LLM Configuration
Large Language Models (LLMs) are the brain of your Saiki agents. This section covers everything you need to know about configuring LLMs for your agents.
Overview
Saiki supports multiple LLM providers out-of-the-box via the Vercel AI SDK. You can also use any OpenAI SDK-compatible provider. You can easily switch between different models and providers without changing your application code.
Quick Start
Basic Configuration
llm:
provider: openai
model: gpt-4.1-mini
apiKey: $OPENAI_API_KEY
systemPrompt: |
You are Saiki, a helpful AI assistant with access to tools.
Use these tools when appropriate to answer user queries.
Environment Variables
Set API keys in your .env
file:
OPENAI_API_KEY=your_openai_key
ANTHROPIC_API_KEY=your_anthropic_key
GOOGLE_GENERATIVE_AI_API_KEY=your_google_key
GROQ_API_KEY=your_groq_key
What's in This Section
Configuration Reference
Complete guide to all LLM configuration options including:
- Required and optional fields
- System prompt configuration
- Provider-specific options
- Advanced settings
Supported Providers
Detailed guide to all supported LLM providers:
- Built-in providers: OpenAI, Anthropic, Google, Groq
- Custom providers: Azure OpenAI, OpenRouter, Together.ai, local models
- Setup instructions for each provider
- Supported models and their capabilities
Key Concepts
Providers vs Models
- Provider: The service hosting the LLM (e.g., OpenAI, Anthropic)
- Model: The specific AI model to use (e.g., gpt-4.1-mini, claude-3-5-sonnet)
System Prompts
System prompts define how your agent behaves. They can be:
- Simple strings for basic use cases
- Advanced configurations with multiple contributors for complex scenarios
Routers
Saiki uses routers to handle LLM requests:
- Vercel router (default): Optimized for performance and reliability
- In-built router: Direct communication with providers
Next Steps
- New to LLMs? Start with the Configuration Reference to understand the basics
- Looking for a specific provider? Check the Providers Guide for setup instructions
- Building an agent? Head to Building with Saiki for implementation patterns