Agent System Prompts

Comprehensive guide to creating effective system prompts for your AI agents

Agent System Prompts

System prompts are the foundational instructions that define how your AI agent behaves, responds, and interacts with users. They act as the “personality” and “expertise” blueprint for your agent.

Understanding System Prompts

Based on the agent configuration interface shown in the screenshots, system prompts are configured in the Behavior tab of your agent settings.

What Are System Prompts?

System prompts are instructions that:

• Define your agent’s role and expertise
• Set behavioral guidelines and conversation style
• Establish knowledge boundaries and capabilities
• Configure response format and structure
• Integrate with knowledge bases and external tools

Core Components Visible in Interface

The configuration interface shows several key elements:

1. Agent Prompt (Give Instructions To Agent): The main system instruction field
2. Prompt Templates: Pre-built templates for common use cases
3. Knowledge Integration: “Use the following pieces of context to answer the question at the end.”
4. Initial Messages: Welcome messages and conversation starters
5. Suggested Messages: Pre-written prompts to help users engage

Prompt Template System

The interface includes a “Prompt templates” button that provides access to pre-built prompt structures. This helps ensure you start with proven prompt patterns.

Template Categories

Based on the interface structure, templates likely include:

• Knowledge Base Agents: For information retrieval and Q&A
• Customer Support: For handling customer inquiries
• Sales Assistants: For lead qualification and product guidance
• Technical Support: For troubleshooting and technical assistance
• General Assistants: For multi-purpose interactions

Knowledge Base Integration

The system prompt field shows integration with knowledge bases through context instructions:

Use the following pieces of context to answer the question at the end.

This instruction tells the agent how to:

• Utilize available knowledge sources
• Reference specific information when responding
• Maintain accuracy by staying within knowledge boundaries
• Provide contextual, relevant answers

System Prompt Best Practices

1. Clear Role Definition

Start your prompt by clearly defining the agent’s role:

You are a knowledgeable customer support agent for [Company Name]. 
Your role is to help customers with questions about our products and services.

2. Behavioral Guidelines

Set clear expectations for how the agent should behave:

Always be helpful, professional, and friendly. 
If you don't know something, admit it and offer to find help.
Keep responses concise but complete.

3. Knowledge Integration

Based on the interface screenshot showing knowledge context integration:

Use the following pieces of context to answer the question at the end.
If the context doesn't contain relevant information, say so clearly.
Always cite your sources when possible.

4. Response Format

Specify how responses should be structured:

Structure your responses with:
- A brief acknowledgment of the question
- Clear, actionable information
- Next steps or follow-up suggestions when appropriate

5. Boundaries and Limitations

Set clear boundaries for what the agent can and cannot do:

You can help with:
- Product information and features
- Troubleshooting common issues
- Account management guidance

You cannot:
- Process payments or financial transactions
- Access private customer data
- Make policy exceptions

Advanced Prompt Engineering

Contextual Instructions

The interface shows advanced configuration options including:

• MCP Agent Integration: Model Context Protocol for external tool access
• Knowledge Base Selection: Specific knowledge sources to reference
• Voice and Model Settings: How the agent processes and responds

Multi-Modal Considerations

For agents with voice capabilities (as shown in the voice selection screenshots):

When responding:
- Keep verbal responses conversational and natural
- Use appropriate pauses and emphasis
- Adapt language complexity to the user's apparent level
- Be more expressive in tone than in text-only interactions

Real-Time Interaction Prompts

For real-time models (GPT Realtime, GPT‑4o Realtime, GPT Realtime Mini, Gemini 2.5 Flash/Flash Lite, Groq-hosted realtime variants):

In real-time conversations:
- Respond quickly and naturally
- Handle interruptions gracefully
- Use natural speech patterns
- Acknowledge non-verbal cues when possible

Industry-Specific Prompt Examples

Customer Support Agent

You are a customer support specialist for [Company]. Your goal is to resolve 
customer issues efficiently and professionally.

Use the following knowledge base context to provide accurate information.
When you don't have specific information, offer to connect the customer 
with a human specialist.

Always:
- Listen carefully to the customer's concern
- Provide step-by-step solutions
- Follow up to ensure the issue is resolved
- Maintain a helpful, patient tone

Sales Assistant

You are a sales consultant helping potential customers understand our products.
Your role is to qualify leads, answer questions, and guide customers toward 
the right solutions.

Use product information from the knowledge base to provide accurate details.
Focus on understanding customer needs before recommending solutions.

Guidelines:
- Ask qualifying questions to understand requirements
- Present benefits, not just features
- Handle objections with empathy and information
- Guide toward appropriate next steps

Technical Support Specialist

You are a technical support expert specializing in [Product/Service].
Your role is to diagnose problems and provide technical solutions.

Reference technical documentation and troubleshooting guides from the 
knowledge base. For complex issues, gather detailed information before 
suggesting solutions.

Approach:
- Ask diagnostic questions to understand the issue
- Provide step-by-step troubleshooting instructions
- Explain technical concepts in accessible language
- Escalate when issues exceed your knowledge scope

Testing and Optimization

Prompt Testing Strategy

1. Role Consistency: Test if the agent maintains its defined role
2. Knowledge Accuracy: Verify responses align with knowledge base content
3. Boundary Respect: Ensure agent stays within defined limitations
4. Tone Consistency: Check that communication style remains consistent
5. Edge Cases: Test unusual or challenging scenarios

Iterative Improvement

The interface provides real-time feedback (as shown with “Agent preview theme updated”), allowing for iterative refinement:

1. Deploy Initial Prompt: Start with a well-structured prompt
2. Monitor Interactions: Track how users interact with the agent
3. Identify Issues: Look for patterns in misunderstandings or errors
4. Refine Instructions: Update the system prompt based on learnings
5. Test Changes: Verify improvements work as expected

Integration with Other Features

Initial Messages Integration

System prompts work with initial agent messages:

• Welcome Message: ”👋 Hi, how can I help you?”
• Suggested Prompts: Pre-written conversation starters
• Context Setting: Establish the interaction framework immediately

Voice Integration

For voice-enabled agents, system prompts should include:

Voice Interaction Guidelines:
- Speak clearly and at a moderate pace
- Use natural conversational patterns
- Pause appropriately for user responses
- Adapt to the user's communication style

Data Collection Integration

When agents collect user data:

Data Collection Protocol:
- Explain why information is being collected
- Request permission before gathering personal data
- Handle sensitive information appropriately
- Confirm information accuracy before proceeding

Common Pitfalls to Avoid

1. Overly Long Prompts: Keep instructions focused and actionable
2. Vague Instructions: Be specific about expected behaviors
3. Conflicting Guidelines: Ensure all instructions align consistently
4. Ignoring User Context: Consider how users will actually interact
5. Static Thinking: Plan for prompt evolution and improvement

Measuring Prompt Effectiveness

Track key metrics to evaluate prompt success:

• User Satisfaction: How well does the agent meet user needs?
• Task Completion: Does the agent help users accomplish their goals?
• Escalation Rate: How often do conversations require human intervention?
• Conversation Length: Are interactions appropriately efficient?
• Return Usage: Do users come back to interact with the agent?

System prompts are the foundation of effective AI agents. Well-crafted prompts create agents that are helpful, reliable, and aligned with your business objectives while providing excellent user experiences.