Managing Prompts

Learn how to create and manage prompts in the Management Portal.

Overview

Prompts are reusable text templates that define agent behavior, tool instructions, and data pipeline processing. They provide a centralized way to manage the instructions that guide AI model responses.

Accessing Prompts

In the Management Portal sidebar, click Prompts under the Agents section
The prompt list loads, showing all available prompts

Prompt List

The table displays:

Column	Description
Name	Unique identifier for the prompt
Category	Type of prompt (Workflow, Tool, DataPipeline)
Description	Purpose of the prompt
Edit	Settings icon to modify the prompt

Searching and Sorting

Use the search box to filter by name or description
Click column headers to sort
Use pagination controls to navigate large lists

Prompt Categories

Category	Usage
Agent Workflow	Main prompts for agent orchestration, routing, and response generation
Agent Tool	Instructions for specific tools (code interpreter, knowledge search)
Data Pipeline	Prompts used in data processing workflows

Creating a Prompt

Click Create Prompt at the top right of the page
Configure the prompt settings:

Prompt Configuration Fields

Field	Description	Requirements
Prompt Name	Unique identifier	Letters, numbers, dashes, underscores only. No spaces.
Description	Purpose and usage notes	Recommended for discoverability
Category	Type of prompt	Select from dropdown
Prompt Prefix	Main prompt content	The actual instructions for the AI

Click Create Prompt to save

Name Validation

The prompt name validates in real-time:

✔️ Green checkmark: Name is available
❌ Red X: Name is invalid or already taken

Editing Prompts

Locate the prompt in the list
Click the Settings icon (⚙️) in the Edit column
Modify the configuration
Click Save Changes

Note: The prompt name cannot be changed after creation. To rename, create a new prompt and update references.

Writing Effective Prompts

Best Practices

Be specific: Clearly define the expected behavior
Include context: Specify the persona, tone, and constraints
Use examples: Include sample inputs and outputs when helpful
Structure clearly: Use formatting (headers, lists) for complex prompts

Example Workflow Prompt

You are a helpful assistant that provides accurate information about company policies.

Guidelines:
- Answer questions concisely and professionally
- If information is not available, clearly state so
- Cite relevant policy documents when applicable
- Do not make assumptions beyond the provided data

Response Format:
- Start with a direct answer
- Provide supporting details
- Include relevant references

Example Tool Prompt

You are a code interpreter tool. Generate Python code to answer the user's question.

Requirements:
- Write clean, well-commented code
- Handle edge cases appropriately
- Return results in a clear format
- Do not execute dangerous operations

Using Prompts in Agents

Prompts are referenced in agent configurations:

Workflow Main Prompt

Defined directly in the agent's workflow configuration as the primary system prompt.

Additional Workflow Resources

Added via the "Add Workflow Resource" dialog:

Resource Type: Prompt
Resource: Select the prompt
Resource Role: Purpose identifier (e.g., router_prompt, final_prompt)

Tool Prompts

Added via the "Add Tool Resource" dialog when configuring agent tools:

main_prompt: Primary tool instructions
router_prompt: Instructions for tool selection

Common Prompt Roles

Role	Purpose
`main_prompt`	Primary instructions for the workflow or tool
`router_prompt`	Instructions for selecting tools or routing
`files_prompt`	Instructions for file identification
`final_prompt`	Instructions for generating final responses

Access Control

Prompt permissions can be managed for each prompt:

Open the prompt for editing
Click Access Control at the top right
Add or remove role assignments

Permission	Description
`FoundationaLLM.Prompt/prompts/read`	View the prompt
`FoundationaLLM.Prompt/prompts/write`	Edit the prompt

Table of Contents