Shieldstral is an advanced, 3B parameter open-weight multimodal guardrail developed by Mistral AI. Designed to redefine how applications handle runtime security, it allows developers to define custom safety policies using natural language directly at inference time.
As a versatile multimodal tool, Shieldstral can seamlessly evaluate text, images, or a combination of both from a single token output. Built for efficiency and local deployment, it is capable of running smoothly on a single 16GB GPU, making it an ideal open-source infrastructure solution for teams prioritizing data security, AI alignment, and compliance.
Shieldstral is a 3B open-weight multimodal guardrail developed by Mistral AI that evaluates text and images for safety at runtime.
Safety policies are defined dynamically in natural language directly at inference time.
Shieldstral is optimized to run locally on a single 16GB GPU.
0 community reviews for Shieldstral
Help the community by sharing your experience. Your review helps others make better decisions.
