AI Trace Logs for Tickets and Chats
AI Trace Logs show exactly how a BoldDesk AI Agent produced a reply in a ticket or chat. For every AI-generated response, you can see how the agent understood the customer’s question, which knowledge libraries it consulted, and which skills, API tools, and MCP tools it used, along with the time each step took.
This article explains how to enable trace logging, open a trace from a ticket or chat and read each part of the AI Trace dialog.
Note:
- AI Trace Logs are available for AI 2.0 Agents, which are offered as an add-on for the Enterprise and Momentum (Legacy) plans.
- Traces are recorded only for responses generated after trace logging is enabled for the agent.
What AI Trace Logs Capture
| Information | Description |
|---|---|
| Query interpretation | How the AI Agent understood the customer’s question or request. |
| Knowledge libraries | The libraries (KB articles, files, URLs) the agent consulted. |
| Tool invocations | The skills, API tools, and MCP (Model Context Protocol) tools the agent called. |
| Execution metrics | Total latency and the duration of each processing step. |
| Agent metadata | The agent’s name, version, and the AI model used. |
Benefits of Using AI Trace Logs
For support agents
- Transparency: See the reasoning and sources behind every AI response.
- Quality assurance: Confirm that responses are based on the correct knowledge and logic.
- Faster debugging: Quickly spot when the AI used the wrong information or tool.
For administrators
- Audit trail: Keep a complete record of AI decisions for compliance reviews.
- Performance monitoring: Track agent latency and identify slow steps.
- Configuration validation: Verify that agents use the intended libraries, skills, and tools.
How to Enable Trace Logging for an AI Agent
Trace logging is configured per agent. Enable it before the agent goes live so that every production response has a trace.
- Go to AI module > Agents and open the AI agent you want to trace.
- Open the agent’s Settings and go to Trace Logging.
- Turn on Trace Logging.
- Select the details to capture:
- LLM Tracing
- Tool Tracing
- Click Save.
Note: Enabling trace logging applies only to new responses. Responses generated while trace logging was off cannot be traced later.
How to View AI Trace Logs in a Ticket
- Open the ticket and go to the Messages section.
- Find the comment marked with the AI icon. This icon indicates an AI-authored comment.
- Click the AI icon on the comment.
- The AI Trace dialog opens and shows how the AI Agent generated that comment.
How to View AI Trace Logs in a Chat
- Open the chat conversation and find the AI-generated message.
- Click the More options (⋯) menu on the message.
- Select AI Activity Logs.
- The AI Trace dialog opens and shows the complete trace for that message.
Understanding the AI Trace Dialog
The AI Trace dialog has three parts: the header, the Sources AI Used summary, and the detail tabs.
Dialog Header
- Agent avatar: The AI agent’s initials or icon (for example, SA).
- Metadata: The response latency and timestamp.
Sources AI Used
This summary shows what the AI Agent relied on to generate the response. All items are clickable links—click any library, skill, or tool to view its details.
| Field | What it shows | If nothing was used |
|---|---|---|
| Understood As | The AI’s interpretation of the customer’s query, shown as a quoted, normalized question. | Blank. Query interpretation capture may be turned off for this agent. |
| Library Used | Clickable links to the knowledge libraries the agent consulted. Multiple libraries are listed if more than one was used. | — |
| Skills Executed | Clickable links to the custom AI skills invoked. Click an entry to expand its execution details. | “No skills executed for this response” |
| API Tools Used | Clickable links to the external APIs or API connections called, with their connection details and parameters. | “No API tools used for this response” |
| MCP Tools Used | Clickable links to the MCP tools invoked, with their configurations and responses. | “No MCP tools used for this response” |
Detail Tabs
Below the summary, two tabs provide deeper technical information.
| Tab | What it shows |
|---|---|
| Technical Details | Agent name and version, model name (for example, gpt-5.2), and latency in milliseconds. |
| Execution Data | Each processing step with its type (such as AGENT or MEMORY) and duration, a timeline of the execution flow, and expandable step details. |
Switching Between JSON and Steps Views
Both tabs let you switch between two views:
- Steps: A chronological timeline of processing steps, with each step’s duration, status, and expandable details. Use this view for day-to-day review.
- JSON: The raw OpenTelemetry trace data, including complete span information, attributes, metadata, and full error details. Use this view for in-depth debugging or export.
