> ## Documentation Index
> Fetch the complete documentation index at: https://docs.roark.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Datasets

> Group calls or chats into named collections for evaluation, comparison, and analysis

## Overview

A **dataset** is a named, project-scoped collection of conversations. Use datasets to group calls or chats you want to evaluate together — a batch of onboarding failures, last quarter's escalations, a golden set for regression testing in [Studio](/documentation/metrics/studio).

Each dataset holds one conversation type — **Calls** or **Chats** — chosen at creation. A dataset is locked to that type; you can't mix modalities in a single dataset.

Datasets live under **Measure → Datasets** in the sidebar.

## Dataset properties

| Property        | Description                                               |
| :-------------- | :-------------------------------------------------------- |
| **Name**        | Required — e.g. "Q1 Support Calls", "Onboarding Failures" |
| **Description** | Optional context for what the dataset contains            |
| **Type**        | Calls or Chats — fixed at creation                        |
| **Count**       | Number of conversations currently in the dataset          |
| **Labels**      | Tags for organizing datasets                              |

<Note>
  Datasets with a **System** badge are managed by Roark and can't be edited or deleted.
</Note>

## Creating a dataset

<Steps>
  <Step title="Open Datasets">
    Go to **Measure → Datasets** and click **New Dataset**.
  </Step>

  <Step title="Pick the conversation type">
    Choose **Calls** or **Chats**. This is permanent — pick based on what you'll be adding.
  </Step>

  <Step title="Name it">
    Add a name and an optional description, then click **Create**.
  </Step>
</Steps>

## Adding conversations

How you add conversations depends on the dataset type:

* **Call datasets** — open the dataset and click **Add Calls** to pick calls from your project.
* **Chat datasets** — add chats from the chat's detail page, not from the dataset panel.

## Managing a dataset

Selecting a dataset from the list opens its detail panel, which shows the name, conversation count, and last-updated time above a table of members. From there you can:

* **Remove conversations** — select rows and bulk-remove them (with confirmation). Removing a conversation from a dataset doesn't delete the conversation itself.
* **Edit** — rename the dataset or update its description.
* **Delete** — remove the dataset entirely.
* **Drill in** — click any row to open that conversation's detail view.

The list page has **All / Calls / Chats** filter pills and a search box to find datasets quickly.

## What's Next

<CardGroup cols={2}>
  <Card title="Studio" icon="flask-conical" href="/documentation/metrics/studio">
    Run a battery of evals against the conversations you've grouped
  </Card>

  <Card title="Collectors" icon="shield-check" href="/documentation/metrics/metric-collectors">
    Run metrics automatically on matching calls and chats
  </Card>
</CardGroup>