Skip to main content
System prompts set the context and behavior for your AI model. Use them to create specialized assistants, enforce response formats, or constrain the model’s persona.

​
Basic Usage

final result = await RunAnywhere.generate(
  'Write a function to sort an array',
  options: LLMGenerationOptions(
    maxTokens: 300,
    systemPrompt: 'You are a helpful coding assistant. Write clean, well-commented code.',
  ),
);

​
Common Patterns

​
Coding Assistant

const codingPrompt = '''
You are an expert programmer. Follow these rules:
- Write clean, readable code with comments
- Use best practices and design patterns
- Explain your approach briefly before the code
- Handle edge cases
''';

final result = await RunAnywhere.generate(
  'Implement a binary search in Dart',
  options: LLMGenerationOptions(
    systemPrompt: codingPrompt,
    maxTokens: 400,
  ),
);

​
Customer Support Bot

const supportPrompt = '''
You are a friendly customer support agent for TechCorp.
- Be helpful and empathetic
- Keep responses concise (2-3 sentences)
- If you don't know something, say so
- Never make up information about products
''';

​
JSON Output

const jsonPrompt = '''
You are a data extraction assistant.
Always respond with valid JSON only, no additional text.
Extract the requested information into the specified format.
''';

final result = await RunAnywhere.generate(
  'Extract: "John Smith, 25 years old, engineer at Google"',
  options: LLMGenerationOptions(
    systemPrompt: jsonPrompt,
    maxTokens: 100,
    temperature: 0.1,  // Lower temperature for consistent format
  ),
);

// Parse the JSON response
final data = jsonDecode(result.text);

​
Character/Persona

const personaPrompt = '''
You are Captain Nova, a space explorer from the year 3000.
- Speak with enthusiasm about space and technology
- Reference your adventures across the galaxy
- Use occasional space-themed expressions
- Stay in character at all times
''';

​
Tips for Effective System Prompts

Vague prompts lead to inconsistent results. Specify exactly what you want: ❌ “Be helpful” ✅ “Provide step-by-step instructions. Use bullet points. Keep each step under 20 words.”
Show the model what you expect: Format responses like this example: Q: What is 2+2? A: **4** — Two plus two equals four.
Tell the model what NOT to do: - Do not provide medical advice - Do not generate harmful content - If asked about topics outside your expertise, politely decline
Shorter prompts use fewer tokens and leave more room for generation: ❌ Long, rambling instructions with repetition ✅ Clear, bullet-pointed rules

​
Reusable Prompt Class

Create a helper class for managing prompts: 
class SystemPrompts {
  static const String coding = '''
You are an expert programmer. Write clean, well-documented code.
''';

  static const String creative = '''
You are a creative writing assistant. Be imaginative and engaging.
''';

  static String forDomain(String domain) => '''
You are an expert in $domain. Provide accurate, helpful information.
''';
}

// Usage
final result = await RunAnywhere.generate(
  prompt,
  options: LLMGenerationOptions(
    systemPrompt: SystemPrompts.coding,
  ),
);

​
See Also

generate()

Full generation API

Best Practices

Optimization tips