Strict / Balanced / Adaptive

Strict

Maximum constraints. The generated prompt instructs the LLM to refuse borderline requests, add disclaimers aggressively, and ask for clarification rather than guessing. Best for high-risk regulated environments.

Balanced

Standard constraints aligned with your policy thresholds. The LLM is instructed to be helpful while respecting boundaries. Recommended for most production applications.

Adaptive

Minimal constraints. The prompt relies on VeldrixAI's runtime enforcement rather than model-side instruction. Best for creative or exploratory use cases where over-refusal is a concern.