How to beginner · 3 min read

How to create validators with Guardrails AI

Q: How to create validators with Guardrails AI

Use Guardrails AI to create validators by defining rules in a railspec YAML file or inline Python decorators that specify constraints on model outputs. These validators enforce output formats, types, and safety checks automatically during generation.

Quick answer

Use Guardrails AI to create validators by defining rules in a railspec YAML file or inline Python decorators that specify constraints on model outputs. These validators enforce output formats, types, and safety checks automatically during generation.

PREREQUISITES

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

Setup

Install the guardrails-ai package and set your OpenAI API key as an environment variable.

bash

pip install guardrails-ai

Step by step

Create a railspec YAML file defining your validators, then load it in Python to validate LLM outputs automatically.

python

import os
from guardrails import Guard

# Define a railspec YAML string with validators
rail_spec = """
- name: user_info
  type: object
  fields:
    name:
      type: string
      description: User's full name
      required: true
    age:
      type: integer
      description: User's age
      required: true
      minimum: 0
"""

# Save railspec to a file
with open("user_info_rail.yaml", "w") as f:
    f.write(rail_spec)

# Initialize Guard with the railspec file
guard = Guard.from_path("user_info_rail.yaml")

# Example LLM output to validate
llm_output = '{"name": "Alice", "age": 30}'

# Validate output
validated_output = guard.validate(llm_output)
print(validated_output)

output

{"name": "Alice", "age": 30}

Common variations

You can create validators inline using Python decorators or extend validation with custom functions. Guardrails supports async usage and integration with OpenAI or other LLMs.

python

from guardrails import guard

@guard(
    {
        "name": "string",
        "age": {"type": "integer", "minimum": 0}
    }
)
def validate_user_info(output: str):
    # This function will automatically validate the output
    return output

# Example usage
result = validate_user_info('{"name": "Bob", "age": 25}')
print(result)

output

{"name": "Bob", "age": 25}

Troubleshooting

If validation fails with ValidationError, check your railspec syntax and ensure the output matches the expected format.
Use guard.validate() exceptions to debug invalid fields.
Ensure your LLM output is valid JSON or matches the schema type.

✅

Key Takeaways

Define validators in a railspec YAML or inline Python to enforce output constraints.
Guardrails AI automatically validates and sanitizes LLM outputs to prevent errors.
Use guard.validate() to check outputs and catch validation errors early.
Validators support types, ranges, required fields, and custom logic.
Guardrails integrates seamlessly with OpenAI and other LLMs for safe generation.

Verified 2026-04 · gpt-4o, gpt-4o-mini

Verify ↗