Editing and Maintaining Prompts

Create and edit AI prompts in PromptOwl - build simple prompts, sequential workflows, or multi-agent supervisors with version control.

This guide explains how to create, edit, version, and maintain prompts in PromptOwl for enterprise users.

---

Overview

Prompts are the core building blocks of your AI workflows in PromptOwl. This guide covers:

• Creating new prompts

• Understanding prompt types (Simple, Sequential, Supervisor)

• Working with blocks and variations

• Managing variables

• Configuring AI models

• Version control and publishing

---

Creating a New Prompt

Starting a New Prompt

1. Click the + New button on the Dashboard

2. Or navigate directly to Create Prompt from the sidebar

Screenshot: Create Prompt Button

Basic Prompt Information

Fill in the foundational details:

Field	Description	Required
Prompt Name	A descriptive name for your prompt	Yes
Description	Brief explanation of what the prompt does	No
Tags	Keywords for organization and search	No

Screenshot: Prompt Header

---

Understanding Prompt Types

PromptOwl supports three types of prompts to handle different complexity levels:

Simple Prompt (Default)

Best for straightforward AI interactions with a single system context.

Use cases:

• Question answering

• Content generation

• Simple analysis tasks

Structure:

• One system context

• Direct user interaction

• Optional RAG (document retrieval)

Screenshot: Simple Prompt

---

Sequential Prompt

Best for multi-step workflows where each step builds on the previous.

Use cases:

• Multi-stage content creation (research → draft → polish)

• Data processing pipelines

• Complex analysis with multiple phases

Structure:

• Multiple blocks executed in order

• Each block can use a different AI model

• Output from one block can feed into the next

Screenshot: Sequential Tab

---

Supervisor Prompt

Best for complex tasks requiring multiple specialized AI agents.

Use cases:

• Multi-agent systems

• Tasks requiring different expertise

• Complex decision-making workflows

Structure:

• One supervisor block orchestrates the workflow

• Multiple agent blocks with specialized roles

• Supervisor decides which agents to invoke

Screenshot: Supervisor Prompt

---

Working with Blocks

Blocks are the building units of Sequential and Supervisor prompts.

Adding a Block

1. Click Add Block at the bottom of the blocks list

2. Enter a block name (e.g., "Research", "Summarize", "Format")

3. Configure the block settings

Screenshot: Add Block

Block Configuration Options

Each block has several configurable options:

Setting	Description
Block Name	Descriptive name for this step
Prompt Source	"Inline" (write here) or "Use Existing" (reference another prompt)
AI Model	Select the model for this block
Tools	Enable tools available to this block
Dataset	Connect documents for RAG
Variables	Map variables specific to this block
Human Message	Optional message shown between blocks

Screenshot: Block Settings

Using Existing Prompts in Blocks

To reuse an existing prompt within a block:

1. Set Prompt Source to "Use Existing"

2. Click Select Prompt

3. Browse and select the prompt to reference

4. Choose the version to use

5. Map any required variables

Screenshot: Select Existing Prompt

---

Working with Variations

Variations allow you to test different prompt versions within the same block.

Creating a Variation

1. Click the + button next to the variations tabs

2. Enter the variation text

3. Optionally name the variation

Screenshot: Add Variation

Switching Between Variations

Click on the variation tab to switch the active version. The active variation is used when:

• Previewing the prompt

• Publishing to production

• Running evaluations

Best Practices for Variations

• Use variations for A/B testing different phrasings

• Keep track of which variation performs best

• Document why each variation was created

---

Managing Variables

Variables make your prompts dynamic and reusable.

Adding a Variable

1. Click Add Variable in the Variables section

2. Enter the variable name (e.g., company_name, tone)

3. Set a default value (optional)

4. Check "Show as question" if users should provide this value

Screenshot: Variable Manager

Using Variables in Prompts

Insert variables using curly braces: {variable_name}

Example:

You are a helpful assistant for {company_name}.
Please respond in a {tone} tone.

Connecting Variables to Documents

To connect a variable to a document or folder:

1. Click Connect Data next to the variable

2. Browse your Data Room

3. Select a document or folder

4. The document content will be injected at runtime

Variable Types

Type	Format	Description
Text	{variable_name}	Simple text replacement
Document	artifact-{id}	Single document content
Folder	folder-{id}	All documents in folder
Block Output	{{block-key}}	Output from previous block (Sequential/Supervisor)

---

Configuring AI Models

Setting the Default Model

1. Click on the Model dropdown in the settings panel

2. Select your preferred AI provider:

   • OpenAI (GPT-4, GPT-4 Turbo, etc.)

   • Anthropic (Claude 3 Opus, Sonnet, Haiku)

   • Google (Gemini Pro, Gemini Ultra)

   • Groq (Llama, Mixtral)

3. Choose the specific model version

Screenshot: Model Selection

Model Settings

Fine-tune model behavior with these settings:

Setting	Description	Range
Temperature	Controls randomness (lower = more focused)	0 - 2
Max Tokens	Maximum response length	Varies by model
Top P	Nucleus sampling threshold	0 - 1

Screenshot: Model Settings

Per-Block Model Configuration

In Sequential and Supervisor prompts, each block can use a different model:

1. Expand the block settings

2. Toggle "Override default model"

3. Select the model for this block

4. Adjust settings as needed

---

Version Control

PromptOwl automatically tracks all changes to your prompts.

Understanding Versions

• Draft Version - Work in progress, not visible to users

• Production Version - The active version users interact with

• Version History - Complete record of all changes

Viewing Version History

1. Open a prompt in edit mode

2. Click the Versions panel on the right

3. Browse all previous versions with:

   • Version number

   • Date created

   • Who created it

   • Change notes

Screenshot: Version Panel

Saving a Draft

Click Save to create a new draft version without publishing. This allows you to:

• Work on changes over multiple sessions

• Test changes before going live

• Collaborate with team members on updates

Publishing a Version

To make your changes live:

1. Click Publish

2. Add change notes describing what's different

3. Confirm the publication

The new version becomes the production version immediately.

Screenshot: Preview Mode

Restoring a Previous Version

To revert to an earlier version:

1. Open the Versions panel

2. Find the version you want to restore

3. Click Publish on that version

4. Confirm the restoration

Note: This creates a new version based on the old one rather than deleting recent versions.

---

Testing Your Prompt

Preview Mode

Test your prompt without affecting production:

1. Click Preview in the editor

2. Enter test values for variables

3. Send test messages

4. Review the responses

Screenshot: Preview Mode

Best Practices for Testing

• Test with various input scenarios

• Try edge cases and unusual requests

• Verify variable substitution works correctly

• Check that document retrieval returns relevant content

---

Prompt Settings

General Settings

Setting	Description
Enable Memory	Remember conversation history
Enable Citations	Show source references for RAG
Enable Tools	Allow AI to use connected tools

Display Settings

Setting	Description
Show in Sidebar	Display prompt in user sidebar
Featured	Highlight on dashboard
Custom Avatar	Set a prompt icon

Screenshot: Prompt Settings

---

Sharing and Permissions

Sharing with Individuals

1. Click the Share button

2. Enter email addresses

3. Select permission level (View, Edit)

4. Click Share

Sharing with Teams

1. Click Share

2. Check "Share with team"

3. Select the team

4. Click Share

Screenshot: Share Modal

---

Maintenance Best Practices

Regular Reviews

• Review prompt performance monthly

• Check annotation feedback for improvement areas

• Update prompts when AI models improve

Documentation

• Use descriptive names and descriptions

• Add change notes when publishing

• Tag prompts for easy organization

Version Management

• Keep production versions stable

• Use drafts for experimental changes

• Document major changes in version notes

---

Troubleshooting

Prompt not responding as expected

1. Check variable values are correct

2. Verify the model settings

3. Review the system context for clarity

4. Test with simpler inputs first

Variables not being replaced

1. Ensure variable names match exactly (case-sensitive)

2. Check for typos in {variable_name} syntax

3. Verify variables have values assigned

Sequential blocks not passing data

1. Use {{block-key}} syntax to reference previous outputs

2. Verify block keys are correct

3. Check that previous blocks are generating output

Model errors

1. Verify your API keys are configured

2. Check if the model is deprecated (warning will show)

3. Try a different model to isolate the issue

---

Related Guides

• Adding and Managing Users

• Adding and Organizing Documents

• Reviewing and Managing Annotations