# Conversations

The Conversations tab is available on every published agent — Chatbot, Agentic Flow, and Voice Agent. It provides a full log of every conversation or session the agent has handled, with detail views showing the complete message transcript, KPI results, and session metadata.

Access is permission-gated: `canViewChatAgentConversations`, `canViewFlowAgentConversations`, or `canViewVoiceAgentConversations` depending on agent type.

For Chatbot and Agentic Flow agents, users see only their own conversations unless their role includes **View All Chat Agent Conversations** or **View All Flow Agent Conversations**. Workspace Admins see all conversations.

[![Conversations list](../images/cmd-conversations-list.png)](../images/cmd-conversations-list.png)

## Conversations List

The list view shows one row per conversation or session. The columns and available filters differ slightly between agent types.

**Chatbot**

Columns:

- **Conversation ID** — unique identifier for the conversation session.
- **Status** — `Completed` or `In progress`.
- **Tag** — `user` (live conversation) or `simulation` (triggered from NeoGPT or the Simulations tab). Displayed as a colour-coded badge.
- **Created At** — timestamp when the conversation started.

Filters:

- **Created on** — date range picker to narrow by conversation start date.
- **Tag** — filter to show only `user` or `simulation` conversations, or both.

Search: by Conversation ID.

Additional controls:

- **Refresh** button — reloads the list.
- **Resync Conversation** button — triggers a background process to resync and update conversation status records. Appears when conversations are present. The process may take some time to complete.
- **Delete** (per row, permission-gated with `canDeleteChatAgentConversations`) — permanently removes the conversation record.

**Agentic Flow**

Columns:

- **Conversation ID** — unique identifier for the conversation.
- **Status** — `Completed` or `In Progress`.
- **Duration** — elapsed time in minutes, or `-` if the session is still in progress.
- **Created At** — timestamp when the session started.

Search: by Conversation ID or User ID.

Users with the `canDeleteFlowAgentConversations` permission get a row menu with **Delete**, plus **Mark as Complete** for conversations still in progress.

**Voice Agent**

The Voice Agent Conversations tab follows the same structure as the Agentic Flow tab, with session-level rows showing ID, status, duration, and timestamp.

## Conversation Detail View

Clicking any row opens the full detail view for that conversation.

[![Conversation detail — header card, transcript, and translation](../images/cmd-conversation-detail.png)](../images/cmd-conversation-detail.png)

**Header card**

The top of the detail view shows a summary card with key metadata. Fields vary by agent type:

*Chatbot:*

- Conversation ID
- Status (Completed / In progress)
- Tag (user / simulation)

*Agentic Flow:*

- Conversation ID
- Status (Completed / In Progress)
- Duration (minutes, or In progress)
- Messages (total message count)
- Created At
- Flow ID

A **Delete** action button appears on the header card for users with the appropriate delete permission.

**Left panel — Results and Metadata**

The left panel (40% width) shows structured data extracted during the conversation:

- **Results** (**KPI Results** on Agentic Flow) — if the agent has KPIs configured, the extracted KPI values from this conversation are displayed here as a JSON viewer with copy-to-clipboard support.
- **Session Metadata** (Agentic Flow only) — additional session-level metadata stored alongside the conversation, rendered as a JSON viewer.

If no KPI results or metadata are present, the left panel is empty.

**Right panel — Message Transcript**

The right panel (60% width) shows the full conversation as a chat history — user messages and agent responses in chronological order, each with a timestamp. Messages are rendered in a familiar chat bubble layout with outgoing (agent) and incoming (user) sides distinguished visually.

### Translation

The transcript panel includes a built-in translation feature for reviewing conversations in other languages:

1. Select a **target language** from the language dropdown.
2. Click **Translate** to translate the entire transcript into the selected language.
3. The panel switches to showing the translated version of each message.
4. Click **Translate** again to toggle back to the original.

The source language is detected automatically and displayed once translation completes. Translation is disabled if the conversation has no messages.

> **Note**
>
> Translation is performed on demand per conversation. The original messages are always stored; translation is a view-only feature and does not modify the stored records.

## Best Practices

- Use the **Tag filter** on Chatbot conversations to separate live user traffic from simulation/test runs — simulation conversations are tagged automatically when triggered from NeoGPT or the Simulations tab.
- Check the **Results** panel (**KPI Results** on Agentic Flow) on high-value conversations to verify that structured outputs were extracted correctly, especially after making changes to the agent’s output schema.
- Use the **Resync Conversation** button if conversations appear stuck in `In progress` status longer than expected — this triggers a backend reconciliation pass.
- For systematic quality reviews, filter to `user` tag only and review a sample of recent conversations regularly to catch unexpected agent behaviour early.