Skip to main content

Chats

Monitor and analyze chat interactions between your users and the AI assistant.

Overview​

The Chats page shows all chat sessions between users and the Deway assistant. Use it to:

  • Monitor how users interact with the AI
  • Identify common questions and topics
  • Find gaps in your Knowledge Base
  • Improve Agent Training based on real conversations

Navigate to Chats in the admin sidebar.

Chat List​

The chat list shows all sessions with filtering and search capabilities.

Each Chat Shows​

FieldDescription
User IDThe identified user (from Deway.identify())
TopicAI-detected topic of the conversation
Message CountNumber of messages in the conversation
PagesWhich pages the user was on during the chat
StartedWhen the conversation began
Last ActivityMost recent message timestamp

Filtering​

Filter chats by:

  • Time Period — Last 24 hours, 7 days, 30 days, or custom date range
  • User ID — Search for a specific user
  • Sort — By start time or last activity, ascending or descending

Chat Details​

Click on a chat to open the full conversation in a side panel.

Message Structure​

Each message shows:

FieldDescription
Actoruser or assistant
ContentThe message text (supports markdown rendering)
TimestampWhen the message was sent
IntentWhat the AI identified the user was trying to do
TopicCurrent conversation topic
PageWhich page the user was on when they sent the message

AI Response Details​

For assistant messages, you can also see:

  • Citations — Knowledge Base sources used to generate the response, with links to the original documentation
  • Topical Structure — Whether the conversation topic is stable, gradually shifting, or abruptly shifting

Rating Responses​

Each assistant message has a star rating control. Use it to grade the quality of an AI response and optionally leave a written note. Ratings and comments are stored against the message and visible to other admins on your team.

  • Click a star (1–5) to rate the response
  • An optional comment popover lets you explain the rating
  • Re-click a star to update; clear it to remove the rating

Use ratings to flag responses that need work — they make it easy to come back later and identify training gaps.

Using Chats to Improve​

Identify Knowledge Gaps​

Look for chats where:

  • The AI couldn't answer questions
  • Users asked follow-up clarifying questions
  • Users seemed confused by responses

Action: Upload additional documentation to your Knowledge Base.

Detect Sensitive Topics​

Look for chats where:

  • Users ask about pricing or billing
  • Users express frustration
  • Users need human support

Action: Add Sensitive Topics to handle these appropriately.

Tune Agent Training​

Look for chats where:

  • The agent's tone or persona feels off
  • Sensitive topics weren't caught or were over-triggered

Action: Adjust Agent Training — refine the personality type, sensitive topic definitions, or add training examples.

Best Practices​

  • Review chats weekly — Look for patterns, not just individual messages
  • Focus on gaps — Prioritize chats where the AI fell short
  • Track improvements — Monitor whether changes to training and knowledge base improve responses over time

Next Steps​