LLM Element

The <llm> element enables AI/Language Model integration:

<llm 
model="account/fireworks/model/deepseek-v3"
temperature={0.7}
includeChatHistory={true}
>
<instructions>
  Process the input according to these instructions.
</instructions>
<prompt>
  {({userInput}) => userInput.message}
</prompt>
</llm>

Prop	Type	Default
`id?`	`string`	-
`model`	`string`	-
`temperature?`	`number`	-
`includeChatHistory?`	`boolean`	-
`stopSequences?`	`string[]`	-
`topP?`	`number`	-
`toolChoice?`	`string`	-
`tools?`	`object[]`	-
`grammar?`	`object`	-
`repetitionPenalty?`	`number`	-
`responseFormat?`	`object \| string`	-

Allowed Children

prompt: The prompt text
instructions: System instructions

Examples

Basic Usage

<llm model="account/fireworks/model/deepseek-v3">
<prompt>
  Summarize the following text: {({userInput}) => userInput.text}
</prompt>
</llm>

With Instructions

<llm 
model="account/fireworks/model/deepseek-v3"
temperature={0.5}
>
<instructions>
  You are a helpful assistant that provides concise answers.
  Always be polite and professional.
</instructions>
<prompt>
  {({userInput}) => userInput.question}
</prompt>
</llm>

Advanced Configuration

<llm
model="account/fireworks/model/deepseek-v3"
temperature={0.8}
topP={0.9}
includeChatHistory={true}
stopSequences={["END", "STOP"]}
repetitionPenalty={1.2}
responseFormat="text"
>
<instructions>
  {({state}) => `
    Current context: ${state.context}
    Previous interactions: ${state.history}
    Respond based on this context.
  `}
</instructions>
<prompt>
  {({userInput, state}) => `
    User query: ${userInput.message}
    Current state: ${state.status}
  `}
</prompt>
</llm>

Usage Notes

Model selection affects capabilities
Temperature controls randomness
Instructions guide model behavior
Prompts can be dynamic
Can include context and history

Best Practices

Model Selection
- Choose appropriate model
- Consider model capabilities
- Balance cost and performance
- Test model behavior
Prompt Engineering
- Write clear prompts
- Include relevant context
- Use consistent format
- Handle edge cases
Parameter Tuning
- Adjust temperature for task
- Set appropriate top-p
- Configure stop sequences
- Test different settings
Error Handling
- Handle model errors
- Provide fallbacks
- Log issues
- Monitor performance

Common Patterns

Conversational Agent

<llm
model="account/fireworks/model/deepseek-v3"
includeChatHistory={true}
>
<instructions>
  You are a helpful assistant. Be concise and friendly.
  Previous conversation: {({state}) => state.chatHistory}
</instructions>
<prompt>
  User: {({userInput}) => userInput.message}
  Assistant:
</prompt>
</llm>

Data Processing

<llm
model="account/fireworks/model/deepseek-v3"
temperature={0.2}
responseFormat={{
  type: "json",
  schema: {
    type: "object",
    properties: {
      category: { type: "string" },
      sentiment: { type: "string" },
      keywords: { type: "array" }
    }
  }
}}
>
<instructions>
  Analyze the text and return a JSON object with:
  - category (topic category)
  - sentiment (positive/negative/neutral)
  - keywords (array of important terms)
</instructions>
<prompt>
  {({userInput}) => userInput.text}
</prompt>
</llm>

Tool Usage

<llm
model="account/fireworks/model/deepseek-v3"
tools={[
  {
    name: "search",
    description: "Search for information",
    parameters: {
      query: "string"
    }
  }
]}
toolChoice="auto"
>
<instructions>
  Use the search tool when you need to find information.
</instructions>
<prompt>
  {({userInput}) => userInput.question}
</prompt>
</llm>

LLM Element

On this page