Promptly Logo

System Prompts

Learn how to use and customize system prompts in Promptly

System prompts are special instructions that guide how AI models respond to your requests. In Promptly, you can customize these prompts to tailor the AI's behavior to your specific needs.

What Are System Prompts?

A system prompt is a set of instructions sent to the AI model before your actual query. It helps:

  • Define the AI's role and personality
  • Set constraints on responses
  • Specify output formats
  • Provide background context
  • Guide the AI's approach to your queries

Think of system prompts as setting up the rules of engagement before starting a conversation.

Accessing System Prompt Settings

To view or modify system prompts in Promptly:

  1. Open Promptly
  2. Click on the Promptly menu in the menu bar
  3. Select Preferences (or use ⌘,)
  4. Navigate to the System Prompts tab

Default System Prompts

Promptly comes with carefully crafted default system prompts for each model category. These prompts are designed to:

  • Maximize the model's helpfulness
  • Encourage concise, relevant responses
  • Ensure appropriate behavior
  • Guide the model to follow your instructions carefully

While the default prompts work well for most use cases, customizing them can significantly enhance the AI's responses for specific tasks.

Customizing System Prompts

To customize a system prompt:

  1. In Preferences > System Prompts, find the model group you want to modify
  2. Click the edit (pencil) icon
  3. Modify the system prompt text
  4. Click "Save" to apply your changes

You can create different system prompts for each model group (OpenAI, Anthropic, Gemini, Ollama).

Effective System Prompt Techniques

Here are some strategies for writing effective system prompts:

Define a Clear Role

You are an expert software developer specializing in Swift and iOS development.
Provide practical, concise code examples and explain complex concepts in simple terms.

Specify Output Format

Format your responses using Markdown. Use code blocks with appropriate syntax highlighting.
For lists, use bullet points. Keep explanations concise and focused.

Set Constraints

Limit your responses to 100 words or less. Focus on actionable advice rather than
theoretical explanations. Do not include links or references unless specifically requested.

Guide Reasoning Approach

Approach problems step-by-step. First identify the core issues, then consider multiple
solutions before recommending the most appropriate one. Explain your reasoning briefly.

Example System Prompts

For Technical Writing

You are a technical writing assistant. Help create clear, concise documentation.
Use simple language, avoid jargon, and structure content logically.
Include examples where helpful. Format responses in Markdown.

For Creative Writing

You are a creative writing partner. Help generate imaginative, original content.
Embrace vivid descriptions, compelling characters, and engaging dialogue.
Provide varied options when asked, and help develop ideas without taking over the creative process.

For Data Analysis

You are a data analysis expert. Help interpret data, identify patterns, and draw conclusions.
Present insights clearly, acknowledging limitations and uncertainties.
When appropriate, suggest visualizations or further analyses that might yield additional insights.

System Prompts and Context Window

Remember that system prompts consume part of the model's context window (the total amount of text it can consider). Very long system prompts may limit the space available for your actual queries and the AI's responses.

For most use cases, aim to keep system prompts under 500 words.

Resetting to Defaults

If you want to revert to Promptly's default system prompts:

  1. Navigate to Preferences > System Prompts
  2. Click the "Reset to Default" button for the specific model group
  3. Confirm your choice

Best Practices

  • Test your system prompts with various queries to ensure they work as expected
  • Make incremental changes and observe how they affect responses
  • Be specific about what you want, but avoid overly restrictive instructions
  • Consider creating different system prompts for different types of tasks
  • Remember that not all models respond the same way to identical system prompts