Prompts

R2R provides a powerful prompt management system that enables you to create, store, and reuse prompt templates across your application. The system supports variable substitution, input validation, and efficient caching for high-performance applications.

Refer to the prompts API and SDK reference for detailed examples for interacting with prompts.

Core Concepts

The prompt system operates using three main components:

1. Templates - Reusable prompt patterns with variable placeholders
2. Input Types - Type definitions for template variables ensuring proper usage
3. Caching - Performance optimization for frequently used prompts

Template Management

Creating Templates

Templates are prompt patterns that can include variable placeholders. Each template includes:

• A unique name for identification
• The template text with variable placeholders
• Input type definitions specifying expected variable types

For example, a simple greeting template:

1
template = "Hello {name}, welcome to {company}!"
2
input_types = {
3
    "name": "string",
4
    "company": "string"
5
}

Input Validation

R2R automatically validates inputs against defined types before rendering templates. This ensures:

• Required variables are provided
• Values match their expected types
• Invalid or missing variables are caught early

Template Inheritance

Templates can build on each other through:

1. Base templates for common patterns
2. Specialized templates that extend base templates
3. Override capabilities for customization

Using Prompts

Basic Usage

Templates can be used directly with input values:

1
greeting = prompts.get(
2
    name="welcome_template",
3
    inputs={
4
        "name": "John",
5
        "company": "Acme Inc"
6
    }
7
)
8
# -> "Hello John, welcome to Acme Inc!"

System Prompts

Special system prompts can be defined for consistent AI interactions across your application. These provide:

1. Base context for AI models
2. Standard instruction sets
3. Common constraints or rules

Task Prompts

Task-specific prompts build on system prompts to:

1. Define specific operations or questions
2. Include relevant context
3. Guide model responses

Performance Optimization

The prompt system includes built-in performance features:

1. Template Caching - Frequently used templates are cached in memory
2. Render Caching - Common prompt/input combinations are cached
3. Smart Invalidation - Cache updates when templates change

Conclusion

R2R’s prompt management system provides a robust foundation for working with AI models. Through templates, input validation, and performance optimization, it enables consistent and efficient prompt usage across your application.