LLM (Agentflow Node)

The LLM node in InnoSynth-Forjinn's Agentflow is a fundamental component for integrating Large Language Models (LLMs) into your workflows. It allows you to leverage the power of various conversational AI models to analyze inputs, generate responses, and perform a wide range of language-based tasks.

Purpose

The LLM node is designed for:

• Text Generation: Creating human-like text for responses, summaries, or creative content.
• Question Answering: Providing answers to user queries based on its training data and provided context.
• Content Analysis: Analyzing text for sentiment, keywords, or other insights.
• Structured Output: Generating responses in a predefined JSON format for programmatic use.

Appearance on Canvas

(Image: Placeholder for the LLM node icon/appearance on the canvas)

Configuration Parameters

The LLM node offers extensive configuration options to control the behavior of the integrated LLM.

1. Model (Required)

• Label: Model
• Type: asyncOptions (loads available Chat Models)
• Description: Select the specific Large Language Model you want to use (e.g., OpenAI Chat Model, Anthropic Chat Model).
• Configuration: You can configure the selected model's specific parameters (e.g., API key, temperature, model name) via the llmModelConfig property, which is loaded dynamically.

2. Messages (Optional)

• Label: Messages
• Type: array of objects
• Description: Pre-define a series of messages (System, Assistant, Developer, User) to set the context, persona, or initial instructions for the LLM. These messages are prepended to the conversation history.
  • Role: system, assistant, developer, user
  • Content: The message text. Supports variables.

3. Enable Memory (Optional)

• Label: Enable Memory
• Type: boolean
• Default: true
• Description: If enabled, the LLM will maintain conversational memory, allowing it to remember past interactions within the current session.

4. Memory Type (Optional)

• Label: Memory Type
• Type: options
• Default: allMessages
• Description: Defines how the LLM manages its conversational history.
  • All Messages: Retrieves all messages from the conversation.
  • Window Size: Uses a fixed window to surface the last N messages.
  • Conversation Summary: Summarizes the entire conversation.
  • Conversation Summary Buffer: Summarizes conversations once a token limit is reached.

5. Window Size (Optional)

• Label: Window Size
• Type: number
• Default: 20
• Description: (Visible when Memory Type is "Window Size") Specifies the number of recent messages to retain in memory.

6. Max Token Limit (Optional)

• Label: Max Token Limit
• Type: number
• Default: 2000
• Description: (Visible when Memory Type is "Conversation Summary Buffer") The maximum token limit before the conversation is summarized.

7. Input Message (Optional)

• Label: Input Message
• Type: string
• Description: Add an additional input message as a user message at the end of the conversation. Useful for dynamic inputs.

8. Return Response As

• Label: Return Response As
• Type: options
• Default: userMessage
• Options: User Message, Assistant Message
• Description: Determines how the LLM's final response is formatted in the output.
  • User Message: The response is treated as if it came from a user.
  • Assistant Message: The response is treated as if it came from an assistant.

9. JSON Structured Output (Optional)

• Label: JSON Structured Output
• Type: array of objects
• Description: Instructs the LLM to generate its output in a specific JSON schema. This is powerful for integrating LLM responses into programmatic workflows.
  • Key: The key name in the JSON output.
  • Type: The data type of the value (String, String Array, Number, Boolean, Enum, JSON Array).
  • Enum Values: (Visible for Enum type) Comma-separated list of allowed enum values.
  • JSON Schema: (Visible for JSON Array type) A JSON schema defining the structure of items within the array.
  • Description: A description of the key, guiding the LLM on what to generate.

10. Update Flow State (Optional)

• Label: Update Flow State
• Type: array of objects
• Description: Allows you to update the runtime state of the workflow during the LLM's execution.
  • Key: The key of the state variable to update.
  • Value: The new value for the state variable. Supports variables and the output of the LLM ({{ output }} or {{ output.jsonPath }}).

Inputs & Outputs

• Input: Typically receives a user message, a prompt, or contextual information from a preceding node.
• Output: Produces the LLM's generated response, which can be plain text or a structured JSON object (if configured). It also includes usageMetadata (token counts).

Example Usage: Summarizing Text

Let's create a simple chatflow to summarize user-provided text.

1. Connect a Start node to the input of an LLM node.
2. Configure Model: Select an OpenAI Chat Model.
3. Configure Messages: Add a System message: "You are a helpful assistant that summarizes text."
4. Configure Input Message: {{question}} (assuming user input is passed as question variable).
5. Connect to Chat Output: Connect the output of the LLM node to a Chat Output node.

Now, when a user provides text, the LLM will summarize it and return the summary.