Open Source

Official SDKs, integrations, datasets, and research from Responsible AI Labs. Everything below is public, MIT-licensed where applicable, and production-tested against the RAIL Score API.

SDKs

Official client libraries for the RAIL Score API.

Python SDK

PyPI versionPyPI downloads

Python 3.8+·License MIT

Sync and async clients, compliance checks, safe-regeneration, policy engine, multi-turn sessions, middleware, agent evaluation (v2.4), and full OpenTelemetry observability. Provider extras for OpenAI, Anthropic, Google Generative AI, and LiteLLM.

pip install rail-score-sdk

Quick start

from rail_score import RailScore

client = RailScore(api_key=RAIL_API_KEY)
result = client.eval(
    content="Our AI system prioritizes user privacy and data security.",
    mode="deep",
)
print(result.scores)

JavaScript / TypeScript SDK

npm versionnpm downloads

TypeScript 5.2, Node 16+·License MIT

Type-safe API coverage for eval, safe-regenerate (server + external mode), compliance, and health. Built-in provider wrappers (OpenAI, Anthropic, Google). Policy engine, session tracking, Langfuse integration, OTEL telemetry. Agent evaluation in v2.4.

npm install @responsible-ai-labs/rail-score

Quick start

import { RailScore } from "@responsible-ai-labs/rail-score";

const client = new RailScore({ apiKey: process.env.RAIL_API_KEY! });
const result = await client.eval({
  content: "Our AI system prioritizes user privacy and data security.",
});
console.log(result.scores);

Integrations

Drop-in modules for common content, data, and workflow platforms.

Drupal module

drupal.org

Drupal 9/10/11 module for automated RAIL evaluation on entity presave, compliance checking across 6 frameworks, audit logging, incident tracking, and a persistent human review queue. Optional integration with the Drupal AI module scores every AI-generated response automatically.

composer require drupal/rail_score

Research & datasets

Peer-reviewed research and public datasets that underlie the RAIL framework.

RAIL paper on arXiv

RAIL paper on arXiv

The peer-reviewed paper describing the RAIL framework and evaluation methodology.

Open →

Datasets on HuggingFace

Datasets on HuggingFace

Public datasets and model artifacts, including RAIL-HH-10K (multi-dimensional safety dataset) and supporting materials for research replication.

Open →

Contribute on GitHub

Issues, pull requests, and feature requests are all welcome on any of the public repositories under the Responsible AI Labs GitHub organization.

github.com/Responsible-AI-Labs →

Prefer to start with a concrete use case? Get in touch or read the documentation.