Vyse

VYSE is a stateful, open-source security middleware for Machine Learning APIs. It operates as an intercepting proxy placed between HTTP clients and ML inference endpoints. VYSE maintains a persistent behavioral profile per session, computes four behavioral signals per request, derives a composite threat score, classifies intent using a zero-shot NLI model, and applies adaptive countermeasures without requiring model retraining, API modification, or proprietary cloud services.