Open-Source Firewall and Guardrails for AI Models

OpenShield’s AI Firewall is a transparent proxy built to protect AI model endpoints the same way modern application security tools protect APIs: by placing a control point in front of the model and enforcing security policy on every request. In a model-protection program, it functions as a front-door shield for LLMs — reducing abuse, constraining risky inputs, and limiting harmful or policy-violating outputs before they ever reach users or downstream systems.

OpenShield focuses on practical, first-line protections that map well to common LLM threat paths. It provides rate limiting (to blunt automated abuse and DoS-style pressure), content filtering and keyword filtering (to reduce prompt-injection patterns, sensitive-data exposure, and prohibited content requests), and tokenizer calculation support (useful for cost/usage controls and guardrails tied to token limits).