Prompt Builder

Picture of Akash

Akash

Updated on July 22, 2024

Creating prompts is a crucial aspect of leveraging Gen-AI. GPTfy now offers a user-friendly, structured UI that simplifies prompt creation, minimizing clicks and saving users significant time. The prompt builder allows users to test prompts directly within the same interface, ensuring expectations and results are met. All functionalities are consolidated on one page, reducing maximum hand-holding to the user or customer.

Before You Begin:

Step 1: Create a new prompt

Initially, while creating the prompt, it will ask for five fields:

Users can fill in the information and click on the Save button. A prompt will be created.

  • Prompt Name: Give it a clear and descriptive title.
  • Target Object: Select the object with which you want to use the prompt (e.g., Account).
  • Data Context Mapping: Choose the specific mapping based on your needs.
  • AI Model: Pick the AI Model that will power your prompt (e.g., GPTfy, OpenAI)
  • Type: Select the type of prompt. The available types include JSON and text.

Text Prompt: Ideal for predefined prompts in your own words. Enter the desired prompt command in the text box during creation.

JSON Prompt: Utilize the product’s full potential with predefined actions. Follow this article for creating components.

After creating the Prompt user will see the detailed UI of Prompt Builder.

The prompt builder UI contains three main tabs explained below, which contain different sections:

  • Configuration
  • Grounding
  • Prompt

Step 2: Configure the prompt

  • Configuration: This tab contains the information needed to create the prompt. The user can configure the prompt using these fields. There are four sections in the Configuration tab.
  • Object- Target object of the prompt
  • Type – The prompt like Text/JSON
  • AI Prompt Name – Shows the Prompt name
  • Status – Shows the status of the prompt like
  • Draft/Active
    Description – Description for the prompt
  • Purpose – This is the multi-select picklist that shows the purpose of the prompt
  • Target Field: If the user wants to save the AI response in a specific field on the record, select the field here.
  • Prompt Request Id – This is the auto-generated unique Prompt Request Id
  • Include Files – The checkbox allows the files to be included in the prompt during execution.
  • Allow User Input lets users manually enter commands while running any prompt.
  • Append Timestamp – By enabling this checkbox, the Date and time will be stamped in the Target field.


Prompt Availability: This section provides Visibility conditions and Profiles to restrict some users from running the prompt as required.

Visibility Condition: This field acts like a filter, controlling when a prompt can be used on a specific record.
Click on the pencil icon as shown above to create the visibility condition.

Specify conditions by using query builder, field Condition Visibility auto-populates.

Profile – This is a multi-select picklist by which the user can decide which profile user can access the prompt.

AI Model – This section contains the details regarding the AI Model settings like Name, temp, Top P, Max Output Tokens

  • AI Model—The user can add the AI Model that will power the prompt (e.g., GPTfy, OpenAI).
  • Temperature—This field Controls the balance between creativity and predictability in the output (0-1). Higher values lead to more diverse responses.
  • Top P – This field decides how many possible words the AI considers, influencing the response’s diversity. Higher values lead to more unique results.
  • Max Output Tokens—This field allows the user to set the maximum length of the generated response.

Other Information – This section contains the field ‘How it works.’ This is a conditional field on the Prompt Object. A value will be considered if it is in the “How it Works” field.

This is a rich text field where the admin can add values to help users understand the prompt more.

Step 3: Grounding (Ethical & Content)

Grounding—This tab allows you to add “extra flavor” to your prompt command, making the response more tailored to your specific needs.

  • Examples might include avoiding bias, promoting fairness, or staying on topic.

Sections:

  • Grounding rules: These are multi-select picklist values. When selected and saved, they are appended to the prompt command during activation, influencing the AI’s generation process.

Locale – GPTfy derives Locale based on how the prompt is called. The locale information is appended to the Grounding rules sent to AI.

  • Locale – Automation: – This Locale is if the prompt is running from AI Mass Processing, Flow, REST API
  • Locale – GPTfy Console: – This Locale is if the prompt is ran from GPTfy User Console
 

Step 4: Prompt Command and Quick testing

Prompt – This tab shows the Prompt command and enables users to search the record and run the prompt.

  • Prompt Command—A prompt Command is a text box where the user can enter the command as expected. If the prompt is a text prompt, a prompt command text box will appear, and if the prompt type is JSON, the prompt component list will appear.

Users can search the target object’s record from the lookup below and run the GPTfy for quick testing, even if the prompt is inactive. A security audit record will be created after execution.

Step 5: Save & Activate

Click “Save” to create your prompt. Once saved, a unique “Prompt Request ID” appears for reference.

Click “Activate” to make it available in the GPTfy Console dropdown. This will trigger a validation check. Upon successful validation, your prompt is ready to use!

Editing the Prompt—There are no separate buttons for editing the prompt. When the user deactivates the prompt, all prompt fields become editable. Upon deactivation, the prompt’s status is set to “Draft.”

The user can then Activate the prompt again and make a clone of it by using the ‘Clone with Related’ button.

Step 6: Version Control and Audits

  • Enable the Prompt versioning from the preferences tab if AI Setting to enable the functionality.
 

Version Control – Every activation creates a JSON version stored under Files for tracking. Track changes and revert to previous versions if needed.

  • The “Reinstate” button allows you to restore an older version.
 

Security Audit—The user can check the security audit records beside the Versions tab. Whenever the user runs the prompt, the security audit record is added.