# LLM
**Action ID:** `llm`
## Description
Use a large language model such as GPT
## Input Parameters
| Name | Type | Required | Default | Description |
| ----------------- | -------- | :------: | ----------------- | -------------------------------------------------------------------------------------------------------------- |
| human\_message | string | ✓ | - | The prompt that is fed to the model (max 640,000 characters) |
| system\_message | string | - | null | Instructions for the AI assistant on how to behave and respond (max 640,000 characters) |
| chat\_history\_id | string | - | null | The chat history ID used to retrieve conversation history. If not provided, a new chat history will be created |
| model | dropdown | - | DeepSeek V4 Flash | The AI model to use for generating the response |
| temperature | number | - | 0.5 | Controls randomness in responses (0.0-1.0). Higher values result in more creative responses |
View JSON Schema
```json
{
"description": "LLM node input.",
"properties": {
"human_message": {
"description": "The prompt that is fed to the model.",
"maxLength": 640000,
"title": "Prompt",
"type": "string"
},
"system_message": {
"anyOf": [
{
"maxLength": 640000,
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "This represents a system message, which tells the model how to behave.",
"title": "System Message"
},
"chat_history_id": {
"anyOf": [
{
"type": "string"
},
{
"type": "null"
}
],
"default": null,
"description": "The chat history id used to retrieve the chat history. If not provided, a new chat history will be created.",
"title": "Chat History ID"
},
"model": {
"default": "DeepSeek V4 Flash",
"description": "The AI model to use to generate the response.",
"enum": [
"OpenAI GPT-4o Mini",
"DeepSeek V4 Flash",
"Google Gemini 2.0 Flash-Lite",
"Google Gemini 2.0 Flash",
"Google Gemini 1.5 Flash",
"Google Gemini 1.5 Pro",
"Anthropic Claude 3 Haiku"
],
"title": "Model",
"type": "string"
},
"temperature": {
"default": 0.5,
"description": "The temperature to use when generating the response. Higher temperatures will result in more creative responses.",
"maximum": 1,
"minimum": 0,
"title": "Temperature",
"type": "number"
}
},
"required": [
"human_message"
],
"title": "LLMQuestionAnsweringInput",
"type": "object"
}
```
## Output Parameters
| Name | Type | Description |
| ----------------- | ------ | --------------------------------------------------------- |
| content | string | The AI-generated response text |
| chat\_history\_id | string | The chat history ID used to maintain conversation context |
View JSON Schema
```json
{
"description": "LLM node output.",
"properties": {
"content": {
"title": "Content",
"type": "string"
},
"chat_history_id": {
"title": "The chat history id used to retrieve the chat history.",
"type": "string"
}
},
"required": [
"content",
"chat_history_id"
],
"title": "LLMQuestionAnsweringNodeOutput",
"type": "object"
}
```
## How It Works
This node processes your text prompts through powerful large language models to generate intelligent responses. When you provide a human message (your question or request), the node sends it to the selected AI model along with any optional system instructions that define the model's behavior. The model analyzes the input, considers any chat history for context continuity, and generates a relevant response. The temperature parameter controls the randomness of outputs, with lower values producing more focused and deterministic responses, while higher values enable more creative and varied outputs.
## Usage Examples
### Example 1: Simple Question Answering
**Input:**
```
human_message: "What are the three primary colors and why are they important in art?"
model: "DeepSeek V4 Flash"
temperature: 0.3
```
**Output:**
```
content: "The three primary colors are red, blue, and yellow. They are important in art because they cannot be created by mixing other colors, and all other colors can be created by combining these primaries in various proportions. This fundamental principle forms the basis of color theory and allows artists to create a full spectrum of colors from just these three base colors."
chat_history_id: "chat_abc123def456"
```
### Example 2: Content Generation with System Instructions
**Input:**
```
human_message: "Write a professional email apologizing for a delayed shipment"
system_message: "You are a professional customer service representative. Write concise, empathetic emails that acknowledge issues and provide solutions."
model: "OpenAI GPT-4o Mini"
temperature: 0.7
```
**Output:**
```
content: "Subject: Sincere Apologies for Your Delayed Shipment\n\nDear Valued Customer,\n\nI sincerely apologize for the delay in your recent order. We understand how frustrating this must be, and we take full responsibility for not meeting our delivery commitment.\n\nYour package is now en route and should arrive within 2-3 business days. To make up for this inconvenience, we've applied a 15% discount to your next purchase.\n\nThank you for your patience and understanding.\n\nBest regards,\nCustomer Service Team"
chat_history_id: "chat_789ghi012jkl"
```
### Example 3: Continuing a Conversation
**Input:**
```
human_message: "Can you elaborate on the mixing ratios?"
chat_history_id: "chat_abc123def456"
model: "Google Gemini 2.0 Flash"
temperature: 0.5
```
**Output:**
```
content: "Certainly! When mixing primary colors, the ratios determine which secondary or tertiary color you'll create. For example:\n\n- Equal parts red + blue = purple\n- Equal parts blue + yellow = green\n- Equal parts red + yellow = orange\n\nBy adjusting the ratios (like 2 parts red to 1 part blue), you can create variations like violet or burgundy. The possibilities are virtually endless, which is why understanding primary colors is so fundamental to color mixing."
chat_history_id: "chat_abc123def456"
```
## Common Use Cases
* **Customer Support Automation**: Generate intelligent responses to customer inquiries, support tickets, and frequently asked questions
* **Content Creation**: Write blog posts, product descriptions, marketing copy, social media content, and advertising text
* **Data Analysis**: Analyze text data, extract insights, summarize reports, and identify patterns in large datasets
* **Code Assistance**: Generate code snippets, explain technical concepts, debug issues, and provide programming guidance
* **Translation and Localization**: Translate content between languages while maintaining context and cultural nuances
* **Educational Content**: Create explanations, tutorials, study guides, and answer student questions on various topics
* **Conversational AI**: Build chatbots, virtual assistants, and interactive conversation systems with maintained context
## Error Handling
| Error Type | Cause | Solution |
| ------------------------ | --------------------------------------------------------- | ---------------------------------------------------------------------------------- |
| Empty Message Error | human\_message parameter is empty or missing | Ensure human\_message contains text before sending to the model |
| Token Limit Exceeded | Input message exceeds 640,000 character limit | Reduce the length of your prompt or split it into multiple requests |
| Invalid Chat History | chat\_history\_id doesn't exist or is expired | Start a new conversation by omitting chat\_history\_id or verify the ID is correct |
| Model Unavailable | Selected model is temporarily down or unavailable | Switch to an alternative model or retry after a few minutes |
| Temperature Out of Range | Temperature value is not between 0.0 and 1.0 | Set temperature to a value within the valid range (0.0-1.0) |
| Rate Limit Error | Too many requests sent in a short time period | Implement request throttling or wait before sending additional requests |
| Invalid System Message | System message contains unsupported content or formatting | Simplify system message and ensure it contains valid instructions only |
## Notes
* **Model Selection**: DeepSeek V4 Flash offers excellent performance for most tasks. Use GPT-4o Mini for cost efficiency, Gemini for multimodal capabilities, and Claude for detailed analysis
* **Temperature Control**: Use 0.0-0.3 for factual, deterministic responses (data analysis, extraction). Use 0.7-1.0 for creative tasks (content generation, brainstorming)
* **System Messages**: Define the model's role, tone, and behavior. Examples: "You are a technical expert", "Respond concisely", "Use a friendly tone"
* **Chat History**: Reuse chat\_history\_id to maintain conversation context across multiple turns. Essential for building conversational experiences
* **Prompt Engineering**: Clear, specific prompts produce better results. Include context, desired format, and any constraints in your prompt
* **Character Limits**: Both human\_message and system\_message support up to 640,000 characters, allowing for extensive context and detailed instructions
* **Response Quality**: More advanced models (Gemini Pro, Claude) provide better reasoning but cost more. Balance quality needs with budget constraints
* **Workflow Integration**: Chain multiple LLM nodes together for complex tasks like draft-review-refine workflows or multi-step analysis
---
# Agent Instructions: Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
Perform an HTTP GET request on the current page URL with the `ask` query parameter:
```
GET https://docs.agenticflow.ai/reference/nodes/llm.md?ask=
```
The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.