Workfx QA Agent

In this section, we introduce a representative use case featuring a simple Agent Flow designed specifically for Workfx-related questions. Key points include:

• The agent receives user input in the form of questions.

• It performs targeted retrieval from the Workfx introduction document using a predefined knowledge search process.

• The agent generates relevant answers based strictly on retrieved content.

• The system is restricted to a single domain to ensure all responses align with Workfx’s official documentation.

Overview of Flow Structure

The execution logic of this flow is as follows:

• Question Classifier Node The flow begins by determining whether the user’s question is related to Workfx.

• Workfx-Related Questions

  • If the question is relevant to Workfx, the system invokes the knowledge search tool defined in the “Tool-Flow Tool” section;

  • The tool performs retrieval and returns related content;

  • The agent then generates a natural language response based on the retrieved result.

• Unrelated Questions

  • If the question is not related to Workfx, a Code Node is used to return a fixed message such as: “Sorry, I can only answer questions related to Workfx.”

Question Classifier Node

Model Configuration

• Model: gpt-4o-mini

Chat History

• Input Variable: "User Query"

Classes

• CLASS 1: Questions about Workfx Introduction

• CLASS 2: Questions having nothing to do with Workfx Introduction

Tool Node

Tool Metadata

• Name: Knowledge Base Search Tool for Workfx

• Description: Introduction KB Search Tool

Tool Selection

• Selected Tool: Workfx Introduction KB Search Tool

Tool Parameters

• Query Prompt: </> system.query

the user's original input (User Query) is passed directly to the knowledge search tool as the query parameter, without any additional processing or intermediate transformation.

Code Node

Configuration

• Input Variables: None (the node operates independently)

• Output Field: result

• Function: Outputs a fixed message when the question is unrelated to WorkFX, informing the user that the system cannot respond.

Output Code Logic

def main() -> dict:
    return {
        "result": "Sorry, this question is not related to WorkFX, so I can't answer it."
    }

LLM Node

Prompt Template

This prompt guides the LLM to generate an answer based on the retrieved content from the tool:

{{#tool-1751255820523.text#}}
Here is some information about WorkFX.
Based on this information, please answer the following question.
{{#system.query#}}

• Line 1: Inserts the content returned by the knowledge search tool (text field)

• Line 2: Provides context that the answer should be based on that information

• Line 3: Inserts the original user query (system.query)

Output Variable

• final_answer (or the default output field used by the platform)

Input Variables

• Implicitly bound:

  • tool-1751255820523.text (from the knowledge tool)

  • system.query (user input)

Answer Node

The final answer in this flow comes from one of two sources:

LLM Node Output – For Related Questions

• Source: LLM Node

• Output Field: final_answer

• Condition: Used when the question is related to WorkFX

• Logic:

  • Read the generated answer from the LLM Node (final_answer)

  • Set it as the response output from the End Node

Code Node Output – For Unrelated Questions

• Source: Code Node

• Output Field: result

• Condition: Used when the question is not related to WorkFX

• Logic:

  • Read the predefined message from Code Node (result)

  • Set it as the response output from the End Node