Portal is an AI security solution designed to protect Large Language Model (LLM) and Small Language Model (SLM) integrations by functioning as a guardrail between the language model and the user interface. It intercepts and mitigates risks to generative AI applications in real-time, specifically targeting prompt injections, jailbreaks, and data leakage. The system scans model inputs for adversarial content and Personally Identifiable Information (PII), while simultaneously evaluating outputs to prevent biased, toxic, or inappropriate statements.
The application integrates into existing AI workflows via an API, providing automated prompt and response filtering. It includes privacy-preserving mechanisms to de-identify sensitive data before it reaches the language model and re-identify it upon return, maintaining full data context. The solution is entirely model-agnostic, allowing it to be deployed alongside any generative AI model or application.
