Guardrails AI Tracing

Instrument LLM applications using the Guardrails AI framework

Google Colabcolab.research.google.com

Setup Guardrails AI with a DatasetEmbeddingsGuard

This guide helps you set up instrumentation for your Guardrails AI application using OpenInference to send traces to Arize AX.

API Key Setup

Before running your application, ensure you have the following environment variables set:

export ARIZE_SPACE_ID="YOUR_ARIZE_SPACE_ID"
export ARIZE_API_KEY="YOUR_ARIZE_API_KEY"
export OPENAI_API_KEY="YOUR_OPENAI_API_KEY" # Needed for the example code
# Add other LLM provider API keys if used by Guardrails

Sign up for a free Arize account. You can find your Arize Space ID and API Key in your Arize account settings.

Install

Install the necessary packages for Guardrails AI, the OpenInference instrumentor, Arize OTel, and OpenTelemetry:

pip install guardrails-ai openinference-instrumentation-guardrails arize-otel

Setup Tracing

Connect to Arize AX using arize.otel.register and apply the GuardrailsInstrumentor.

import os
from arize.otel import register
from openinference.instrumentation.guardrails import GuardrailsInstrumentor

# Setup OTel via Arize's convenience function
tracer_provider = register(
    space_id=os.getenv("ARIZE_SPACE_ID"),   
    api_key=os.getenv("ARIZE_API_KEY"),     
    project_name="my-guardrails-app"  
)

GuardrailsInstrumentor().instrument(tracer_provider=tracer_provider)

Run Guardrails Example

Now you can run your Guardrails AI code as usual. The instrumentor will capture relevant trace data.

# Ensure OPENAI_API_KEY is set in your environment for this example
import openai
from guardrails import Guard
from guardrails.hub import TwoWords # Example validator

# Initialize Guardrails
guard = Guard().use(
    TwoWords() # Using a simple validator from the hub
)

# Make a call through Guardrails
# This will also make an underlying call to the OpenAI LLM
response = guard(
    llm_api=openai.chat.completions.create,
    prompt="What is another name for America?",
    model="gpt-3.5-turbo",
    max_tokens=1024,
    # You can add instructions or other parameters as needed by your spec
)

# Print the validated (or processed) response
if response.validation_passed:
    print(f"Validated Output: {response.validated_output}")
else:
    print(f"Validation Failed. Raw LLM Output: {response.raw_llm_output}")
    print(f"Validation Details: {response.validation_details}")

Observe in Arize AX

After running your Guardrails AI application, traces will be sent to your Arize project. You can log in to Arize AX to:

• Visualize the execution flow of your Guardrails, including the steps and validators.

• Inspect the inputs and outputs of the underlying LLM calls.

• Analyze latency and errors.

• Evaluate the effectiveness of your guards.

Resources

• Guardrails AI Documentation

• OpenInference Guardrails Instrumentor

• Example Colab Notebook (Focuses on a specific guard type but shows instrumentation)