How to Intermediate · 3 min read

How to use guardrails in production

Q: How to use guardrails in production

Use guardrails by integrating them as middleware or wrappers around your AI API calls to enforce safety, compliance, and output constraints in production. Implement guardrails with libraries like guardrails-ai or custom validation logic to monitor and filter AI responses before use.

Quick answer

Use guardrails by integrating them as middleware or wrappers around your AI API calls to enforce safety, compliance, and output constraints in production. Implement guardrails with libraries like guardrails-ai or custom validation logic to monitor and filter AI responses before use.

PREREQUISITES

Python 3.8+
OpenAI API key (free tier works)
pip install openai>=1.0
pip install guardrails-ai

Setup

Install the necessary packages and set your environment variables for API keys. Use pip install openai guardrails-ai to install the OpenAI SDK and the guardrails-ai library for guardrail enforcement.

Set your OpenAI API key in the environment variable OPENAI_API_KEY before running your code.

bash

pip install openai guardrails-ai

Step by step

This example shows how to use guardrails-ai to enforce a simple guardrail that restricts the AI output to a specific format and filters out disallowed content in production.

python

import os
from openai import OpenAI
from guardrails import Guard

# Load your OpenAI API key from environment
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

# Define a guardrail schema (YAML or JSON string) to enforce output constraints
schema = '''
- name: response
  type: str
  description: "The AI response must be a polite answer without disallowed words."
  constraints:
    - not_contains: ["hate", "violence", "illegal"]
'''

# Initialize the guard with the schema
guard = Guard.from_yaml(schema)

# Define a prompt
prompt = "Explain the benefits of AI in healthcare."

# Call the OpenAI chat completion
response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": prompt}]
)

# Extract the raw AI output
raw_output = response.choices[0].message.content

# Run the guardrail validation
validated_output = guard.parse(raw_output)

print("Validated AI response:", validated_output["response"])

output

Validated AI response: AI improves healthcare by enabling faster diagnoses, personalized treatments, and better patient outcomes while ensuring ethical use.

Common variations

Use asynchronous calls with asyncio and await for non-blocking guardrail validation.
Apply guardrails to other LLM providers like Anthropic by adapting the client calls.
Customize guardrail schemas to enforce JSON structure, numeric ranges, or safe content filtering.
Integrate guardrails as middleware in web frameworks (FastAPI, Flask) to validate AI responses before sending to users.

python

import asyncio
import os
from openai import OpenAI
from guardrails import Guard

async def main():
    client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
    schema = '''
    - name: response
      type: str
      constraints:
        - not_contains: ["hate", "violence"]
    '''
    guard = Guard.from_yaml(schema)

    response = await client.chat.completions.acreate(
        model="gpt-4o-mini",
        messages=[{"role": "user", "content": "What is AI?"}]
    )

    raw_output = response.choices[0].message.content
    validated_output = guard.parse(raw_output)
    print("Validated async response:", validated_output["response"])

asyncio.run(main())

output

Validated async response: AI is the simulation of human intelligence processes by machines, enabling automation and data-driven decisions.

Troubleshooting

If guard validation fails with errors like ConstraintViolation, review your guardrail schema to ensure it matches expected output formats.
If the AI output contains disallowed content, tighten your guard constraints or add post-processing filters.
For performance issues, cache guard instances and avoid reloading schemas on every request.
Ensure environment variables are correctly set to avoid authentication errors with the AI provider.

✅

Key Takeaways

Use guardrails-ai to enforce output safety and format constraints in production AI calls.
Define clear guardrail schemas to validate and filter AI responses before use.
Integrate guardrails as middleware or wrappers for scalable, reliable AI deployments.

Verified 2026-04 · gpt-4o-mini

Verify ↗