---
title: Conversations
description: "Every text agent conversation, with turns, tokens, response time and tool calls, and a detail view that replays the thread turn by turn."
---

<Badge variant="accent">Text agents</Badge>

**Conversations**, under **Observability** in the sidebar, is the text-agent
counterpart of [Call Logs](/operate/call-logs). A conversation is one thread with a
text agent, whether it came from the web widget, a chat channel such as Slack or
WhatsApp, or a scheduled run.

## Statistics

Five cards summarise the list: **Conversations**, **Total Turns**, **Total Tokens**,
**Avg Response** (seconds from a user message to the agent's reply) and
**Avg Turns / Conv**. They cover every conversation in the workspace, whatever
source is selected in the table.

![The Conversations statistics and table.](/media/operate/conversations-table.webp)

## The table

| Column | What it shows |
|---|---|
| **Agent** | The text agent. |
| **Channel** | Where the conversation happened: **Web** for the widget, **Try Chat** for the agent's own **Chat** test, the platform (**Slack**, **Telegram**, **Teams**, **WhatsApp** or **Discord**) for a bot deployment, and **Scheduled** for a scheduled run. |
| **Turns** | User messages answered. |
| **Tokens** | Total tokens across the thread. |
| **Status** | The outcome of the latest turn: **Success**, **Error**, **Aborted**, or **Pending** while a tool call waits for approval. |
| **Started**, **Last activity** | Relative times; hover for the exact date and time. |

The **All sources** selector above the table narrows the list to **Web**,
**Channel** or **Scheduled**. **Refresh** reloads it. The table shows 10
conversations a page, with **Previous**, **Next** and the page numbers in the
footer.

## Conversation detail

Choose a row to open the thread. The title is the agent's name, a link that opens
it in a new tab. Under it: the provider's logo and the model that answered (for
example **gpt-5.4-mini**), the channel, the start time, and the thread id, which you
can copy; a direct-message thread also carries a **DM** badge. **Playground** opens
the same agent's **Chat** test with this thread loaded, so you can continue the
exchange yourself.

![A conversation with its per-turn footer and the Summary and Token usage cards.](/media/operate/conversation-detail.webp)

Every agent reply ends with a footer: response time, tokens as `in / out`, the tools
it called, the model, and the turn's status. A turn waiting for a person to approve
a tool call shows **Awaiting approval**, and the turn that resumes it **Action
approved**. A failed turn shows its error, or **Credits exhausted** when the account
ran out of credits.

Cards on the right sum up the thread:

- **Summary**: **Turns**, **Steps** and **Tool calls**.
- **Token usage**: **Total**, **Input**, **Output** and **Avg / turn**.
- **Redacted PII**: the kinds of personal data removed, when any were.
- **Tools used**: a chart of how often each tool was called, when any were.

## Related

- [Quickstart: your first text agent](/get-started/quickstart-text-agent)
- [Bot deployments](/deploy/bot-deployments/overview)
- [Call logs](/operate/call-logs) for voice agents
- [Usage and history](/account/usage-and-history)