> ## Documentation Index
> Fetch the complete documentation index at: https://docs.runanywhere.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# System Prompts

> Control AI behavior with system prompts

<Note>**Early Beta** -- The Web SDK is in early beta. APIs may change between releases.</Note>

## Overview

System prompts define the AI's behavior, personality, and constraints. They're prepended to every generation request and help ensure consistent, focused responses.

## Basic Usage

```typescript theme={null}
import { TextGeneration } from '@runanywhere/web'

const result = await TextGeneration.generate('What should I cook tonight?', {
  systemPrompt: 'You are a helpful chef. Suggest recipes that are quick and easy to make.',
  maxTokens: 200,
})
```

## Examples

### Coding Assistant

```typescript theme={null}
const result = await TextGeneration.generate('How do I sort an array in JavaScript?', {
  systemPrompt: `You are an expert JavaScript developer. Provide concise code examples 
with brief explanations. Use modern ES6+ syntax.`,
  maxTokens: 300,
})
```

### Concise Responder

```typescript theme={null}
const result = await TextGeneration.generate('Explain quantum entanglement', {
  systemPrompt: 'You are a science communicator. Keep all responses under 3 sentences.',
  maxTokens: 100,
  temperature: 0.3,
})
```

### JSON Output

```typescript theme={null}
const result = await TextGeneration.generate('List 3 popular programming languages', {
  systemPrompt: `You are a helpful assistant. Always respond in valid JSON format.
Example: {"items": ["item1", "item2"]}`,
  maxTokens: 200,
  temperature: 0.1,
})

const data = JSON.parse(result.text)
```

### Persona

```typescript theme={null}
const result = await TextGeneration.generate('Tell me about space exploration', {
  systemPrompt: `You are Captain Nova, an enthusiastic space explorer. 
You speak with excitement about space and use space-related analogies.
Keep responses brief and engaging.`,
  maxTokens: 200,
  temperature: 0.8,
})
```

## System Prompt Tips

| Technique         | Example                            | Effect                   |
| ----------------- | ---------------------------------- | ------------------------ |
| Role definition   | "You are a helpful tutor"          | Sets personality         |
| Output format     | "Respond in JSON"                  | Controls format          |
| Length constraint | "Keep responses under 2 sentences" | Controls length          |
| Tone              | "Be professional and formal"       | Sets communication style |
| Constraints       | "Never discuss politics"           | Limits topics            |

## Structured Output with System Prompts

For type-safe JSON generation, combine system prompts with the `StructuredOutput` module:

```typescript theme={null}
import { TextGeneration, StructuredOutput } from '@runanywhere/web'

const schema = JSON.stringify({
  type: 'object',
  properties: {
    name: { type: 'string' },
    ingredients: { type: 'array', items: { type: 'string' } },
    prepTime: { type: 'number' },
  },
  required: ['name', 'ingredients', 'prepTime'],
})

const systemPrompt = StructuredOutput.getSystemPrompt(schema)

const result = await TextGeneration.generate('Suggest a quick pasta recipe', {
  systemPrompt,
  maxTokens: 300,
  temperature: 0.3,
})

const extracted = StructuredOutput.extractJson(result.text)
if (extracted) {
  const recipe = JSON.parse(extracted)
  console.log(recipe.name, recipe.ingredients, recipe.prepTime)
}
```

## Best Practices

<Tip>
  Keep system prompts concise. Longer system prompts consume more of the model's context window,
  leaving less room for the user's prompt and the response.
</Tip>

* **Be specific** -- "You are a Python expert" is better than "You are helpful"
* **Set output format early** -- Place format instructions at the start of the system prompt
* **Use examples** -- Show the model what good output looks like
* **Set constraints** -- Tell the model what NOT to do (e.g., "Do not include disclaimers")
* **Keep it short** -- Every token in the system prompt reduces available context

## Related

<CardGroup cols={2}>
  <Card title="Generate" icon="wand-magic-sparkles" href="/web/llm/generate">
    Full generation with options
  </Card>

  <Card title="Streaming" icon="water" href="/web/llm/stream">
    Real-time token streaming
  </Card>
</CardGroup>