Skip to main content

Agent Instance View

Overview​

The Agent Instance View provides administrators and audit users with a comprehensive interface to inspect all details of a conversation between a user and an agent. This view is essential for:

  • Troubleshooting and debugging
  • Cost and token usage analysis
  • Understanding agent behaviour
  • Compliance and audit purposes
  • User support and performance monitoring

How to Access the Agent Instance View​

To view the details of a specific agent instance:

  1. Navigate to the Instance List:

    • Go to the AI Agents List and click the Instances button on the agent card.
    • Here, you will see a table listing all agent instances.

    Agent Card Access

  2. Open the Details Panel:

    • Locate the agent instance you want to inspect.
    • Click the details icon (usually found at the right end of the row) to open the Agent Instance View for that conversation.

    Details Icon

Top Section: Basic Information​

At the top of the screen, three summary cards present key information about the agent execution:

Agent Instance Summary

1. Agent & Model​

  • Agent: Name and version (e.g., "My Agent", v185 Published). The version label (e.g., "Published") indicates if the agent is in production or draft.
  • Model: The AI model used for the conversation (e.g., "GPT-5.4 (Latest) [OpenAI]").

2. Time​

  • Start Date: When the conversation began (with timestamp).
  • End Date: When the conversation ended (with timestamp).
  • Duration: Total elapsed time for the session.
  • Status: Whether the conversation is open or closed.

3. Costs & Tokens​

  • Input: Number of input tokens and associated cost (tracks user prompt size and cost impact).
  • Output: Number of output tokens and associated cost (tracks agent response size and cost impact).
  • Services: Cost for additional services (e.g., image generation, transcription, or other billable events).
  • Total: Total cost and total tokens used for the conversation.
  • View more: Click to see a detailed cost breakdown.

Main Section: Detailed Conversation View​

The lower part of the screen provides in-depth details for each message, helping you analyse agent behaviour and conversation flow.

Detailed Conversation View

4. Chat History​

Displays the chronological exchange between the agent and the user. Each message includes:

  • Sender: Agent or user, with avatar and role label for clarity.
  • Message: Preview of the message content.
  • Timestamp: When the message was sent.

5. Message Details Panel​

The right panel contains several tabs for in-depth analysis of each individual message:

Summary​

  • Date: Timestamp of the selected message or event.
  • Total tokens: Number of tokens used in the message/event.
  • Service cost: Cost attributed to the service for this message/event.
  • Total cost: Cumulative cost up to this point.
  • Duration: Time taken for the agent to respond.
  • Message preview: Content of the selected message.

Summary Tab

Input​

  • Shows a JSON object with the exact API request sent when executing this agent. Useful for debugging and replicating the agent's input.

Input Tab

Chat History (Tab)​

  • Shows a detailed JSON view of the full chat history up to this point in the conversation. Useful for reconstructing the conversation context.

Chat History Tab

Execution Logs​

  • Displays the duration and status of different events in the conversation.

Execution Logs Tab

Skills​

  • Lists all skills executed in this conversation, with access to detailed execution information for each one.

Skills Tab

Skills Details

Knowledge​

  • Shows all relevant sections of knowledge retrieval, including which knowledge sources were accessed and what content was used. See Knowledge Sources.

Knowledge Tab

Knowledge Details

Input Token Usage​

  • For user messages, shows a detailed breakdown of input token consumption for processing that message.

Input Token Usage Tab

Sensitive Data Obfuscation​

This tab displays all sensitive data that has been automatically detected and obfuscated during the conversation. For each detected value, you can see:

  • Original Value: The original string that was redacted.
  • Obfuscated Value: The redacted string as it appears in the conversation.
  • Detection Score: The detected score for the original value.

Note: This tab is only visible to users with the Sensitive Data Obfuscation role.

Use this tab to audit how sensitive information is handled, verify that redaction is working as intended, and support compliance reviews.

Sensitive Data Obfuscation Tab

Guardrails​

The Guardrails tab provides visibility into any policy or safety rules that were triggered during the conversation. When a guardrail violation occurs, this tab will show:

  • Violated Rule: The specific rule or policy that was broken.
  • Description: Details about the nature of the violation.
  • Score/Severity: A score or severity level indicating the impact or risk of the violation.

This information helps administrators and compliance teams monitor for unsafe, inappropriate, or non-compliant behaviour, and take corrective action if needed.

Guardrails Tab

User and Session Metadata​

At the bottom of the view, additional metadata is displayed:

  • User Identifier: The user's email or unique identifier.
  • Logged User: The system account used for the session (may differ from the end user in delegated or support scenarios).
  • IP Address: The user's IP address at the time of the conversation.
  • Origin: The web origin of the session (e.g., https://hub.serenitystar.ai).
  • Referrer: The referring page or application, useful for tracking how the user accessed the conversation.