# Activity & chat logs

> Read every real conversation your agent has had. Filter chat logs by date, channel, feedback, confidence, sentiment, and topic; inspect the full transcript with citations and confidence; turn weak answers into better data; and export logs as CSV or JSON.

Where [Analytics](/docs/agents/analytics) gives you the numbers, **Activity & chat logs** gives you the source material: every real conversation your agent has had. This is where you discover what customers actually ask, catch wrong answers, and find the exact gaps to fix.

> **THE SINGLE BEST HABIT FOR A GREAT AGENT:** Read your chat logs weekly. The questions customers actually ask — and the answers the agent actually gave — reveal more than any amount of guessing about what to add to your knowledge base.

## What's in a log entry

Each conversation in the list shows an at-a-glance summary, and opening it reveals the full transcript. Per conversation you get:

- **Preview** — the opening customer message, so you can scan the list quickly.
- **Channel and status** — where it happened (widget, email, WhatsApp…) and whether the bot handled it, escalated, or resolved.
- **Message count** — how many turns the conversation took.
- **Feedback** — thumbs-up and thumbs-down counts on the agent's answers.
- **Confidence** — a score (high / medium / low) reflecting how well-grounded the answers were, based on the citations and stored per-answer scores.
- **Sentiment and topic** — the conversation's classified mood and subject.

## Filtering logs

The log list supports a rich set of filters so you can zero in on the conversations that matter. Combine them freely.

| Filter | Use it to find |
| --- | --- |
| Date (from / to) | Conversations in a specific window — e.g. since a policy change. |
| Source (channel) | Conversations from one channel, like only the website widget or only email. |
| Feedback | Answers rated positive, negative, or with no rating. |
| Confidence | High-, medium-, or low-confidence conversations — low confidence flags likely gaps. |
| Sentiment | Positive, neutral, or negative conversations. |
| Topic | All conversations about a given subject, like returns or shipping. |
| Search | Free-text search across the conversation content. |

> **THE FASTEST PATH TO A FIX:** Filter to **low confidence** or **thumbs-down**. Those conversations are almost always a missing or unclear source — read a few, find the pattern, and add a Q&A pair.

## Inspecting a conversation

Open any conversation to read the full transcript turn by turn, with the agent's citations visible. This shows you not just *what* the agent said but *what it answered from* — the same grounding you see in the [playground](/docs/agents/playground), now for real traffic.

## Turning logs into better data

Chat logs feed directly into improving the agent:

- **Suggestions** on the Data sources tab automatically surface knowledge gaps drawn from these conversations — low-confidence answers and thumbs-down replies, each with the original question.
- **Improve answer** lets you correct a weak reply and pin it as a high-priority Q&A pair so it's retrieved first next time.

See [Data sources](/docs/agents/data-sources) for both workflows in detail.

## Exporting logs

You can export an agent's chat logs as **CSV** or **JSON** for offline analysis, reporting, or feeding into other tools. The export respects the agent scope and includes the conversation data behind the dashboard view.

> **INFO:** CSV is convenient for spreadsheets and quick pivot tables; JSON preserves the full structured conversation data for programmatic analysis.

## Managing conversations

From a conversation you can hand it off to a human — escalating it into your [Help Desk](/docs/help-desk/overview) as a ticket. And from the agent's danger zone you can wipe all conversations and messages if you need a clean slate (for example, after testing).

> **WIPING CONVERSATIONS IS PERMANENT:** Deleting an agent's conversations removes them and their messages for good, and the data won't appear in analytics or exports afterward. Export first if you might need the history.

## What's next

- [Analytics](/docs/agents/analytics) — The aggregate metrics behind these conversations.
- [Data sources](/docs/agents/data-sources) — Close the gaps your logs reveal with Suggestions and Improve answer.
- [Playground](/docs/agents/playground) — Reproduce and fix a bad answer interactively.
- [Help Desk](/docs/help-desk/overview) — Handle the conversations that escalate to a human.