Insight IT Solutions PrismGuard

PrismGuard is an open-source firewall designed to safeguard large language model (LLM) applications. It offers robust protection against prompt injection attacks, intercepting malicious inputs before they reach your underlying models. Key features include: rules-first approach for clear decision-making, direct allow/block responses with a resolution_gate indicating the specific rule triggered, and optional local ONNX support for enhanced performance. This ensures transparent auditing and eliminates ambiguity often associated with probability scores. Built for production environments, PrismGuard is self-hosted and highly performant, demonstrating superior cold holdout results compared to other solutions. It can be seamlessly integrated at various points in your application architecture, such as at agent entry, alongside RAG chunks, or as a sidecar. The tool is licensed under Apache-2.0, providing flexibility and community support. PrismGuard is ideal for engineering teams and developers building secure LLM copilots, RAG systems, digital assistants, and vertical agents who require a dependable and auditable defense against prompt manipulation. It offers a straightforward `pip install` process to quickly deploy and protect your applications.