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

# Advanced Features

Prompt Studio offers advanced capabilities for test data management, collaboration, version control, and API integration.

***

## Test Data

Populate variables in your prompts using imported or AI-generated datasets.

### Import Test Data

Upload a CSV file to populate variable values automatically.

**Requirements:**

* Column names must match variables exactly (case-sensitive): `{{Name}}` requires a column named "Name"
* Data must start from the first row and column
* Maximum 10 rows imported at once

**Steps:**

1. Click **Test data > Import test data**.
2. Upload a CSV file or select from existing datasets.
3. Preview the data and click **Proceed**.

Variables are populated from matching columns and substituted during prompt execution.

### Generate Test Data

Create synthetic datasets using AI—no manual data entry required.

**Limits:** Maximum 5 rows generated at once.

**Steps:**

1. Enter a prompt with variables (for example, `{{Job Title}}`, `{{Experience}}`).
2. Click **Test data > Generate test data**.
3. Select a model, specify row count, and click **Generate**.
4. Preview the dataset and click **Apply value to the variables**.

## Sharing

Share prompts with other users for collaboration. Version history and settings are shared based on commit status.

| Scenario              | What's Shared                                     |
| --------------------- | ------------------------------------------------- |
| Before committing     | Inputs, outputs, settings (no version history)    |
| After committing      | Full version history with all changes             |
| Multiple contributors | Owner designation preserved; all versions visible |

**Steps:**

1. Click the **three dots** icon > **Share**.
2. Select users (must be part of your account).
3. Assign a role and click **Share**.

### Roles and Permissions

| Role                    | Permissions                                                    |
| ----------------------- | -------------------------------------------------------------- |
| **Full (Owner)**        | View, edit, restore, commit, delete; manage users and API keys |
| **Edit (Collaborator)** | All Owner permissions except delete                            |
| **View (Viewer)**       | View prompts and versions only                                 |

***

## Versioning

Track prompt iterations by committing versions. Each commit creates a shareable record of the prompt's evolution.

**Key points:**

* Versions are auto-named sequentially (V1, V2, V3...)
* Must generate output before committing
* Latest committed version is the default (can be changed)
* Use any version as a draft without altering the original
* No limit on number of versions

**Steps:**

1. Generate output, then click **Commit**.
2. Click **Versions** to view all saved versions.
3. Select a version and click **Use as a draft** to edit it.
4. Use **Mark as default version** to set your preferred version.

## Prompt API

Access prompts programmatically via version-specific API endpoints.

**How it works:**

* API endpoint is generated after the first commit
* Calls return both SystemPrompt and HumanPrompt
* If no version specified, returns the default version
* Supports cURL, Python, and Node.js

**API Keys:**

* Create multiple keys per endpoint
* Keys can be copied once and deleted, but not reused
* Deleting a key invalidates all external integrations using it

Access via the **Prompt API** icon on the canvas.

## Bookmark a Model

Mark your preferred model for consistent output. The bookmarked model is saved with committed versions.

**Steps:**

1. Click the **Bookmark model with its settings** icon.
2. If committing without a bookmark, you'll be prompted to select a preferred model.

The preferred model appears in the Versions dialog.

## Draft History

Capture complete canvas snapshots—prompts, variables, and outputs—at different points in time.

**Steps:**

1. Click the **three dots** icon > **Draft history**.
2. View saved drafts with their inputs and outputs.
3. Click **Restore** to revert to a previous state.

| Feature           | What's Saved                                           |
| ----------------- | ------------------------------------------------------ |
| **Draft History** | Full canvas: prompts, variables, and generated outputs |
| **Versions**      | Prompts only (no outputs)                              |

## Regenerate Output

Re-run prompts selectively to refine results without regenerating everything.

**Regeneration levels:**

* **Cell-level**: Regenerate a single output
* **Column-level**: Regenerate all outputs for one model

**Use cases:**

* Fine-tune prompts for better quality
* Compare model performance
* Reduce bias with prompt adjustments
* Preserve good outputs while fixing specific issues