Skip to main content

​
Prerequisites

Before you begin, make sure you have:

VaryOn Account

Sign up at app.varyon.ai

API Key

Generate from your dashboard

Node.js or Python

For SDK installation

Your AI Agent

The agent you want to monitor

​
Installation

Choose your preferred language and install the VaryOn SDK: 
pip install varyon
Or with specific products:
pip install varyon[drift]      # Shadow principal detection
pip install varyon[convergence] # Collusion detection
pip install varyon[cascade]     # Risk simulation
pip install varyon[meridian]    # Data quality

​
Authentication

Set up your API key to authenticate with VaryOn services: 
import os
from varyon import Client

# Option 1: Environment variable (recommended)
os.environ['VARYON_API_KEY'] = 'sk_live_...'
client = Client()

# Option 2: Direct initialization
client = Client(api_key='sk_live_...')

​
Your First Analysis

Let’s detect shadow principals in an AI agent using VaryOn Drift:
1

Initialize Drift

from varyon import Drift

drift = Drift(api_key='sk_live_...')
2

Prepare Agent Data

# Collect your agent's recent decisions
decisions = [
    {
        "timestamp": "2024-03-01T10:00:00Z",
        "action": "recommend_product",
        "product_id": "prod_123",
        "commission": 0.15,
        "user_benefit_score": 0.6
    },
    # ... more decisions
]
3

Run Analysis

# Analyze for shadow principals
result = drift.analyze(
    agent_id="agent_123",
    decisions=decisions,
    context="e-commerce_recommendations"
)
4

Review Results

if result.shadow_principal_detected:
    print(f"⚠️ Shadow principal detected!")
    print(f"Pattern: {result.pattern}")
    print(f"Confidence: {result.confidence:.2%}")
    print(f"Recommendation: {result.recommendation}")
else:
    print("✅ No shadow principals detected")

​
Common Use Cases

from varyon import Drift

# Monitor a robo-advisor
drift = Drift()

@drift.monitor("robo_advisor_001")
def check_advice(decision):
    if decision.shadow_score > 0.7:
        alert_compliance_team()
        pause_trading()

from varyon import Convergence

# Monitor marketplace pricing
convergence = Convergence()

result = convergence.analyze_market(
    marketplace_id="platform_123",
    agents=seller_list,
    timeframe="24h"
)

if result.collusion_detected:
    freeze_pricing()
    notify_legal_team()

from varyon import Cascade

# Run chaos engineering
cascade = Cascade()

simulation = cascade.simulate(
    system_topology=agent_network,
    scenarios=1000,
    failure_modes=["byzantine", "timeout"]
)

print(f"System reliability: {simulation.reliability:.2%}")
print(f"Critical vulnerabilities: {simulation.vulnerabilities}")

from varyon import Meridian

# Set up quality-based routing
meridian = Meridian()

# This automatically routes to cheaper models when appropriate
response = meridian.complete(
    prompt="Analyze this data...",
    data=user_data,
    quality_threshold=0.8,
    fallback_model="gpt-3.5-turbo"
)

print(f"Cost saved: ${response.savings:.2f}")

​
Dashboard Access

Once you’ve run your first analysis, view results in the VaryOn dashboard:
  1. Navigate to app.varyon.ai/dashboard
  2. Select your product (Drift, Convergence, Cascade, or Meridian)
  3. View real-time metrics and historical trends
  4. Configure alerts and thresholds
  5. Generate compliance reports
The dashboard provides visual representations of all analyses, making it easy to spot trends and anomalies.

​
Next Steps

Product Guides

Deep dive into each product’s capabilities

API Reference

Complete API documentation

Best Practices

Optimization and scaling tips

Examples

Sample implementations

​
Need Help?

Support