Prompt Studio

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.

​

Key Features

• 65+ pre-built templates for common use cases.
• Multi-model comparison and testing (up to 5 models simultaneously).
• Variable support to test multiple scenarios at once.
• AI-assisted prompt generation and synthetic test data creation.
• Version control, draft history, and collaboration.
• Model performance analytics: response time and token usage.

​

Access Prompt Studio

1. Log in to your AI for Process account and select Prompts on the top navigation bar.
2. On the Prompts dashboard, select a tab:
   • All prompts: All available prompts.
   • My prompts: Prompts you created or saved.
   • Shared prompts: Prompts shared with you for use or collaboration.
   Each tab shows the prompt name, prompt text, and creator name.
3. Click New prompt.
4. In the New prompt dialog, enter a name and click Proceed. The Prompt landing page opens.

​

Create a Prompt

​

Generate a Prompt

1. Click Generate a prompt.
2. In the Prompt generator dialog, select a model and enter your instruction.
3. Review the AI-generated prompt.
4. Click Proceed to copy the prompt to the canvas. Customize it as needed.

​

Start from Scratch

​

Prompt Library

1. Click Prompt library.
2. In the Prompt library dialog, select a tab:
   • My templates: Templates you previously saved.
   • All templates: All available templates.
   
3. Click a template to preview it, then click Use template.
4. The template loads on the prompt canvas. Customize it as needed.

​

Work on the Prompt Canvas

Add prompts → Apply variables → Select models → Generate output

​

Add Prompts

1. In the System prompt field, assign a role to the model. This field is optional—use the toggle to enable or disable it.
2. In the Prompt field, enter your instructions. Click Generate Prompt to expand a short instruction into a more detailed prompt.
3. Optionally, define a Response JSON schema to structure model responses. If the selected model supports response formatting, the schema is applied directly. If not, the schema is included with the prompt, and the model responds in the requested format if it can. Without a schema, the model responds in plain text. Supported types: String, Boolean, Number, Integer, Object, Array, Enum, and anyOf. For schema syntax, see Defining JSON schema. Resolve any schema errors before proceeding.

​

Apply Variables

1. In the Prompt field, add variables in double curly braces. For example, {{xyz}}. The Variables column appears automatically.
2. In the Variables window, assign a value to each variable. Click Add an empty row to add multiple rows.
3. Verify that the variables are substituted correctly in the Prompt window.

​

Select Models

1. In the prompt canvas, click the Select Model field and choose a model and connection.
2. To add more models, select from the columns on the right.
3. Click the model settings icon to adjust parameters: temperature, top k, top p, and max tokens.

​

Generate Output

• Model responses.
• Total input and output tokens.
• Time taken to generate each response.

​

Prompt Canvas Options

​

Model Column Options

​

Top Toolbar Options

​

Advanced Features

​

Bookmark a Model

1. Click the Bookmark model with its settings icon on the model column. The icon changes to Model bookmarked with its settings after bookmarking.
2. If you commit without bookmarking, a dialog prompts you to select a preferred model.
3. After committing, click Versions to see the preferred model recorded in the version history.

​

Draft History

​

Regenerate Output

• Fine-tune specific prompts for better quality.
• Compare model performance on the same prompt.
• Adjust prompts to reduce bias.
• Experiment with specific cells while preserving other outputs.

​

Manage Prompt Versions

• Committing: You must generate output before committing. The first commit creates V1; subsequent commits create V2, V3, and so on. Version names are assigned automatically.
• Default version: The latest committed version is the default. To change it, select a version and click Mark as a default version.
• Using versions as drafts: Load any version as a draft to edit without affecting the original. Commit the draft to create a new version.

1. Click Commit on the prompt canvas. The system saves the current prompt, system prompt, and preferred model as a new version.
2. Click Versions to view the version history.
3. Select a version and click Use as a draft to load it on the canvas for editing.

​

Share Prompts

1. Open the prompt and click the three dots icon > Share.
2. In the Share dialog, select users from your account. To add users not in your account, go to Settings.
3. Assign a role and click Share. The system notifies selected users about the shared prompt and their permissions.

​

Prompt Roles and Permissions

​

Prompt API Endpoint

• Each prompt has one API endpoint. By default, it returns the latest version. Set a default version to control which version the endpoint serves.
• A successful API request returns the SystemPrompt and HumanPrompt from the specified version.
• Edit the endpoint’s query parameters to target a specific version. Without a version parameter, it returns the default version.
• You can create multiple API keys per endpoint. Each key can be copied once and deleted but cannot be reused.
• Deleting an API key invalidates it in all external systems where it was used.

​

Best Practices

• Start with templates: Use the prompt library for common use cases to accelerate prompt creation.
• Use variables: Run multiple prompts simultaneously and ensure consistent testing across scenarios.
• Bookmark strong models: Track better-performing models by bookmarking them.
• Use version control: Commit versions for significant changes and log updates clearly.
• Export results as CSV: Save important outputs for sharing and analysis.
• Save successful prompts: Save effective prompts as templates in the prompt library for future use.