HiddenLayer MSFT Agent Framework Guardrails

HiddenLayer MSFT Agent Framework Guardrails provides AI safety middleware for the Azure Agent Framework, enabling real-time detection and mitigation of malicious inputs, prompt injections, and unsafe AI agent behaviors.

Features

• Input Scanning: Analyze user messages for prompt injections and malicious content
• Output Filtering: Monitor and filter AI-generated responses
• Redaction Support: Automatically redact sensitive or unsafe content
• Streaming Compatible: Works with both streaming and non-streaming responses
• Easy Integration: Simple middleware integration with Azure Agent Framework

Installation

pip install hiddenlayer-msft-agent-framework-guardrails

Configuration

Set the following environment variables:

# Required: Your HiddenLayer API credentials
export HIDDENLAYER_API_ID="your-api-id"
export HIDDENLAYER_API_KEY="your-api-key"

# Optional: Project-specific configuration
export HIDDENLAYER_PROJECT_ID="your-project-id"
export HIDDENLAYER_REQUESTER_ID="your-app-name"

Usage

Basic Example

import asyncio
from agent_framework.openai import OpenAIChatClient
from hiddenlayer_msft_agent_framework_guardrails.middleware import (
    HiddenlayerChatMiddleware,
    HiddenLayerParams,
)

# Configure HiddenLayer parameters
params = HiddenLayerParams(model="gpt-4o-mini")


async def main():
    # Create agent with HiddenLayer guardrails
    agent = OpenAIChatClient().as_agent(
        name="MyAgent",
        model="gpt-4o-mini",
        instructions="You are a helpful assistant.",
        middleware=[
            HiddenlayerChatMiddleware(params),
        ],
    )

    # Run query
    query = "Hello!"
    print(f"User: {query}")
    result = await agent.run(query)
    print(f"Result: {result}")


if __name__ == "__main__":
    asyncio.run(main())

Advanced Examples

Streaming Responses

async def streaming_example():
    agent = OpenAIChatClient().as_agent(
        name="MyAgent",
        instructions="You are a helpful assistant.",
        middleware=[
            HiddenlayerChatMiddleware(params),
        ],
    )

    query = "Hello!"
    print(f"User: {query}")
    print("Agent: ", end="", flush=True)

    async for chunk in agent.run(query, stream=True):
        if chunk.text:
            print(chunk.text, end="", flush=True)
    print()

Capability Matrix

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Support

For issues, questions, or contributions:

• GitHub Issues: Report a bug or request a feature
• Email: sdks@hiddenlayer.com

Related Projects

• Azure Agent Framework
• HiddenLayer SDK