Shieldstral
Adaptive, local and multimodal AI moderation
Shieldstral is an open-weight multimodal safety model from Mistral designed to evaluate text, images, and combined inputs. Instead of relying on fixed safety categories, it lets developers define policies in natural language at inference time and returns a calibrated safety score. Released under Apache 2.0, its compact architecture can run locally, making it practical for building adaptive and privacy-focused moderation systems.