Creating AI Agents
Getting Started with Agent Creation
Creating a new AI agent is the first step to automating customer interactions. This guide walks you through the complete process.Navigation to Create Agent
Path:/agents
→ Click “New Agent” button → Fill out agent creation form
Required Information
Essential Fields
- Agent Name: A unique identifier for the agent (must be unique within your organization)
- Agent Instructions: Detailed behavioral guidelines that define how the agent should respond and behave
Optional Configuration
- Agent Description: Brief description of the agent’s purpose and role
- Initial capabilities and tools: Basic features to enable during creation
- Knowledge collections: Connect existing knowledge bases
- WhatsApp channel connections: Link to specific phone numbers
Step-by-Step Creation Process
1. Basic Information Setup
- Navigate to the agents list page (
/agents
) - Click the “New Agent” button
- Enter a unique agent name
- Provide clear, detailed agent instructions
- Add an optional description
2. Configure Core Capabilities
During creation, you can enable basic capabilities:- Handover to Humans: Allow agent to transfer conversations to human operators
- Chat Guard: Protect against inappropriate content
- Follow-up Messages: Automatically re-engage customers
- Auto-Close: Automatically close inactive conversations
- Unsubscribe Handling: Allow customers to opt out
3. E-commerce Features (Optional)
If your business involves product sales:- Product Recommendations: Enable product suggestions
- Checkout Functionality: Allow direct ordering through chat
- Order Tracking: Provide order status updates
4. Connect Knowledge Collections
- Select existing knowledge collections to connect
- Create new collections if needed
- Collections provide the agent with information to answer customer questions
5. Setup WhatsApp Channels
- Connect specific WhatsApp business numbers
- Ensure proper WABA (WhatsApp Business Account) configuration
- Test connectivity before going live
6. Save and Deploy
- Review all configurations
- Save the agent
- The agent will be created and ready for use
Best Practices for New Agents
Writing Effective Agent Instructions
- Be Specific: Clearly define the agent’s role and responsibilities
- Include Examples: Provide sample responses for common scenarios
- Set Boundaries: Define what the agent should and shouldn’t do
- Use Consistent Tone: Maintain brand voice throughout interactions
- Keep It Updated: Regularly refine based on performance
Initial Configuration Recommendations
- Start Simple: Begin with basic capabilities and add advanced features gradually
- Test Thoroughly: Verify each capability works as expected
- Monitor Performance: Track metrics after deployment
- Gather Feedback: Collect user feedback for improvements
Common Creation Issues
Agent Name Already Exists
Problem: “Agent name must be unique” error Solution: Choose a different, more descriptive nameInstructions Too Vague
Problem: Agent responses are inconsistent or unhelpful Solution: Provide more detailed, specific instructions with examplesMissing Required Fields
Problem: Cannot save agent configuration Solution: Ensure agent name and instructions are providedKnowledge Collection Connection Fails
Problem: Collections don’t connect properly Solution: Verify collection exists and you have proper permissionsPost-Creation Steps
1. Test Your Agent
- Use the “Test Agent” feature in the sandbox
- Try various conversation scenarios
- Verify responses match expectations
2. Monitor Initial Performance
- Check conversation metrics
- Monitor handover rates
- Review customer feedback
3. Iterate and Improve
- Refine instructions based on real interactions
- Add or remove capabilities as needed
- Update knowledge collections with new information
Next Steps
After creating your agent:- Configure Advanced Settings: Explore the Configuration tab for fine-tuning
- Connect Additional Tools: Add marketplace tools for extended functionality
- Setup Performance Monitoring: Establish metrics tracking
- Train Your Team: Ensure human operators understand handover procedures