LLM

LLM Node

The LLM (Large Language Model) node is responsible for generating responses based on user input using a language model. It performs natural language processing tasks according to the agent’s purpose.

LLM Model

Select the LLM model that the agent will use to generate responses.

Available models may vary depending on your billing plan (for details, please contact the sales team).

Refer to the following model guides to select the LLM best suited for your service:

GPT

Claude

Xionic

Xionic is Sionic’s proprietary model. (Contact the sales team if you wish to use it.)

LLM Parameters

Top P

Top P controls the randomness of the generated text using nucleus sampling.
The model selects words randomly from the smallest possible set whose cumulative probability equals P%.
For example, with Top P = 0.9, the model only considers the words whose cumulative probability reaches 90%.
A higher Top P (closer to 1) results in more diverse and creative outputs.

Maximum Length

Temperature

Frequency Penalty

Presence Penalty

Stop Sequences

Seed

Prompt Configuration

Customize the instruction prompts used by the model during generation.

Prompts are divided into SYSTEM and USER roles:

SYSTEM Prompt: Defines the model’s tone, behavior, and personality.
It provides overarching guidance for how the model should respond.

USER Prompt: Contains the actual user query or task request.
The model generates responses based on this input.

Additional Features

Function Call

Allows the model to call predefined functions for integration with external systems.

Each function requires parameters such as name, description, and properties.

Knowledge Chunk Format

Defines how retrieved knowledge is passed into the LLM.

{{Index}} — Automatically numbers retrieved items.

{{Filename}} — Displays the source file name.

{{Page}} — Shows page metadata (may not exist for all Korean documents).

{{Content}} — Includes the main retrieved content.

History Turn Count

Specifies how many previous conversation turns the model can reference.

A higher value improves contextual understanding but increases token usage.

LLM Node#

LLM Model#

LLM Parameters#

Prompt Configuration#

Additional Features#

Function Call#

Knowledge Chunk Format#

History Turn Count#