System Instructions
Out of the box, AI agents are generalists - they can engage in conversation but lack specific focus or consistent behavior patterns. What if you want your agent to be a dedicated customer service expert who always maintains a friendly tone? Or a technical specialist who follows strict protocols? That's where system instructions come in - they transform general-purpose AI into focused, reliable assistants that consistently follow your rules and guidelines.
What are System Instructions?
Think of system instructions as the "operating manual" for your AI agent - they determine everything from how it processes information to how it responds to requests.
Unlike one-time commands or chat-specific instructions, system instructions create a consistent personality and behavior pattern that persists across all interactions.
How System Instructions Work
When you run an AI agent on MindPal, the system instructions become the system prompt for the AI model. This system prompt is consistently maintained throughout all interactions with the agent, whether you're:
- Chatting with the agent individually
- Running the agent as part of a multi-agent workflow
- Using the agent across different sessions
Configuring System Instructions
In MindPal, system instructions are divided into two main sections:
1. Background
This section defines who your agent is and what it is supposed to do. Include information such as:
- Role and identity: e.g., "You are an experienced content strategist"
- Behavioral guidelines: e.g., "You communicate in a friendly yet professional manner"
- Task parameters: e.g., "You always start by understanding the target audience"
- Knowledge boundaries: e.g., "You have deep knowledge of SEO best practices"
- Tool usage guidelines: e.g., "You must use both Tavily and Exa Search tools at once" or "You must break down your search with Tavily tool in multiple search queries"
2. Desired Output Format
This section specifies how your agent should structure its responses. You can define:
- Response structure: How information should be organized
- Required components: What must be included in each response
- Formatting preferences: How to present different types of information
- Specific templates: Any standardized formats to follow
- Output constraints: Any limitations on response length or content
Best Practices
Do's ✓
- Be specific: Define clear, unambiguous instructions
- Stay focused: Keep the agent's purpose narrow and well-defined
- Include examples: Provide sample scenarios and desired responses
- Set boundaries: Clearly outline what the agent should not do
- Update regularly: Refine instructions based on performance
Don'ts ✗
- Avoid contradictions: Don't give conflicting instructions
- Don't be vague: Unclear instructions lead to inconsistent performance
- Don't overload: Too many instructions can confuse the agent
- Don't assume knowledge: Explicitly state required information
- Don't skip testing: Always verify how instructions affect behavior
Remember: Well-crafted system instructions are the foundation of a high-performing AI agent. Take time to design, test, and refine them for optimal results.