Concepts

The RAIL Score API is built around four core primitives. Understanding these makes the rest of the SDK and API reference much easier to navigate.

Evaluation

How the RAIL Score API analyses content across 8 responsible-AI dimensions. Understand basic vs deep mode, what each dimension measures, and how scores are calculated.

8 dimensionsbasic / deepscore tiers

Safe Regeneration

The evaluate → check → regenerate feedback loop. RAIL scores your content, identifies which dimensions fall below your thresholds, then rewrites the content to fix them — automatically.

feedback loopthresholdsauto-rewrite

Middleware

Drop-in provider wrappers that intercept every LLM response and attach RAIL scores before returning it to your application — no manual eval calls required.

wrappersOpenAI / Gemini / Anthropicauto-scoring

Policy Engine

Declarative rules that translate RAIL scores into actions. Define thresholds per dimension and specify what happens when they fail — block, warn, flag, or allow — at the call level or across a full session.

rulesblock / warn / flagsessions

How They Fit Together

These four concepts form a layered system — each one builds on the previous:

Evaluation is the foundation. It answers: “How responsible is this content?” across 8 dimensions.

Safe Regeneration uses evaluation in a loop. It answers: “How do I automatically fix content that fails?”

Middleware makes evaluation invisible. It answers: “How do I score every response without writing eval calls everywhere?”

Policy Engine enforces rules on evaluation results. It answers: “What should my application do when scores fall below acceptable levels?”

Already understand the concepts?

API Reference →Python SDK →JavaScript SDK →AI Chatbot Tutorial →