# How to use Ask AI for Custom Data?

**Ask AI** on the landing screen opens Instance Agent with your bot context. Describe your goal — the agent helps design field structure (`columns_json`) for a new table or suggests process connection steps.

## When you need this

- You are unsure which fields a directory or request table needs.
- You want a draft structure faster than **From scratch**.
- You need process block hints for an existing table (**Process usage** tab).

## Key terms

- **Ask AI** — option on **New model**; hint: describe the model to Instance Agent for structure help.
- The agent may ask clarifying questions before suggesting fields.
- You must **review and save** in the **Custom Data model editor** — the agent does not replace **Review changes before save**.
- On **Process usage** — separate **Ask AI** buttons to connect the table to a process.
- The old **AI Magic** drawer is **removed** from the UI — use **Ask AI** via Instance Agent instead.
- Requires the new Custom Data UI ([how to enable](./enable-custom-data-ui-v2.md)).

## Before you start

- [x] **Menu → Custom Data**.
- [x] Instance Agent is available on your instance.
- [x] You can describe the task clearly.

## Step-by-step

### New table

1. Open `/templates` or click **Create table**.
2. Choose **Ask AI**.
3. In the agent chat, describe:
   - Table purpose (directory, requests, CSAT).
   - Data to store (fields, types, uniqueness).
   - Links to other tables if needed.
4. Answer the agent’s follow-ups.
5. Apply the suggested structure in the editor — check **Fields**, **Model metadata**.
6. Click **Save** and confirm changes if prompted.

### Existing table + process

1. Open the table → **Edit** → **Process usage** tab.
2. Click **Ask AI** in the **Ask AI** block.
3. Describe what the process should do (write, read, update).
4. Follow agent recommendations in the process builder.

## What happens next

A draft structure or integration plan is ready for your review. After save, the table works like any Custom Data model.

## How to verify

- Editor fields match your task.
- Field types are semantically correct (email, date, association).
- The process saves/reads data in a test dialog.

## Related guides

- [Create a table](./create-model.md)
- [Use the process tab on a table](./use-scenario-tab-on-model.md)
- [Configure field type](./configure-field-type.md)