The Conversations section on the BoldDesk AI Agent Builder page provides a history of interactions processed by an AI 2.0 Agent. It helps administrators review customer questions, AI-generated responses, and the context available during each interaction. Reviewing conversations can help identify unclear instructions, missing or outdated knowledge, unexpected AI tool behaviour, unsupported assumptions, and incomplete fallback responses.

This article explains how to access AI 2.0 Agent conversations, review the available interaction details, and use the findings to improve AI Agent configuration.

Prerequisites

Before reviewing AI 2.0 Agent conversations, ensure that:

• The AI 2.0 Agent has processed at least one interaction.
• Your AI Agent Builder Access is set to Full Access.
• You have access to the applicable AI 2.0 Agent.
• Trace Logging was enabled before the interaction if you need to review detailed execution information.

For permission configuration instructions, explore How to Configure AI Agent Builder Access Permission.

Reviewing AI Agent Conversations

The Conversations section lists interactions processed by the selected AI 2.0 Agent. For each interaction, you can review a message preview, brand, date and time, channel, AI Agent version, customer messages, AI-generated responses, and available trace information.

You can open the Conversations section from the AI Agents page or directly from Agent Builder.

Method 1. Open Conversations from the AI Agents Page

1. Go to AI Center > AI Agents.

2. Locate the required AI 2.0 Agent.

3. Open More options.

4. Select View Conversations.

   
   
   

Method 2. Open Conversations from Agent Builder

1. Open the required AI 2.0 Agent in Agent Builder.

2. Under Channels in the left navigation menu, select Conversations.

   
   
   

Find and Review a Conversation

After opening Conversations:

1. Use Search to locate the required conversation, if necessary.

2. Review the conversation preview, brand, date and time, channel, and AI Agent version.

3. Select a conversation from the list.

4. Review the customer messages and AI-generated responses in the conversation details panel.

   
   
   

Review Conversation Details

A selected conversation can provide the following information:

Review the complete conversation because earlier messages can affect how the AI 2.0 Agent interprets and answers later requests.

Review AI Traces

AI Traces provide execution-level information about an AI-generated response.

To review AI Traces:

1. Open the required conversation.

2. Locate the applicable AI-generated response.

3. Select the Traces icon below the response.

   
   
   

4. Review the Overview section.

5. Expand the required entries under Traces.

6. Close the panel after completing the review.

   
   
   

AI Trace Overview

AI Trace Stages

Delete an AI Agent Conversation

To delete a conversation:

1. Open Conversations for the required AI 2.0 Agent.

2. Select the conversation you want to remove.

3. Click Delete conversation.

   
   
   

4. Review the confirmation message.

5. Confirm the deletion.

   
   
   

Use Conversations and Traces to Improve the AI Agent

Use the conversation history and AI Traces to identify which part of the AI 2.0 Agent configuration requires improvement.

Troubleshooting

AI Traces Are Not Available

If AI Traces are not displayed for a conversation, confirm that:

• Trace Logging was enabled before the interaction occurred.
• The required LLM Tracing or Tool Tracing options were enabled.
• The selected message is an AI-generated response.
• You have access to the applicable AI 2.0 Agent.

An AI Trace Stage Shows a Failure

If a trace stage reports a failure:

1. Open the applicable trace.
2. Expand the failed trace stage.
3. Review the available error details and execution information.

Depending on the failed stage:

If the customer receives an error response or fallback response, review the configured Custom Error Message and Fallback Instruction.

AI Tool Activity Is Not Displayed

If AI Tool activity is not displayed in traces, confirm that:

• The AI 2.0 Agent invoked an AI Tool during the selected interaction.
• The MCP Server or API Action was linked to the AI Agent version that processed the interaction.
• Log Tool Schemas and Log Tool Responses were enabled.
• The interaction occurred after Tool Tracing was enabled.

The AI Agent Returns Unexpected Responses

If the AI Agent generates inaccurate, inconsistent, or unexpected responses:

• Review the conversation history to understand the customer intent and available context.
• Review the configured Instructions for unclear, conflicting, or incomplete guidance.
• Verify that the required knowledge exists in the linked Libraries.
• Review trace details to determine whether the correct knowledge sources and tools were used.
• Confirm that Grounded response only is enabled if responses should be limited to available knowledge.

The AI Agent Cannot Answer Questions That Exist in Libraries

If the AI Agent cannot answer questions that are covered by existing knowledge:

• Verify that the correct Libraries are linked to the AI Agent.
• Review the retrieval-related trace details.
• Increase the Number of Chunks, if appropriate.
• Confirm that the required content exists, is published, and is up to date.
• Review the response and trace information to determine whether the AI Agent retrieved the expected knowledge.

Responses Are Missing Relevant Conversation Context

If the AI Agent appears to ignore earlier messages:

• Review the memory.history.load trace stage.
• Confirm that conversation history was loaded successfully.
• Review the configured History message limit.
• Check whether the required messages fall outside the loaded history range.

AI Responses Are Slow

If AI-generated responses take longer than expected:

• Review the overall trace Duration.
• Review execution times for each trace stage.
• Identify whether delays occur during agent.invoke, tool execution, or knowledge retrieval.
• Review linked tools, MCP Servers, API Actions, and model settings that may increase execution time.

Fallback Responses Occur More Frequently Than Expected

If the AI Agent frequently returns fallback responses:

• Review the conversation and trace details to determine why a response could not be generated.
• Verify that the required knowledge is available in linked Libraries.
• Review AI Tool configuration and tool availability.
• Review the configured Fallback Instruction.
• Check for tool failures, retrieval failures, or instruction conflicts in the trace data.

Frequently Asked Questions

1. Where can I view AI 2.0 Agent conversations?
   Go to AI Center > AI Agents, open More options for the required AI 2.0 Agent, and select View Conversations.

2. Can I review AI execution details for a response?
   Yes. Click the Traces icon below the applicable AI-generated response.

3. What information is available in AI Traces?
   AI Traces can display the execution status, Session ID, Message ID, duration, start and end times, Agent ID, and individual execution stages.

4. What do the trace-stage durations represent?
   Each duration indicates how long the corresponding execution stage took. For example, agent.invoke shows the time spent processing the request and generating the response.

5. How do I apply improvements identified from a conversation?
   Update an editable draft, test the changes in Playground, publish the updated version, verify its channel mapping, and monitor new conversations.

Related Articles

1. How to Test an AI 2.0 Agent in BoldDesk
2. AI Agent Versioning in BoldDesk AI 2.0
3. How to Configure AI Agent Builder Access Permission