Open-source multi-layered prompt-injection detector combining heuristics, vector similarity, LLM checks, and canary tokens to protect apps from adversarial prompts.