Shieldstral

Shieldstral is a 3B open-weight multimodal guardrail from Mistral, engineered to enhance content safety and policy enforcement. Key capabilities include: * Define safety policies using natural language * Evaluate text and images from a single token output * Operate locally on a single 16GB GPU * Perform at inference time for real-time content review This robust system allows developers and platforms to implement content guidelines with precision and efficiency. By processing both textual and visual inputs, Shieldstral provides a comprehensive safety layer for various applications. Its ability to run locally on standard hardware makes it accessible for integration into diverse operational environments without requiring extensive computational resources. Shieldstral is ideal for content moderation teams, platform administrators, and developers building user-generated content applications. It ensures adherence to community standards and regulatory requirements by providing a powerful, adaptable mechanism for content filtering and policy enforcement across multimodal data.