OpenAI Agents SDK for Python guardrails

Overview

This demo is a practical guide that demonstrates:

Three types of AI guardrails: Input validation, output checking, and real-time streaming monitoring
Integration of the OpenAI Agent SDK for Python with Trigger.dev for production AI workflows
Triggering Python scripts from tasks using our Python build extension
Educational examples of implementing guardrails for AI safety and control mechanisms
Real-world scenarios like math tutoring agents with content validation and complexity monitoring

Guardrails are safety mechanisms that run alongside AI agents to validate input, check output, monitor streaming content in real-time, and prevent unwanted or harmful behavior.

GitHub repo

<Card title="View the OpenAI Agent SDK Guardrails repo" icon="GitHub" href="https://github.com/triggerdotdev/examples/tree/main/openai-agent-sdk-guardrails-examples"

Click here to view the full code for this project in our examples repository on GitHub. You can fork it and use it as a starting point for your own project. </Card>

Video

<video controls className="w-full aspect-video" src="https://github.com/user-attachments/assets/9b1e55c7-467d-4aca-8b4a-a018014c0827"

</video>

Relevant code

Trigger.dev Tasks

inputGuardrails.ts - Passes user prompts to Python script and handles InputGuardrailTripwireTriggered exceptions
outputGuardrails.ts - Runs agent generation and catches OutputGuardrailTripwireTriggered exceptions with detailed error info
streamingGuardrails.ts - Executes streaming Python script and parses JSON output containing guardrail metrics

Python Implementations

input-guardrails.py - Agent with @input_guardrail decorator that validates user input before processing (example: math tutor that only responds to math questions)
output-guardrails.py - Agent with @output_guardrail decorator that validates generated responses using a separate guardrail agent
streaming-guardrails.py - Processes ResponseTextDeltaEvent streams with async guardrail checks at configurable intervals (example: stops streaming if language is too complex for a 10-year-old)

Configuration

trigger.config.ts - Uses the Trigger.dev Python extension