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

# AI-Driven Dataset Onboarding

> Use the AI Assisted flow in the Create Dataset wizard to describe a source in plain language and auto-fill connection and dataset settings

# AI-driven dataset onboarding

AI Onboarding is an assisted setup flow built into Liberator's **Create Dataset** wizard. After you pick a source type, choose **AI Assisted** to describe your data source in plain language (or paste connection details). The assistant tests the connection and fills the standard wizard steps for you to review before creating.

<Note>
  Available in Liberator **2.3** and later. The flow is labeled **BETA**. Manual setup remains available for every connector.
</Note>

## Who can use it

Admins and Super Admins who can create connections and datasets. AI Assisted is offered for every connector type, including S3, Snowflake, SFTP, SQL Server, MySQL, Oracle, PostgreSQL, OneDrive, SharePoint, Azure Blob, Google Cloud Storage, CIFS, FTPS, Local File, and HubSpot.

## Choose AI Assisted or Manual

1. Go to **Datasets → Add Dataset** (or start from **Connections**).
2. Select a source type.
3. Choose a setup flow:

| Flow                   | When to use it                                                                                                         |
| ---------------------- | ---------------------------------------------------------------------------------------------------------------------- |
| **AI Assisted (BETA)** | Paste credentials and a short description; let Liberator fill connection, config, dataset details, and column metadata |
| **Manual**             | Step through every connection and dataset field yourself                                                               |

## Describe your source

Choosing **AI Assisted** opens a chat panel with a prompt such as **Describe your \<Source Type> Dataset**. You can:

* Paste raw connection details (host, keys, bucket, folder/prefix, table name, and so on)
* Describe the source in plain language
* Include wildcard or prefix file patterns (for example, `reports/mag_7_fake_*.csv`) — patterns are preserved exactly in the folder/prefix field

If you do not include a dataset name, the assistant asks for one before continuing. The dataset name becomes the queryable identifier.

## What the assistant fills in

On a successful connection test, the assistant populates the standard wizard steps:

| Wizard step         | What AI fills                                                                  |
| ------------------- | ------------------------------------------------------------------------------ |
| **Basic Info**      | Dataset name (if provided or confirmed in chat)                                |
| **Connection**      | Credentials, endpoint, bucket/account, and related connection fields           |
| **Configuration**   | Default timezone, detected date/time column(s) and format, sample-rows preview |
| **Dataset Details** | Suggested queryable key                                                        |
| **Column Metadata** | Column type, display name, and an AI-generated description for each column     |
| **Webhooks**        | Left for you to configure if needed                                            |

Every AI-filled field stays editable. Nothing is locked until you click **Create Connection** / **Create Dataset** (or the equivalent continue action) at each step.

## Review before creating

The assistant asks you to review and edit anything that looks off. Typical checks:

* Connection test reported success
* Folder/prefix or table matches the intended source
* Date/time column and format look correct
* Sample rows show the expected data
* Column descriptions and the queryable key are usable as-is or edited

## In-flow help

You can ask clarifying questions about the current wizard (for example, what a translation column is) without restarting onboarding. The assistant stays scoped to connecting and configuring **this** dataset — off-topic or destructive requests (such as deleting other datasets) are declined and the conversation is redirected to the outstanding connection details.

## After creation

The new dataset appears in the **Datasets** list with the name you confirmed and the correct source-type badge. You can still edit connection or dataset settings later through the normal Liberator UI.

## Related

<CardGroup cols={2}>
  <Card title="Column description suggestions" icon="wand-magic-sparkles" href="/administration/column-description-suggestions">
    AI-assisted and frequency-ranked column descriptions
  </Card>

  <Card title="Datasource overview" icon="database" href="/datasource-config/overview">
    Connection + dataset architecture and source types
  </Card>

  <Card title="AI-based reporting" icon="chart-mixed" href="/administration/ai-based-reporting">
    Natural-language usage and entitlement reports
  </Card>

  <Card title="What's new in 2.3" icon="rocket" href="/whats-new/liberator-2.3">
    Liberator 2.3 release notes
  </Card>
</CardGroup>