Conversations

Conversations is the operational center of the agent dashboard. It is where you verify whether the agent is actually helping users.

Inbox Workspace

The inbox stays beside the selected conversation, so you can review messages without losing your search, filters, or page. Conversations opens with no selection. Choose a row to open its transcript, or use a conversation's direct link.

On desktop, drag the dividers to resize the inbox and conversation details. Arrow keys also adjust their widths; double-click a divider to restore its default. Reset layout restores both widths and opens the details panel without changing conversations or clearing inbox filters. Layout adjustments last while this workspace is open.

On smaller canvases, details move to a drawer. On mobile, the inbox and transcript appear one at a time; Back to inbox returns to your current search and filters.

What The Inbox Supports

The conversation list includes filters for:

• status
• channel
• assignee
• dates
• message count
• search by customer name, email, phone, or conversation ID

Use All, Waiting, and Mine for daily work. More views include AI-handled, human-owned, closed, empty, and test conversations. Playground and Voice Lab tests stay separate from the customer views. Sort by latest or oldest activity, and use the inbox's page controls to review the complete result set.

Use it to move quickly from "something feels off" to the exact conversations that explain why.

Conversation Detail View

Open a conversation to inspect the full operating context, not just the transcript.

Use the details control beside the conversation heading to open or hide customer context. AI debug information and execution traces remain expandable under the relevant message.

The detail view can include:

• full message history
• channel type and state
• assignment details
• takeover, resume-AI, and close actions
• end-user profile information
• feedback tied to that conversation
• voice session history and recording playback
• tool-call diagnostics from voice sessions
• observe mode for watching without taking over

Status And Channel Clues

The exact labels depend on the conversation state, but the big distinction is simple:

• AI is still handling the conversation
• a human is expected or already active
• the conversation has been closed

Channels also matter. Website text, website voice, WhatsApp, and Messenger can behave differently, so always check the channel before assuming a prompt or retrieval problem.

Common Use Cases

Review a bad answer

Open the conversation and inspect:

• what the user actually asked
• whether retrieval found the right context
• whether an interceptor should have fired
• whether the conversation should have escalated earlier

Simple triage checklist

When something looks wrong, try this order:

1. confirm what the user actually needed
2. check whether the knowledge existed in Sources
3. check whether an Interceptor should have overridden the path
4. check whether Agent Settings made the default behavior too loose or too strict
5. check whether the case really needed Takeover Queue

Handle a live escalation

• claim or open the conversation
• take over as a human
• reply directly during a web-text human takeover assigned to you; attachments use the normal upload and inspection flow
• continue WhatsApp and Messenger replies in their channel rather than in the dashboard
• hand control back to AI only when the user is stable again

Debug voice issues

• open the voice session details
• review the recording if available
• inspect tool calls and latency
• verify whether the issue was speech, retrieval, or tool behavior

Best Practices

• review real conversations every week even if metrics look healthy
• use filters to isolate one failure pattern at a time
• pair conversation review with Feedbacks and Analytics
• update Sources, Interceptors, or Agent Settings only after you understand the failure mode

Related Docs

Takeover Queue

Find escalations and claim conversations needing human attention.

Feedbacks

Review low-rated chats and comments to find quality issues.

Users

End-user profiles, channel identities, history, and metadata.