Conversations

The Conversations section gives you a complete, searchable history of every interaction your agent has had. Use it to review quality, debug failures, edit knowledge base content, and escalate to a human operator when needed.

Why review conversations?

• Quality control — read AI responses and ensure accuracy.
• Debug failures — see exactly where and why a conversation went wrong.
• Improve knowledge — edit FAQ answers and document content directly from conversation context.
• Identify patterns — spot recurring questions that should be added to the knowledge base.
• Audit trail — maintain a full record of every user interaction.

Conversation list

The list shows all conversations for the selected agent, with the most recent first.

Column	Description
Date	When the conversation occurred.
Chat ID	Unique identifier for the conversation.
Contact	The user’s display name or external ID.
Origin	Channel source: WhatsApp, Instagram, or Chat.
Status	Current status with color-coded badge.
Actions	Button to open the full conversation detail.

Statuses

Status	Meaning
Completed	Conversation finished successfully.
Failed	An error occurred during the conversation.
Expired	The user session timed out before completing.

Pagination

Conversations are loaded 20 per page. Navigation buttons let you move between pages, and the total count is always visible.

Filtering and search

Combine multiple filters to find specific conversations:

• Chat ID — search for a conversation by its unique identifier.
• Origin — filter by channel (WhatsApp, Instagram, Chat, or All).
• Status — filter by Completed, Failed, Expired, or All.
• Contact — filter to show only conversations from a specific contact (activated from the detail modal).

Changing any filter resets pagination to the first page. Filters persist during your session.

Conversation detail

Click Details on any conversation to open the full detail modal.

Metadata

The top of the modal shows:

• Chat ID — unique conversation identifier.
• Status — current status badge.
• Origin — channel name.
• Occurred At — full date and time.
• Contact — display name and external ID, with a link to navigate to the contact’s profile. You can also click View all conversations to filter the list to only this contact’s history.

Error information

When a conversation has a Failed status, an additional section appears showing:

• Stage — where in the conversation flow the error occurred.
• Error Code — machine-readable error identifier.
• Error Message — full error description.
• Last User Message — the message that triggered the error.

This information helps you understand what went wrong and whether the knowledge base, prompt, or external integration needs adjustment.

Messages

All messages in the conversation are displayed in chronological order, color-coded by type:

Type	Color	Description
User	Blue	Messages sent by the end user.
Assistant	Gray	AI agent responses.
Human Assistant	Purple	Responses from a human operator during handoff. Shows the operator’s name.
Tool	Yellow	Results from tool or function calls (e.g., calendar booking, payment lookup).
RAG Context	Violet	Knowledge base retrieval results showing which documents the agent used to answer.

Knowledge base context (RAG)

When the agent retrieves information from your knowledge base, a RAG Context message shows:

• Query — what the agent searched for.
• Retrieval Time — how long the search took.
• Documents — each retrieved document with:
  • Source file name and page number.
  • Relevance Score — how closely the document matched the query (0–100%).
  • Tags associated with the document.

Inline editing

You can edit knowledge base content directly from the conversation view — no need to leave and find the document manually:

• FAQ documents — edit the question and answer.
• PDF documents — edit the extracted text content.

After saving, a green Edited badge confirms your changes. If a document was modified after the conversation occurred, an amber Updated indicator is shown.

Tip

Inline editing is the fastest way to fix incorrect answers. Read the conversation, see what the agent retrieved, and fix it on the spot.

Escalating to an operator

If a conversation needs human attention, you can escalate it directly from the detail modal:

• Click Escalate to Operator to create a handoff session and notify your operator team.
• If a handoff is already active, a green Active operator session badge is shown instead.
• If the user’s session has expired, the escalation button is replaced with a notice explaining the user disconnected.

Note

Escalation requires an active operator team. See Operators to set up your team and Operator Handoff for handoff configuration.

Best practices

• Review failed conversations regularly — filter by “Failed” status to catch issues early and fix the underlying cause (missing knowledge, prompt gaps, or integration errors).
• Use inline editing — when you spot an incorrect answer, edit the source document right from the conversation. This improves future responses immediately.
• Filter by contact — if a user reports a problem, filter their conversations to see the full history and context.
• Check RAG scores — low relevance scores (below 50%) may indicate the knowledge base needs better content for that topic.
• Monitor handoff patterns — frequent escalations on the same topic suggest the knowledge base or prompt needs improvement.

Next steps

• Operators — set up your team to handle escalated conversations.
• Knowledge Base — add and manage the documents your agent uses to answer questions.
• FAQ — create question-answer pairs for common topics.
• Campaigns — proactively reach out to your contacts.