, it can cause token waste if the guardrail raises an error after the agent has already started running.

---

Problem:

The documentation says:

“If the guardrail detects malicious usage, it can immediately raise an error, which stops the expensive model from running and saves you time/money.”

However, in practice:

• Guardrails and agent run in parallel
• Tokens are already consumed before the guardrail finishes
• So it does not actually save cost, and the doc feels misleading

---

Suggestion:

Please add a config option like this:

agent = Agent(
    guardrail_mode="parallel"  # Default (current behavior)
    # or
    guardrail_mode="sequential"  # Run guardrail first, then agent
)



This would let developers choose between:

Speed-first use cases (chatbots etc.)

Safety and cost-first use cases (finance, education, critical apps)

Why It Matters:
Flexibility is power. Different use cases need different trade-offs.
Giving this small config would make the SDK much more powerful and customizable.

Demo Video:
🎥 (https://drive.google.com/file/d/1JQPPz1hta7usMcxXA78GnqioFw09zc6B/view?usp=sharing)

Final Thought:
This is not a bug. It's a smart default.
But giving developers the choice will make it a great SDK for everyone.

Thank you for this amazing tool!

### Additional context

_No response_

[RobertCraigie]

This is not the agents repository, please report this there. https://github.com/openai/openai-agents-python