Login
Updated on November 3, 2026

Prompts

Estimated reading: 7 minutes

A prompt is a set of instructions or input provided to an AI model to perform a specific task. It tells the model what to do and can include context, examples, and expected output requirements to guide the response. Prompts can be designed for tasks such as summarizing documents, extracting information, answering questions, and generating content.

Why Prompts Matter

Well-designed prompts help you:

a. Control Behavior: Specify the desired tone, format, and output structure.
b. Improve Consistency: Reduce variations in model responses.
c. Reuse Logic: Apply the same prompt across multiple workflows and services.
d. Iterate Quickly: Update prompt instructions without modifying application code.

Prompt Management Options

a. Add Prompt: Create and configure a prompt manually by defining its instructions, content, and execution settings.
b. Automation: Create and manage prompts automatically as part of workflows or predefined processes, for example generating prompts from a template as part of a deployment pipeline.

How to Create a New Prompt

The Prompts feature lets you create, configure, version, and manage prompts for use across applications and workflows. Prompts are stored and managed in the Control Plane, the central place where your prompts and their versions are kept.

Step 1: Name

Enter a unique name for the prompt. The name identifies the prompt and its versions. Use forward slashes (/) to organize prompts into a folder-like structure, for example support/ticket_summary.

Step 2: Prompt Type and Content

Select the type of prompt to create.

a. Text Mode

Text Mode creates a prompt from a single text template. It works well for single-task prompts such as summarizing, classifying, or rewriting a piece of text.

Features:

i. Supports dynamic placeholders in the {{variable_name}} format.
ii. Supports inline references to existing Control Plane text prompts.
iii. Lets you reuse prompt templates across applications and workflows.

Example:

Summarize the following content: {{document}}

Here, {{document}} is a placeholder that is replaced with the actual input during execution.

b. Chat Mode

Chat Mode creates structured, multi-message prompts that represent a conversation between different roles. Each message is an individual entry assigned to a specific role, and it provides instructions, user input, example responses, or results from external tools.

i. Message

A message is an individual entry in a chat prompt that is assigned to a specific role. Each message contributes to the conversation by providing instructions, user input, example responses, or results from external tools. Together, the messages define the conversation flow that guides the model’s response.

Supported Message Types

Message Type Purpose Example
System Defines the assistant's role, behavior, tone, and overall instructions. You are a support assistant. Answer politely and keep replies under 100 words.
Developer Specifies application-level instructions, constraints, and response requirements. Always respond in JSON with the keys "summary" and "priority".
User Represents the user's input, request, or question. Summarize the following ticket: {{document}}
Assistant Provides example or previous assistant responses to guide output generation. { "summary": "Login issue after password reset", "priority": "high" }
Tool Contains results returned by external tools, APIs, databases, or integrations for the model to use. { "order_id": "1042", "status": "shipped" }

You can add multiple messages to define the conversation flow.

ii. Placeholders

Placeholders pass dynamic values into a prompt at runtime. They are replaced with the corresponding inputs during execution, which makes prompts reusable and flexible. They work in both Text Mode and Chat Mode.

Examples: {{input}}, {{context}}, {{document}}

Step 3: Add Prompt Reference

Prompt references let you reuse an existing Control Plane prompt inside another prompt, which reduces duplication and simplifies maintenance. Both text and chat prompts can include references, but the referenced prompt must be a text prompt.

To add a prompt reference:

1. Click Add Prompt Reference to open the Add inline Control Plane prompt reference popup.
2. Select an existing prompt from the Prompt Name dropdown.
3. Choose how the referenced prompt is resolved in the Reference By field.
4. Select the required label or version.
5. Review the Tag Preview, then click Insert to add the reference to the current prompt.

Tag Preview: Shows the reference tag that will be inserted into the prompt. This applies to both Label and Version references.

Reference By

Comparison Label Version
Resolves to The prompt version currently assigned to the label (for example, production or latest). One specific, fixed version.
When it changes Updates automatically when the label is moved to a newer version. Never changes, even when newer versions are released.
Best for Always using the current approved prompt. Keeping behavior stable and predictable.

Label example:

@@@Prompt:name=Ticketing_Orchestrator|label=production@@@

This tag references the Ticketing_Orchestrator prompt using the production label. At runtime, it is replaced with the content of the prompt version that has that label.

Note: Only Control Plane text prompts can be referenced inline.

Step 4: Configuration

The Configuration section accepts a valid JSON object that defines execution settings and model-specific parameters, such as response length and randomness.

Example:
{
“temperature”: 0.7,
“max_tokens”: 500
}

i. temperature controls the randomness of the generated response.
ii. max_tokens sets the maximum number of tokens allowed in the output.

The supported parameters and their names depend on the model and execution environment. For example, some models use max_completion_tokens instead of max_tokens.

Step 5: Commit Message

The Commit Message describes the changes made in the current prompt version. Clear commit messages improve traceability and help you track revisions over time.

Example: Updated prompt instructions to improve response quality.

After You Commit

Committing saves your changes as a new version of the prompt. You can then assign a label such as production to that version, or reference it directly by version number.

Show Filters

After creating and saving a prompt, use the Show Filters option on the Prompts page to find prompts that match specific criteria. Filters can be applied individually or combined to narrow the list of displayed prompts.

Type: Use the Type filter to display prompts based on their type.

Mode: Select one of the following modes:
Select: Choose prompt types from a predefined list.
Text: Enter prompt types manually and apply text-based filtering.

Select Mode: When Select is selected, the available prompt types appear as checkboxes:
Text: Displays text prompts.
Chat: Displays chat prompts.

Select one or both options:

a. Selecting Text displays only text prompts.
b. Selecting Chat displays only chat prompts.
c. Selecting both options displays all prompts.
d. Clearing a selected option removes that prompt type from the filter criteria.

Text Mode
When Text is selected, enter a value and specify the filter condition.

Filter Condition
a. Contains: Displays prompts matching the entered value.
b. Does Not Contain: Excludes prompts matching the entered value.

Enter Value: Enter a prompt type and click Add to include it in the filter criteria. Multiple values can be added to refine the results.
Examples:
Contains + text: Displays only text prompts.
Does Not Contain + text: Excludes text prompts from the results.

Labels: Use the Labels filter to display prompts based on their assigned labels.
Available labels include:
a. Latest: Identifies the latest version of a prompt.
b. Production: Identifies the production version of a prompt.

Match: When multiple labels are selected, choose how the filter should be applied:
ALL: Displays only prompts that contain all selected labels.
SOME: Displays prompts that contain at least one of the selected labels.

Tags: Use the Tags filter to display prompts associated with specific tags. Enter a tag value and click Add to include it in the filter criteria. Multiple tags can be added to further refine the results.

Version: Use the Version filter to display prompts within a specific version range.
Min.: Specifies the minimum version number to include.
Max.: Specifies the maximum version number to include.

Enter the values directly or use the slider to select a version range between 1 and 100.

Clear All
Click Clear all to remove all applied filters and restore the complete list of prompts. These filters help locate prompts based on their type, labels, tags, and version range.

Next - Prompt Management Playground
Share this Doc

Prompts

Or copy link

CONTENTS

Powered By Sutherland

Robility Chatbot
Robility Assistant
Online