JevShield is an open-source security firewall that detects prompt injections, jailbreaks, data exfiltration, and malicious tool calls in LLM applications. It uses Ollama's Jev-style decision models to return structured security assessments, which a deterministic Python policy engine evaluates to allow, review, or block requests.