Understanding Conversation AI Agent Logs

What Are Conversation AI Agent Logs?

Agent Logs provide a detailed, behind-the-scenes record of how your Conversation AI handled an interaction with a contact. Instead of just seeing the final message that was sent, you can review the entire decision-making process, including the contact's messages, the AI's internal steps, any tools it used, and the timing of each action. This transparency is essential for understanding AI performance, troubleshooting unexpected responses, and ensuring your automated conversations are working as intended.

Key Benefits of Using Agent Logs

Reviewing Agent Logs helps answer critical questions about your AI's behavior, such as:

• What specific path did the AI take to generate a response?
• Did it successfully call the correct tool or action, like checking a calendar or searching a knowledge base?
• What information was passed into an action and what was returned?
• Why might a response have been delayed?
• Did the conversation end or was it handed over to a human agent as expected?

Since AI conversations can follow many different paths based on user input, these logs are invaluable for pinpointing exactly what happened in any given interaction.

Core Concepts in Agent Logs

To effectively use the logs, it helps to understand how activity is organized:

• Conversation: The complete series of messages exchanged between a contact and the AI agent.
• Turn: A single cycle within a conversation: a contact message is received, processed by the AI, and a response is sent.
• Step: An individual action taken by the AI during a turn, such as invoking a tool, parsing information, or generating a response.

How to Access Agent Logs

You can review Conversation AI Agent Logs from several places in your CRM:

1. The Main Agent Logs Page

Navigate to AI Agents > Agent Logs. This is the central hub for reviewing all AI sessions. Here you can view lists of sessions and contacts, apply filters, see performance metrics, and drill into detailed logs.

2. From the Conversations Inbox

When viewing a conversation thread in your inbox that involved the Conversation AI, you can open the related Agent Log directly. This lets you investigate the AI's handling of a specific message without leaving the conversation context.

3. From Contact Records

Agent Logs are accessible from individual contact records and side panels. This is useful when you want to see all AI interactions for a specific person. You can also customize side panel layouts to include the Agent Logs panel for quick access.

Reviewing Logs: Sessions View

The Sessions view presents each AI interaction as a separate row in a list. You can see key details like the AI agent name, contact name, communication channel, timestamp, and status. Clicking on a session opens its detailed log. Use this view when you need to find a specific interaction using filters for time, contact, agent, or channel.

Filtering Sessions

Use the available filters to narrow down the list and find the exact logs you need. You can filter by criteria such as:

• AI Product (e.g., Conversation AI)
• Channel (e.g., SMS, Live Chat)
• Specific Contact
• Date and Time Range
• Specific AI Agent

For example, you could filter to show only Conversation AI sessions from live chat for a particular contact.

Reviewing Logs: Contacts View

The Contacts view organizes AI activity by contact rather than by individual session. This is helpful when you want to start with a person's name and see all their related AI interactions in one place. If a contact has reached out multiple times, this view consolidates that history.

Inside a Detailed Log

Opening a specific session log reveals two primary areas for analysis:

Raw Conversation and Execution Timeline

This section shows the complete message exchange between the contact and the AI, alongside a step-by-step timeline of the AI's internal execution. You can expand any step to see more details.

Reviewing a Specific AI Message

By selecting a particular AI response from the log, you can examine the exact process behind it. This includes:

• Tool and Action Calls: See which tools or actions the AI attempted to use and the data involved.
• Parsed and Raw Views: Review how the AI interpreted the contact's message (parsed intent) alongside the original text.
• Latency and Response Timing: View timestamps to understand how long each part of the process took, which can help identify performance bottlenecks.