🔓 Open Source

ai+me maintains an open source library that implements a contextual firewall for LLM-based applications using the LLM-as-a-judge concept. This library helps safeguard your AI systems from unintended prompts such as jailbreaking attempts, malicious inputs, and other security threats.

🛡️ AIandMe FirewallOSS

The AIandMe FirewallOSS (opens in a new tab) is an open-source library that acts as a middleware layer to contextually filter and validate user prompts. It ensures that AI agents adhere to their intended business scope and operational boundaries through a reflection approach where an LLM acts as a judge.

Key Features

  • Scope Validation: Ensures user prompts align with the AI agent's defined business scope
  • Intent Filtering: Allows only prompts that match predefined permitted intents
  • Restricted Action Blocking: Blocks prompts that attempt to trigger restricted actions
  • Two-Step Assessment: Initial quick decision followed by detailed explanation for efficiency
  • Multiple LLM Provider Support: Works with OpenAI and Azure OpenAI (with more providers coming soon)

How It Works

The library operates in two asynchronous steps for low latency:

  1. Initial Assessment: Quickly delivers a decision regarding the three categories: off_topic, violation, or restriction
  2. Explanation: Provides a detailed explanation of the verdict using streaming capabilities

The library does not function as a proxy - instead, it analyzes user prompts and provides flags indicating potential issues. It's the responsibility of the LLM application developer to determine how to handle flagged prompts based on their specific requirements.

🚀 Getting Started

Install the library using pip:

pip install aiandme

The library supports both self-hosting with your own LLM providers and integration with the ai+me platform for enhanced features and logging capabilities.

📄 License

The AIandMe FirewallOSS library is released under the MIT license, making it free for both personal and commercial use.

🤝 Community

Join the ai+me community to contribute, get support, and stay updated:

The open source library represents our commitment to transparency and community-driven development in AI security.

👥 Community & Support