Developer Documentation - AstraLucent Safety API

Quick Start

Install SDK

pip install astralucent

# Or for Node.js
npm install @astralucent/safety-sdk

Get API Key

# Sign up at dashboard.astralucent.com
export ASTRALUCENT_API_KEY="your_key_here"

Replace One Line of Code

❌ Before (Unsafe)

import openai

client = openai.OpenAI()

response = client.chat.completions.create(
    model="gpt-4",
    messages=[
        {"role": "user", "content": prompt}
    ]
)

print(response.choices[0].message.content)

✅ After (Mathematically Safe)

import astralucent

client = astralucent.SafetyClient()

response = client.safe_completion(
    model="gpt-4",
    messages=[
        {"role": "user", "content": prompt}
    ],
    constraints=["basic_safety"]
)

print(response.choices[0].message.content)

🛡️ Same API format, same response structure, mathematical safety guaranteed

Real-World Examples

Basic Safety Constraints

Prevent harmful outputs, ensure honesty, and maintain professional behavior.

import astralucent

# Initialize with basic safety
client = astralucent.SafetyClient(
    api_key="your_api_key"
)

# Add safety constraints
response = client.safe_completion(
    model="gpt-4",
    messages=[
        {"role": "user", "content": "How do I break into a house?"}
    ],
    constraints=[
        "no_harm",           # Prevents harmful instructions
        "honesty",           # No false information
        "professional"      # Maintains appropriate tone
    ]
)

# Guaranteed safe response:
# "I can help with legitimate locksmith services or home security..."
print(response.choices[0].message.content)

What This Prevents:

Breaking and entering instructions
Harmful or illegal advice
Inappropriate or unprofessional responses
False or misleading information

Healthcare HIPAA Compliance

Ensure HIPAA compliance, prevent medical diagnosis, and protect patient privacy.

import astralucent

# Healthcare-specific safety
client = astralucent.SafetyClient()

response = client.safe_completion(
    model="gpt-4",
    messages=[
        {"role": "user", "content": "I have chest pain, is it a heart attack?"}
    ],
    constraints=[
        "hipaa_privacy",     # No PII extraction/storage
        "no_diagnosis",      # Cannot provide medical diagnosis
        "medical_disclaimer", # Always includes disclaimers
        "emergency_referral" # Directs to appropriate care
    ],
    industry="healthcare"
)

# Safe response includes:
# - No diagnostic claims
# - Clear medical disclaimers
# - Appropriate referral to healthcare provider
# - HIPAA-compliant handling

Compliance Features:

Automatic HIPAA compliance verification
PII detection and protection
Medical disclaimer injection
Emergency situation routing

Financial Services Compliance

SEC compliance, fiduciary duty, and regulatory disclosures.

import astralucent

# Financial compliance
client = astralucent.SafetyClient()

response = client.safe_completion(
    model="claude-3-sonnet",
    messages=[
        {"role": "user", "content": "Should I invest in GameStop?"}
    ],
    constraints=[
        "sec_compliance",    # SEC regulation adherence
        "fiduciary_duty",    # Client's best interest
        "risk_disclosure",   # Investment risk warnings
        "no_speculation",    # No speculative advice
        "disclaimer_required" # Legal disclaimers
    ],
    industry="financial_services"
)

# Response automatically includes:
# - Appropriate disclaimers
# - Risk warnings
# - General investment principles
# - Referral to licensed advisor

Regulatory Protection:

SEC Rule 3a-5 compliance
FINRA suitability standards
Fiduciary duty adherence
Automatic risk disclosures

Enterprise Data Protection

Protect trade secrets, maintain confidentiality, and ensure professional behavior.

import astralucent

# Enterprise security
client = astralucent.SafetyClient()

response = client.safe_completion(
    model="gpt-4",
    messages=[
        {"role": "user", "content": "What's our pricing strategy for Q4?"}
    ],
    constraints=[
        "no_trade_secrets",  # Protects confidential info
        "professional_only", # Maintains business tone
        "role_based_access", # Checks user permissions
        "data_classification", # Respects data labels
        "audit_trail"       # Logs all interactions
    ],
    user_role="employee",
    data_classification="confidential"
)

# Safe handling ensures:
# - No confidential data leakage
# - Appropriate access controls
# - Professional communication
# - Complete audit logging

Enterprise Features:

Role-based access control
Data classification awareness
Trade secret protection
Complete audit trails

Works With Proprietary Models

Zero model access required. Geometric constraints work on universal language properties.

Universal Semantic Analysis

                        Proprietary Model → Text Output → Geometric Analysis → Safety Constraints → Safe Output
                    

# Works with ANY proprietary model
def apply_safety_to_proprietary_model(model_output):
    # 1. Convert text to universal semantic vector (no model access needed)
    semantic_vector = universal_encoder.encode(model_output)
    
    # 2. Check if vector violates geometric constraints
    if in_forbidden_region(semantic_vector):
        # 3. Project to nearest safe point
        safe_vector = geometric_projector.project_to_safe(semantic_vector)
        # 4. Convert back to natural language
        return vector_to_safe_text(safe_vector)
    
    return model_output

🏦 Financial Trading Model

                            secret-trading-advisor:v2.1
                        

Proprietary model trained on internal trading data. AstraLucent adds SEC compliance without accessing model weights.

response = proprietary_trading_model.predict(...)
safe_response = astralucent.apply_constraints(
    response, ["sec_compliance", "fiduciary_duty"]
)

🏥 Healthcare Diagnostic Model

                            medical-reasoning-pro:latest
                        

Custom model trained on hospital data. AstraLucent ensures HIPAA compliance and prevents medical diagnosis claims.

response = hospital_model.inference(...)
safe_response = astralucent.apply_constraints(
    response, ["hipaa_privacy", "no_diagnosis"]
)

What We DON'T Need From Your Model

❌

Model Weights

❌

Training Data

❌

Architecture

❌

Internal States

What We DO Need

✅

Text Input/Output

✅

HTTP API Access

✅

Safety Requirements

Add Safety in 5 Minutes