For the complete documentation index, see llms.txt. This page is also available as Markdown.

Conversation Viewer

View complete conversations and their details

Overview

The Conversation Viewer is a page in the MaiAgent admin console where administrators and customer service staff can review complete conversation records more clearly and in greater detail. It lets you focus on every detail of a single conversation, including message content, the timeline, and tool usage records.

The Conversation Viewer is particularly useful for quality monitoring, customer service training, and issue tracking, helping you analyze conversations more effectively.


Open the Conversation Viewer

From the inbox

  1. Open the conversation list

    • Click Customer Service Conversations in the left menu.

    • Click All Conversations.

  2. Select a conversation to view

    • Find the conversation in the list.

    • Click it to show the full conversation on the right.

Conversation content on the right of the All Conversations page: a user message and an AI reply, with the reply marked as having used the knowledge base and completed steps, and the sources and thumbs-up and thumbs-down icons below
Open a conversation: the AI reply shows the knowledge base used, the steps performed, and the reference sources

Interface

Main sections

Section
What it shows

Conversation information panel

Basic conversation information - User name - Conversation platform (LINE / Facebook / Web Chat) - Start time - Conversation status

Message timeline

The complete conversation in chronological order - User messages - AI replies - System messages - Timestamps

Conversation details

Advanced information - AI assistant used - Tool call records - Knowledge base query results - Token usage

Message display

  • User messages: Appear on the left with a white background.

  • AI replies and support staff replies: Appear on the right with a light blue background. The replier (platform name or support staff name) is shown below.

  • System events: Appear in the center in small gray text, such as handoff notices, a support agent taking over, and status changes.

  • Timestamps: The sent time appears below each message.


Advanced features

Tool usage records

When an Agent searches the knowledge base or calls a tool while replying, the reply is marked above with Used knowledge base, Used tool, and Completed N/N steps. Expand the steps to see what each step searched for and what each tool returned.

Knowledge base citation tracking

If an AI reply cites knowledge base content, numbers in the reply mark the citation positions, and the References section below lists the cited files and reference nodes. Click one to see the cited passage.

User thumbs-up and thumbs-down

The thumbs-up or thumbs-down that users give an AI reply in Web Chat, the Agent marketplace, or MaiGPT, along with the reasons they select and any additional text in the “Provide feedback” window after a thumbs-down, are recorded on that reply and can be reviewed here one reply at a time. For the overall ratio, see Usage Analytics: User Satisfaction. For how users do this on their side, see MaiGPT: Give Thumbs-Up, Thumbs-Down, and Feedback on Replies.

Token usage statistics

For conversation-level cost, see the Credit consumption summary in the Details panel: the total, average, highest, and lowest Credits for each AI reply turn, with CSV export. To see the Token count, processing time, and execution steps of a single reply turn, open the conversation detail page in AgentOps. See Automated Evaluation and AI Assistant Monitoring.

Cost statistics help you find unusually expensive replies, so you can then adjust Agent settings or the knowledge base.


Use cases

Scenario 1: Quality monitoring

Need: A supervisor checks conversation quality each day to confirm that AI replies meet standards.

Steps:

  1. Randomly select 10 conversations from the inbox each day.

  2. Review each conversation in the Conversation Viewer.

  3. Check whether:

    • AI replies are accurate.

    • The tone is appropriate.

    • Tools were used correctly.

    • Knowledge base citations are relevant.

  4. Mark conversations that need improvement.

  5. Give feedback to the AI training team.

Scenario 2: Customer service training

Need: Train new customer service staff with real conversation examples.

Steps:

  1. Select strong examples, such as complex issues handled well.

  2. Explain each step using the Conversation Viewer.

  3. Cover how to:

    • Understand customer needs.

    • Use tools effectively.

    • Guide the conversation.

    • Close the conversation properly.

  4. Have new staff practice similar scenarios.

Scenario 3: Troubleshooting

Need: A user reports an incorrect AI reply, and you need to find the cause.

Steps:

  1. Find the conversation using search.

  2. Review it in detail in the Conversation Viewer.

  3. Analyze the issue:

    • Check whether the user's question was understood correctly.

    • Review which tools the AI called.

    • Confirm whether the knowledge base search results were relevant.

    • Review the AI's reasoning process.

  4. Identify the root cause (insufficient knowledge base content / incorrect tool settings / prompt needs improvement).

  5. Make the appropriate changes.


Get conversation question-and-answer records

MaiAgent offers two ways to obtain records, depending on your needs:

Need
Recommended method
Output

Analyze AI question-and-answer records in bulk

Filter and export from “AgentOps > AI Assistant Monitoring”

Excel

Periodically sync AI question-and-answer records to an internal system

Use the question-and-answer records API, and handle pagination and storage yourself

JSON

Bulk export

  1. Open AgentOps > AI Assistant Monitoring in the left menu.

  2. Filter by Agent, member, date, keyword, or other criteria.

  3. Click Export in the upper-right corner to create an export request.

  4. Open AgentOps > Export Requests in the left menu. Once the status is Completed, download the Excel file. The User column shows who created each export.

Each row in the bulk Excel export represents a “user question and AI reply” exchange and is suitable for quality analysis. It is not a complete archive of all messages and original attachments.

Exports currently include only interactive conversations. If you filter the source to Scheduled Execution in AI Assistant Monitoring, the export button is disabled. See Filter Conversation Records by Member and Source.

Retrieve via API

To sync records regularly or use a custom JSON format, call GET /api/v1/records/ with your organization's API Key. Retrieve AI question-and-answer records by page and filter by AI assistant, date, or keyword.

Authenticate with Authorization: Api-Key <API Key> and follow the pagination information in each response until there is no next page. For the full parameters and response fields, see the MaiAgent API documentation.

Handling attachments

  • The bulk Excel export and question-and-answer records API do not include the original files users uploaded as attachments.

  • There is currently no public API for customers to download conversation attachments in bulk. If you need to retain attachments, contact the MaiAgent team to discuss a suitable data delivery method.


FAQ

Q: How is the Conversation Viewer different from a regular conversation window?

A: The Conversation Viewer provides more complete information, including tool usage records, knowledge base citation tracking, and Token usage statistics. It is useful for analysis, monitoring, and training. A regular conversation window is suited to day-to-day replies.

Q: Can I reply directly in the Conversation Viewer?

A: No. The Conversation Viewer is read-only and is intended for review and analysis. To reply, return to the conversation list.

Q: Does the Conversation Viewer show deleted messages?

A: No. It shows only messages that currently exist. Deleted messages cannot be recovered there.

Q: What if the Conversation Viewer loads slowly?

A: Possible causes and solutions:

  • Long conversation: Conversations with more than 100 messages may take longer to load. Please wait a moment.

  • Network issues: Check your network connection.

  • Large attachments: Too many images or files can slow loading.


Last updated

Was this helpful?