| Feature | Description |
|---|---|
| 🎯 8 RAIL Dimensions | Evaluate content across Reliability, Accountability, Interpretability, Legal Compliance, Safety, Privacy, Transparency, and Fairness |
| ⚡ Multiple Evaluation Tiers | Choose from basic, dimension-specific, custom, weighted, detailed, advanced, and batch evaluation |
| 🤖 AI Generation | Generate RAG-grounded responses, reprompt suggestions, and protected content |
| ✅ Compliance Checks | Built-in support for GDPR, HIPAA, CCPA, and EU AI Act compliance |
| 📊 Batch Processing | Evaluate up to 100 items per request efficiently |
| 🔒 Type-Safe | Full typing support with structured dataclasses for better IDE experience |
| 🔄 Auto-Retry | Built-in error handling and automatic retries |
| 📈 Usage Tracking | Monitor credits, usage history, and API health |
pip install rail-scorefromrail_scoreimportRailScore# Initialize clientclient=RailScore(api_key="your-rail-api-key")
# Evaluate contentresult=client.evaluation.basic("Our AI system ensures user privacy and data security.")
# Access scoresprint(f"Overall RAIL Score: {result.rail_score.score}")
print(f"Confidence: {result.rail_score.confidence}")
print(f"Privacy Score: {result.scores['privacy'].score}")
print(f"Credits Used: {result.metadata.credits_consumed}")- Installation
- Configuration
- Evaluation API
- Generation API
- Compliance API
- Utilities
- Error Handling
- Response Structure
- Use Cases
- Development
- Contributing
- License
- Support
fromrail_scoreimportRailScoreclient=RailScore(
api_key="your-rail-api-key",
base_url="https://api.responsibleailabs.ai", # Optionaltimeout=60# Request timeout in seconds
)Getting an API Key: Visit responsibleailabs.ai to sign up and get your API key.
Evaluate content across all 8 RAIL dimensions:
result=client.evaluation.basic(
content="Your AI-generated content here",
weights=None# Optional custom weights
)
# Access resultsprint(result.rail_score.score) # Overall score (0-10)print(result.rail_score.confidence) # Confidence (0-1)# Individual dimensionsfordim_name, dim_scoreinresult.scores.items():
print(f"{dim_name}: {dim_score.score} (confidence: {dim_score.confidence})")
print(f" Explanation: {dim_score.explanation}")
ifdim_score.issues:
print(f" Issues: {', '.join(dim_score.issues)}")
# Metadataprint(f"Request ID: {result.metadata.req_id}")
print(f"Credits Used: {result.metadata.credits_consumed}")
print(f"Processing Time: {result.metadata.processing_time_ms}ms")Evaluate on one specific dimension only:
result=client.evaluation.dimension(
content="We collect user data with consent",
dimension="privacy"# One of: reliability, accountability, interpretability,# legal_compliance, safety, privacy, transparency, fairness
)
print(result['result']['score'])
print(result['result']['explanation'])Evaluate only specific dimensions:
result=client.evaluation.custom(
content="Healthcare AI system",
dimensions=["safety", "privacy", "reliability"],
weights={"safety": 40, "privacy": 35, "reliability": 25}
)
print(result.rail_score.score)
print(result.scores.keys()) # Only evaluated dimensionsCustom dimension weights:
weights= {
"safety": 30,
"privacy": 25,
"reliability": 20,
"accountability": 15,
"transparency": 5,
"fairness": 3,
"inclusivity": 1,
"user_impact": 1
}
result=client.evaluation.weighted("Content here", weights=weights)Get detailed breakdown with strengths and weaknesses:
result=client.evaluation.detailed("AI model description")
summary=result['result']['summary']
print(f"Strengths: {summary['strengths']}")
print(f"Weaknesses: {summary['weaknesses']}")
print(f"Improvements needed: {summary['improvements_needed']}")Ensemble evaluation with higher confidence:
result=client.evaluation.advanced(
content="Critical AI system",
context="Healthcare decision support system"# Optional
)
print(result.rail_score.confidence) # Typically 0.90+Evaluate multiple items in one request:
items= [
{"content": "First AI-generated text"},
{"content": "Second AI-generated text"},
{"content": "Third AI-generated text"}
]
result=client.evaluation.batch(
items=items,
dimensions=["safety", "privacy", "fairness"],
tier="balanced"# "fast", "balanced", or "advanced"
)
print(f"Processed: {result.successful}/{result.total_items}")
foritem_resultinresult.results:
print(f"Score: {item_result.rail_score.score}")Evaluate RAG responses for hallucinations:
result=client.evaluation.rag_evaluate(
query="What is the capital of France?",
response="The capital of France is Paris.",
context_chunks=[
{"content": "Paris is the capital city of France."},
{"content": "France is a country in Western Europe."}
]
)
metrics=result['result']['rag_metrics']
print(f"Hallucination Score: {metrics['hallucination_score']}") # Lower is betterprint(f"Grounding Score: {result['result']['grounding_score']}") # Higher is betterprint(f"Overall Quality: {metrics['overall_quality']}")Generate context-grounded responses:
result=client.generation.rag_chat(
query="What are the benefits of GDPR compliance?",
context="GDPR provides data protection and privacy rights to EU citizens...",
max_tokens=300,
model="gpt-4o-mini"
)
print(result.generated_text)
print(f"Tokens used: {result.usage['total_tokens']}")
print(f"Credits: {result.metadata.credits_consumed}")Get improvement suggestions:
current_scores= {
"transparency": {"score": 4.5},
"accountability": {"score": 5.0}
}
result=client.generation.reprompt(
content="AI makes decisions automatically",
current_scores=current_scores,
target_score=8.0,
focus_dimensions=["transparency", "accountability"]
)
suggestions=result['result']['improvement_suggestions']
print(suggestions['text_replacements'])
print(suggestions['expected_improvements'])Generate content with safety filters:
result=client.generation.protected_generate(
prompt="Write a description for an AI hiring tool",
max_tokens=200,
min_rail_score=8.0
)
print(result.generated_text)
print(f"RAIL Score: {result.rail_score}")
print(f"Safety Passed: {result.safety_passed}")result=client.compliance.gdpr(
content="We collect user emails for marketing purposes",
context={"data_type": "personal", "region": "EU"},
strict_mode=True# Use 7.5 threshold instead of 7.0
)
print(f"Compliance Score: {result.compliance_score}")
print(f"Passed: {result.passed}/{result.requirements_checked}")
forreqinresult.requirements:
print(f"{req.requirement} ({req.article}): {req.status}")
ifreq.status=="FAIL":
print(f" Issue: {req.issue}")# CCPAresult=client.compliance.ccpa("Content here")
# HIPAAresult=client.compliance.hipaa("Healthcare AI system")
# EU AI Actresult=client.compliance.ai_act("AI system description")credits=client.get_credits()
print(f"Plan: {credits['plan']}")
print(f"Monthly Limit: {credits['credits']['monthly_limit']}")
print(f"Used This Month: {credits['credits']['used_this_month']}")
print(f"Remaining: {credits['credits']['remaining']}")usage=client.get_usage(limit=50, from_date="2025-01-01T00:00:00Z")
print(f"Total records: {usage['total_records']}")
print(f"Total credits used: {usage['total_credits_used']}")
forentryinusage['history']:
print(f"{entry['timestamp']}: {entry['endpoint']} - {entry['credits_used']} credits")health=client.health_check()
print(f"Status: {health['ok']}")
print(f"Version: {health['version']}")fromrail_scoreimport (
RailScore,
AuthenticationError,
InsufficientCreditsError,
ValidationError,
RateLimitError,
PlanUpgradeRequired
)
client=RailScore(api_key="your-api-key")
try:
result=client.evaluation.basic("Your content")
exceptAuthenticationError:
print("Invalid API key")
exceptInsufficientCreditsErrorase:
print(f"Not enough credits. Balance: {e.balance}, Required: {e.required}")
exceptValidationErrorase:
print(f"Invalid parameters: {e}")
exceptRateLimitErrorase:
print(f"Rate limit exceeded. Retry after: {e.retry_after} seconds")
exceptPlanUpgradeRequired:
print("This endpoint requires a Pro or higher plan")All endpoints return responses with this structure:
{
"result": {
"rail_score": {"score": 8.7, "confidence": 0.90},
"scores": {
"privacy": {"score": 9.1, "confidence": 0.94, "explanation": "..."},
...
},
"processing_time": 2.5
},
"metadata": {
"req_id": "abc-123",
"tier": "pro",
"queue_wait_time_ms": 1200.0,
"processing_time_ms": 2500.0,
"credits_consumed": 2.0,
"timestamp": "2025-11-03T10:30:00Z"
}
}fromrail_scoreimportRailScoreclient=RailScore(api_key="your-key")
# Check user-generated content for safetyresult=client.evaluation.dimension(
content="User comment here",
dimension="safety"
)
ifresult['result']['score'] <7.0:
print("Content flagged for review")
print(f"Issues: {result['result']['issues']}")# Evaluate multiple pieces of contentitems= [{"content": text} fortextincontent_list]
result=client.evaluation.batch(
items=items[:100], # Max 100 itemsdimensions=["safety", "fairness", "privacy"]
)
# Filter by scoresafe_content= [
items[i]
fori, resinenumerate(result.results)
ifres.rail_score.score>=7.5
]# Check GDPR complianceresult=client.compliance.gdpr(
content="AI system for user profiling",
context={"purpose": "marketing", "data_type": "personal"}
)
ifresult.failed>0:
print("GDPR compliance issues found:")
forreqinresult.requirements:
ifreq.status=="FAIL":
print(f"- {req.requirement}: {req.issue}")- Python 3.8+
- requests >= 2.28.0
# Clone repository
git clone https://github.com/Responsible-AI-Labs/rail-score.git
cd rail-score
# Install in development mode
pip install -e ".[dev]"# Run tests
pytest
# Format code
black rail_score/
# Type checking
mypy rail_score/We welcome contributions! Please see our Contributing Guide for details.
This project is licensed under the MIT License - see the LICENSE file for details.
- Documentation: https://responsibleailabs.ai/docs
- Issues: https://github.com/Responsible-AI-Labs/rail-score/issues
- Email: support@responsibleailabs.ai
- Website: https://responsibleailabs.ai
If you find RAIL Score useful, please consider giving it a star! ⭐
Made with ❤️ by Responsible AI Labs
Website • Documentation • GitHub • Twitter