Agentic Workflow for Splitting Long Responses

Implementing Dynamic JSON Parsing in an Agentic Workflow for Splitting Long Responses

This guide outlines how to build an agentic workflow that dynamically splits a long AI-generated message into smaller, sequentially displayed individual messages using structured JSON parsing.

1. Introduction

This workflow is designed to:

    1.    Generate a long response in a structured JSON format using the OpenAI: Generate text action.

    2.    Split the response into smaller, more digestible parts using the Get Data from JSON action.

    3.    Present the split content as individual, sequential messages for a better user experience.

2. Workflow Structure

An agentic workflow for splitting long messages includes:

    1.    OpenAI: Generate text node

Produces a structured JSON response where each key represents a smaller part of the overall message.

    2.    Get Data from JSON action

Extracts each section of the JSON response and maps it to individual variables.

    3.    Send Messages nodes

Sequentially displays each part of the message to avoid overwhelming the user with one long block of text.

3. Step-by-Step Implementation

Step 1: Configure the OpenAI: Generate Text Node

    1.    Add an OpenAI: Generate text node to your workflow.

    2.    In the Custom Instructions/Prompt section, instruct the AI to format its response as JSON with clearly defined parts. For example:

Please split your response into multiple parts as a JSON object, with keys like "part1", "part2", "part3", etc. Each part should contain concise information. Example format:

{

  "part1": "This is the introduction...",

  "part2": "This is the detailed explanation...",

  "part3": "This is the conclusion..."

}

    3.    Save the AI-generated JSON response to a custom variable, such as {{ai_json}}.

Step 2: Use the Get Data from JSON Action to Extract Sections

    1.    Add the Get Data from JSON action immediately after the OpenAI: Generate text node.

    2.    In the JSON Data field, reference the variable where the JSON response is stored (e.g., {{ai_json}}).

    3.    Map each JSON key to custom variables:

    •    .part1 → MessagePart1

    •    .part2 → MessagePart2

    •    .part3 → MessagePart3

    4.    Add as many mappings as needed to cover all the parts of the JSON response.

Example JSON:

{

  "part1": "Step 1: Understand the basics of nutrition.",

  "part2": "Step 2: Include more vegetables in your meals.",

  "part3": "Step 3: Stay consistent with healthy eating habits."

}

Mapped variables:

    •    {{MessagePart1}} → “Step 1: Understand the basics of nutrition.”

    •    {{MessagePart2}} → “Step 2: Include more vegetables in your meals.”

    •    {{MessagePart3}} → “Step 3: Stay consistent with healthy eating habits.”

Step 3: Display the Extracted Sections as Split Messages

    1.    Add Send Messages nodes to display the extracted sections sequentially.

    2.    Insert each mapped variable into separate messages:

    •    First message: {{MessagePart1}}

    •    Typing delay

    •    Second message: {{MessagePart2}}

    •    Typing delay

    •    Third message: {{MessagePart3}}

This approach ensures that each part of the response is displayed as an individual message, breaking up the long content for clarity.

4. Best Practices

    •    Validate JSON Structure: Ensure the AI consistently generates valid JSON by providing clear and explicit prompts.

    •    Set Typing Delays: Add slight typing delays between messages to make the interaction feel natural.

    •    Handle Varying Lengths: If the number of parts varies, create fallback logic or extra mappings to handle additional or missing sections.

    •    Test Thoroughly: Run tests with varied inputs to ensure the splitting logic works consistently.

5. Example Workflow

User Input:

“Can you explain the steps to improve focus and productivity?”

AI JSON Output:

{

  "part1": "Step 1: Set clear and achievable goals.",

  "part2": "Step 2: Minimize distractions by creating a focused workspace.",

  "part3": "Step 3: Use techniques like the Pomodoro method to stay productive.",

  "part4": "Step 4: Take breaks to avoid burnout and maintain focus."

}

Displayed Output:

    1.    Message 1: Step 1: Set clear and achievable goals.

    2.    Message 2: Step 2: Minimize distractions by creating a focused workspace.

    3.    Message 3: Step 3: Use techniques like the Pomodoro method to stay productive.

    4.    Message 4: Step 4: Take breaks to avoid burnout and maintain focus.

6. Conclusion

This agentic workflow effectively splits long responses into smaller, user-friendly messages, ensuring clarity and engagement. By using structured JSON parsing, you can control how information is presented, delivering a smooth and intuitive user experience. 

To get started quickly, use this free template: Dynamic Message Splitting Workflow Template.