Core Concepts

Understanding SkyAIApp

Deep dive into SkyAIApp's core concepts to build efficient, reliable AI applications.

What is SkyAIApp?

SkyAIApp is an AI orchestration platform that helps developers build, deploy, and scale production-grade AI applications through intelligent routing, agent runtime, and enterprise-grade security controls.

18-35% Cost Savings

< 1.2s P95 Latency

99.5% Success Rate

Smart Routing

Smart Routing is the core capability of SkyAIApp. It automatically selects the optimal AI model to handle requests based on your goals (cost, quality, stability) and strategies.

How it works

1Analyze request features: complexity, length, type
2Evaluate available models: performance, cost, availability
3Apply optimization strategy: balance metrics based on goal
4Intelligent fallback: auto-switch when primary model unavailable
5Continuous learning: optimize based on historical data

// Simple routing example
const response = await sky.route({
  goal: "cost",              // Optimize for cost
  strategy: "balanced",       // Balanced strategy
  messages: [...],
  
  // Optional: Manual fallback chain
  fallback: {
    models: ["gpt-5.2-instant", "claude-haiku-4.5"],
    maxRetries: 3
  }
});

Optimization Goals

Cost

Minimize API call costs. Ideal for large-scale batch processing and simple queries.

Use cases:

Data labelingContent classificationSimple Q&A

Quality

Pursue highest output quality. Ideal for complex reasoning and creative generation.

Use cases:

Code generationDeep analysisCreative writing

Stability

Ensure service reliability and consistency. Ideal for production and critical business scenarios.

Use cases:

Customer supportFinancial analysisMedical consultation

Routing Strategies

Strategies define how to balance different metrics under a given goal.

balanced

Balanced

Balance between cost, quality, and latency

cost-optimized

Cost-Optimized

Aggressively minimize cost, may sacrifice some quality

quality-first

Quality-First

Prioritize most powerful models

latency-optimized

Latency-Optimized

Choose fastest responding models

💡 Tip: You can use custom strategies or create strategy templates in the dashboard. Learn more →

Agent Runtime

Agent Runtime enables you to build AI agents that can execute multi-step tasks, use tools, and make autonomous decisions.

Key Features

Tool Calling

Integrate external APIs, databases, search engines

Multi-step Reasoning

Auto-plan and execute complex tasks

Sandbox Execution

Safely isolated code execution

State Management

Maintain context across steps

const agent = sky.createAgent({
  tools: ["web_search", "calculator", "code_exec"],
  maxSteps: 10,
  onStep: (step) => {
    console.log(`${step.number}. ${step.action}`);
  }
});

const result = await agent.run({
  task: "Find Bitcoin price and calculate 10% commission"
});

Enterprise Guardrails

Built-in security mechanisms to protect your applications from harmful content, privacy leaks, and other risks.

PII Filtering

Auto-detect and handle personally identifiable information

Supports: Email, phone, SSN, credit cards, etc.

Content Moderation

Filter harmful, inappropriate, or policy-violating content

Supports: Hate speech, violence, sexual content, etc.

Topic Filtering

Restrict discussions on specific topics or domains

Supports: Illegal activities, weapons, adult content, etc.

Security best practices guide →

Semantic Caching

Semantic caching not only matches identical queries but also recognizes semantically similar requests, dramatically improving cache hit rates.

Traditional Caching

✓ "What is AI?"

✗ "What's artificial intelligence?"

✗ "Explain AI to me"

Semantic Caching

✓ "What is AI?"

✓ "What's artificial intelligence?"

✓ "Explain AI to me"

Significant Cost Reduction

Average 34% cache hit rate, saving 30-40% on API call costs.

const response = await sky.route({
  goal: "cost",
  messages: [...],
  cache: true,  // Enable semantic caching
});

// Check cache hit
console.log("Cache hit:", response.routing.cache_hit);

Next Steps

Now that you understand the core concepts, you can start building your first AI application.

Quick Start

5-minute integration

Routing Strategies

Advanced optimization

Authentication

Overview

Was this page helpful?

Let us know how we can improve