Skip to main content

Overview

Configuration is where you define your knowledge agent’s personality, behavior, and first impression. These settings shape how your agent communicates and what users expect when they start a conversation. All configuration happens in the Introduction and Sample Questions tabs of the knowledge agent builder.

System Instructions (The Most Important Setting)

System instructions are the “brain” of your knowledge agent. This is where you define:
  • Who the agent is
  • What it does
  • How it should behave
  • When to use tools
  • What boundaries exist
Think of system instructions as the agent’s job description and operating manual combined.

Anatomy of Good System Instructions

Structure:

Example: Research Assistant

Example: Marketing Assistant

Example: Development Assistant

Tips for Writing System Instructions

Be specific about tools:
Define the workflow:
Set clear boundaries:
Use examples in instructions:

Welcome Message

The welcome message is the first thing users see when they open a chat with your agent. It should:
  • Set expectations for what the agent can do
  • Show personality/tone
  • Encourage engagement
  • Include a clear call-to-action

Good Welcome Messages

Research Assistant:
Marketing Assistant:
Development Assistant:

Welcome Message Best Practices

Do:
  • List specific capabilities (not vague “I can help”)
  • Use bullet points for scannability
  • Match your brand tone (professional, friendly, technical, etc.)
  • End with a question to prompt engagement
Don’t:
  • Write long paragraphs
  • Make promises you can’t keep
  • Use overly formal or stiff language (unless that’s your brand)
  • Forget to mention key capabilities

Prompt Hint

The prompt hint appears in the input field as placeholder text. It guides users on what to ask or how to phrase requests.

Effective Prompt Hints

Examples:

Best Practices

Be specific:
Show format:
Match your agent’s purpose:

Sample Questions

Sample questions appear as clickable buttons when users first see your agent. They:
  • Show what the agent can do
  • Provide examples of how to phrase requests
  • Make it easy to get started (no typing needed)
  • Set expectations for complexity

How to Write Sample Questions

Make them specific and actionable:
Show different capabilities: Don’t make all sample questions do the same thing. Showcase variety:
Match real use cases: Use questions you actually expect users to ask:

Sample Questions Format

In the builder, enter one question per line:
These will appear as individual clickable buttons in the chat interface.

Prompt Filtering (Content Moderation)

Prompt filtering helps prevent misuse of your agent. You can set content guidelines that the agent will follow.

When to Use Prompt Filtering

Use filtering when your agent:
  • Is public and could be misused
  • Handles sensitive topics
  • Should stay on-topic
  • Needs to avoid certain subjects

Example Filters

Stay on-topic:
Professional boundaries:
Brand protection:

Where to Add Filtering

Include filtering rules in your System Instructions:

Configuration Templates by Use Case

Personal Clone / Expert Assistant

Collaborative Builder

Domain Expert Tool

Testing Your Configuration

After setting up your configuration:
  1. Start fresh conversation - Test the welcome message
  2. Click sample questions - Verify they work as expected
  3. Test prompt hint - Does it guide users appropriately?
  4. Test system instructions:
    • Does the agent stay in character?
    • Does it use tools when appropriate?
    • Does it follow the workflow you defined?
    • Does it respect boundaries?
  5. Test edge cases:
    • Ask off-topic questions (does filtering work?)
    • Request things outside capabilities (does it admit limitations?)
    • Give ambiguous requests (does it ask clarifying questions?)

Common Configuration Mistakes

Mistake #1: Vague System InstructionsBad: “You are helpful and answer questions.”Good: “You are a research assistant. When users ask you to research topics, first search your knowledge base, then use the Web Research workflow if needed. Present findings in bullet points with sources.”Fix: Be specific about what, when, and how.
Mistake #2: Not Mentioning ToolsBad: System instructions don’t mention the workflow agents you enabled.Good: “Use the ‘Email Sender’ workflow when users ask you to send emails. Use the ‘Data Analyzer’ workflow to process and analyze datasets.”Fix: Explicitly tell the agent when to use each tool.
Mistake #3: Overlong Welcome MessagesBad: 5 paragraphs explaining every feature in detail.Good: 3-4 bullet points highlighting key capabilities + a question.Fix: Users won’t read long welcomes. Keep it scannable.
Mistake #4: Generic Sample QuestionsBad: “Help me with my work” / “Answer questions” / “Do research”Good: “Research top 10 AI companies and their funding rounds” / “Create a content calendar for April”Fix: Make sample questions specific enough to be useful examples.

Next Steps

Now that you understand configuration:

Add Knowledge

Train your agent with domain expertise

Enable Tools

Give your agent action-taking capabilities

Best Practices

Learn advanced prompt engineering techniques

Troubleshooting

Fix common configuration issues
Remember: Configuration is iterative. Start with a basic setup, test with real users, and refine based on what works. Your system instructions will evolve as you learn how users interact with your agent.