KnowledgeFocus LLM Node

Definition

KnowledgeFocus LLM is a pre-configured LLM node in GoInsight. It includes a Q&A template and is only used in Interactive Flow. It comes with default system prompts and restrictions, making it more suitable for building knowledge base-based chatbots. Compared with the General LLM node, KnowledgeFocus LLM focuses more on answering based on knowledge base content, reducing irrelevant responses or answers that deviate from the knowledge base.

Differences Between KnowledgeFocus LLM Node and General LLM Node

KnowledgeFocus LLM is an LLM node with built-in templates and system restrictions. It is more suitable for knowledge base Q&A scenarios.

General LLM Node is more flexible and is suitable when users need to write their own instructions, organize context, and control model outputs.

In short:

• If you are building a knowledge base-based chatbot, KnowledgeFocus LLM is recommended.
• If you need more freedom to design model behavior and output logic, you can use the General LLM node.

How to Configure

On the editing page of an Interactive Flow, right-click on the canvas or click the + icon at the end of the previous node to add a KnowledgeFocus LLM node.

1. Model

Select an available model based on the actual task. The available models depend on the administrator's configuration in Model Management and any custom models already connected by the enterprise.

2. Knowledge Retrieval Results

Knowledge Retrieval Results are typically used to receive output from an upstream Knowledge Retrieval node, passing relevant knowledge base content to the KnowledgeFocus LLM.

To have the KnowledgeFocus LLM answer questions based on these results, you must insert this variable into the prompt. Otherwise, the model will not automatically use the retrieved content as response context, even if the variable is configured.

3. User

The User field uses the user's Query from the Start node by default. It represents the current question submitted by the user.

4. Prompt and System Prompt

The KnowledgeFocus LLM node includes Prompt and System Prompt.

Prompt is used to supplement the chatbot's role, skills, response style, business requirements, and other instructions. Users can write the Prompt directly in the prompt editor, or type / or { to insert special variable blocks or variables from upstream nodes.

If you are not sure how to write the Prompt, you can use the AI Writing feature to let AI generate a Prompt based on your business scenario.

System Prompt is used to control the chatbot's basic behavior and response boundaries. KnowledgeFocus LLM includes built-in system instructions that restrict the response scope, making it more suitable for Q&A based on knowledge base content.

If Override Default Instructions is enabled, the custom System Prompt will override the default system instructions. After it is enabled, the chatbot may no longer be fully constrained by the original knowledge base response scope and security restrictions, so use this option with caution.

5. Recommend relevant queries

After this option is enabled, KnowledgeFocus LLM can include follow-up questions related to the current question when answering the user. This helps users continue asking questions or explore related content.

If you do not need to display follow-up questions in the response, you can disable this option.

6. Token Distribution

Token Distribution is used to adjust the token allocation ratio between context content and the chatbot's response.

• Context: Controls the token ratio available for context content.
• Response: Controls the token ratio available for the chatbot's response.

If the retrieved knowledge base content is long, you can increase the Context ratio as needed. If you want the chatbot to generate a more complete response, you can increase the Response ratio as needed.

7. Memory

After Memory is enabled, the KnowledgeFocus LLM node can include historical conversation content each time it runs. This helps the model understand previous context and improves response consistency in multi-turn conversations.

Chat history supports two configuration methods:

• Specified: Directly specify the number of historical messages to reference.
• Interpolated: Pass historical messages through a variable.

8. Output Variable

The output of the KnowledgeFocus LLM node is usually connected to a subsequent Answer node or End node to return the model-generated content to the user.

KnowledgeFocus LLM also supports referencing upstream node variables in the Prompt, such as the output of a Knowledge Retrieval node, to provide response context for the model.

9. Error Handling

The KnowledgeFocus LLM node supports error handling.

When an error occurs during node execution, you can choose one of the following options:

• No Handling: If the node fails, the workflow run will be interrupted.
• Default Value: If the node fails, the workflow continues to run and uses the configured default value as the node output.
• Error Handling Branch: If the node fails, the current path is interrupted and redirected to the configured error handling branch.

Summary

KnowledgeFocus LLM is a knowledge base Q&A LLM node designed for Interactive Flow.

It is suitable for building chatbots based on enterprise knowledge bases. When configuring it, focus on Model, Knowledge Retrieval Results, Prompt, System Prompt, Token Distribution, Memory, and Error Handling.