Training Flora Ai Agents

Training: Using Debug Mode to Train Better Responses

Debug Mode is a powerful tool that helps admins understand how the chatbot is using the knowledge base and the ChatGPT API to generate answers. By seeing what the system “sees” and sends, you can more effectively train and improve your chatbot’s responses.

What Is Debug Mode?

Debug Mode is an admin-only view that exposes:

• The knowledge base (KB) search results used for each user query

• The data exchanged with the ChatGPT API

This information is visible only to admins and never to end users. It is designed for troubleshooting, analysis, and improving training—not for regular user interaction.

How to Enable Debug Mode

1. Open your chatbot and go to the Details tab.

2. Click Advanced Settings.

3. Scroll to the very bottom of the page.

4. Check the box next to Debug Mode to enable it.

Debug Mode takes effect immediately for admins. There is no impact on what end users see.

Who Can See Debug Information?

• Admins: Can see all debug information, including KB search results and API payloads.

• End users: Cannot see any debug information. Their experience remains unchanged.

This separation ensures that sensitive or technical details are only visible to authorized team members.

What You See in Debug Mode

When Debug Mode is enabled and a user sends a message, the system performs several steps. Debug Mode lets you see key parts of this process.

1. Knowledge Base Search Results

For every user message, the system:

1. Takes the user’s message as a query.

2. Searches the knowledge base for the most relevant content.

3. Returns the top 3 results that:

• Are most similar to the user’s message, and/or

• Provide the most useful context for answering the question.

In the Debug Mode section, these KB results are the first item you will see. Each result typically includes:

• The title or identifier of the KB entry

• A snippet or preview of the content

• A relevance score or ranking (if available)

This helps you verify:

• Whether the right KB articles are being retrieved

• Whether important content is missing or not being surfaced

• Whether your KB structure and wording are aligned with user queries

2. Data Exchanged with the ChatGPT API

Debug Mode also shows the data sent to and received from the ChatGPT API, such as:

• The prompt or messages sent to the model (including the selected KB context)

• The model’s response

• Any system or configuration parameters relevant to the request

This view helps you:

• Understand how KB content is being framed for the model

• See why the model produced a particular answer

• Identify if you need to adjust instructions, tone, or formatting

How Debug Mode Helps You Train Better Responses

By combining KB search visibility with API-level details, Debug Mode supports a more systematic training process:

• Identify gaps in the knowledge base

• If the top 3 KB results are irrelevant or weak, you may need to:

  • Add new articles

  • Update existing content

  • Improve titles, headings, or keywords

• Improve content quality and structure

• Ensure that key information is:

  • Clear and concise

  • Located in dedicated, well-titled articles

  • Written using the same terms your users use

• Refine chatbot behavior

• If the model’s answer is off despite good KB results, consider:

  • Adjusting system instructions or policies

  • Clarifying how the bot should use KB content

  • Adding examples or templates in your configuration

Using Debug Mode regularly while testing common user journeys and edge cases will help you iteratively improve both your knowledge base and your chatbot’s responses.