Creating Your First Agent
Complete guide to creating a production-ready VOISA AI agent
Creating Your First Agent - Complete Guide
This comprehensive guide walks you through creating a production-ready VOISA AI agent from scratch.
📋 Before You Begin
Prerequisites
- Active VOISA AI account
- Clear understanding of your business needs
- Basic information about your services/products
- Sample FAQs from your customers
Preparation Checklist
- Business hours and timezone
- Services/products list
- Common customer questions
- Preferred communication style
- Any existing scripts or guidelines
🎯 Agent Creation Process
Phase 1: Planning Your Agent
Define Your Agent's Purpose
Before creating, answer these questions:
- Primary Goal: What should the agent accomplish?
- Target Audience: Who will interact with it?
- Key Tasks: What specific actions should it handle?
- Success Metrics: How will you measure success?
Example Use Cases
- Customer Support: Answer FAQs, provide information
- Appointment Booking: Schedule meetings, manage calendar
- Order Taking: Process orders, check availability
- Lead Qualification: Gather information, qualify prospects
Phase 2: Creating the Agent
Step 1: Navigate to Agent Creation
- Log in to voisa.ai/admin
- Click "Agents" in the left sidebar
- Click "+ Create New Agent" button
Step 2: Basic Information
Agent Name
Choose a descriptive internal name:
Good: "Restaurant Reservation Assistant"
Bad: "Agent 1"
Description
Write a clear description for your team:
"Handles restaurant reservations, provides menu information,
and answers questions about dietary restrictions and special events."
Industry Selection
Select your primary industry from:
- Healthcare & Medical
- Hospitality & Food Service
- Retail & E-commerce
- Professional Services
- Real Estate
- Financial Services
- Education & Training
- Beauty & Wellness
- Automotive
- And 25+ more options
Tags (Optional)
Add tags for organization:
reservation, customer-service, german, restaurant
Step 3: Voice Configuration
Selecting the Perfect Voice
-
Browse Available Voices
- Click the voice selector dropdown
- Listen to each voice sample
- Consider your brand personality
-
Voice Selection Criteria
- Gender: Match your brand preference
- Age: Young, middle-aged, or mature
- Tone: Professional, friendly, or casual
- Language: Native German speakers available
-
Recommended Voices by Industry
- Healthcare: Calm, professional, clear
- Restaurants: Warm, friendly, enthusiastic
- Financial: Professional, trustworthy, precise
- Retail: Energetic, helpful, approachable
Voice Settings Configuration
Stability (0.0 - 1.0) Controls voice consistency:
0.5: More expressive, varied0.75: Balanced (recommended)1.0: Very consistent, less natural
Similarity Boost (0.0 - 1.0) How closely to match selected voice:
0.5: Some variation allowed0.75: Close match (recommended)1.0: Exact match attempt
Speed (0.5 - 2.0) Speaking rate:
0.8: Slow, clear (elderly audience)1.0: Normal pace (recommended)1.2: Slightly faster (younger audience)
Speaker Boost
- Enable for clearer audio in noisy environments
- Disable for natural sound in quiet settings
Step 4: Conversation Design
First Message (Greeting)
Create a welcoming, informative greeting:
Template:
"[Greeting], [Business Name]. [Offer help]"
Examples:
Professional:
"Good day, Smith & Associates Law Firm.
How may I assist you today?"
Friendly:
"Hello! Welcome to Pizza Paradise!
Would you like to place an order or make a reservation?"
Medical:
"Hello, this is Dr. Miller's office.
How can I help you with your appointment or medical inquiry?"
System Prompt Configuration
The system prompt defines your agent's behavior, knowledge, and personality.
Structure:
1. Role Definition
2. Business Information
3. Capabilities
4. Constraints
5. Personality Traits
6. Specific Instructions
Example System Prompt:
You are a professional customer service agent for [Business Name],
a [business type] located in [city].
BUSINESS INFORMATION:
- Hours: Monday-Friday 9am-6pm, Saturday 10am-2pm
- Services: [List main services]
- Location: [Full address]
- Contact: [Email and phone]
YOUR CAPABILITIES:
- Answer questions about our services
- Schedule appointments
- Provide pricing information
- Give directions to our location
YOU CANNOT:
- Make medical diagnoses
- Provide legal advice
- Process payments
- Access personal records
PERSONALITY:
- Be friendly and professional
- Use clear, simple language
- Show empathy and understanding
- Be patient with customers
INSTRUCTIONS:
- Always confirm important information
- Offer to transfer to a human for complex issues
- Thank customers for calling
- End calls politely
Generate vs. Manual Prompt
Using AI Generation:
- Click "Generate System Prompt"
- Describe your business:
"We are a dental practice specializing in family dentistry. We offer cleanings, fillings, crowns, and cosmetic procedures. We accept most insurance and offer payment plans." - Review and customize the generated prompt
Manual Creation:
- Use the template above
- Be specific and comprehensive
- Include edge cases
- Add compliance requirements
Step 5: Model Selection
Recommended Models
Gemini 2.5 Flash Lite ⚡
- Fastest response time
- Good for simple queries
- Latency 400ms to 500ms
- Recommended for: High-volume, simple interactions
Gemini 2.5 Flash ⚡
- Fast response time
- Good for normal queries
- Latency 500ms to 700ms
- Recommended for: High-volume, simple interactions
GPT-4o 🎯
- Balanced performance
- Excellent comprehension
- Versatile
- Recommended for: General use, complex conversations
Claude 4 Sonnet 🏥
- HIPAA compliant
- Excellent reasoning
- Medical-grade security
- Recommended for: Healthcare, legal, financial
- Recommended for Tool Calling
Temperature Setting (0.0 - 1.0) Controls response creativity:
0.0: Very focused, consistent0.3: Professional, predictable0.5: Balanced (recommended)0.7: Creative, varied1.0: Very creative, unpredictable
Step 6: Turn Detection
Configure when the agent should respond:
Turn Detection Mode:
- Silence-based: Waits for pause in speech
- Server VAD: Voice Activity Detection
Turn Timeout (seconds):
5: Quick responses, may interrupt7: Balanced (recommended)10: Patient, allows long pauses
End Call on Silence:
-1: Never auto-end30: End after 30 seconds of silence60: Patient auto-end (recommended)
Step 7: Knowledge Base
Adding Knowledge Sources
Document Upload:
- Click "Upload Documents"
- Select files (PDF, DOC, TXT)
- Supported content:
- Service catalogs
- FAQs
- Policies
- Price lists
URL Content:
- Click "Add URL"
- Enter website URLs
- System extracts content automatically
- Great for:
- About pages
- Service descriptions
- Blog posts
Direct Text Input:
- Click "Add Text"
- Enter information directly
- Format clearly:
BUSINESS HOURS: Monday-Friday: 9:00 AM - 6:00 PM Saturday: 10:00 AM - 2:00 PM Sunday: Closed SERVICES: - General Consultation: €50 - Extended Session: €80 - Emergency Service: €120
Organizing Knowledge
Categories:
- Services
- Pricing
- Policies
- FAQs
- Contact Information
Best Practices:
- Keep information current
- Use clear headings
- Include examples
- Add context
Step 8: Privacy & Compliance
Data Retention Settings
Retention Days:
7: Minimal retention30: Standard (recommended)90: Extended for analysis365: Long-term storage
Privacy Options:
- Zero Retention Mode: No data storage
- Delete Audio: Remove recordings after processing
- Delete Transcripts: Remove text after session
Authentication & Security
Enable Authentication:
- Require login for widget access
- Useful for internal tools
Allowed Hosts: Whitelist domains:
https://yourdomain.com
https://www.yourdomain.com
https://app.yourdomain.com
Step 9: Evaluation & Analytics
Enable Evaluation
Toggle on to track quality metrics
Custom Evaluation Criteria
Add metrics to track:
Customer Satisfaction:
Name: "Customer Satisfaction"
Prompt: "Rate how satisfied the customer seemed
based on their tone and responses (1-10)"
Task Completion:
Name: "Task Completion"
Prompt: "Did the agent successfully complete
the customer's request? (yes/no)"
Data Collection Points
Define what to collect:
Customer Name:
Identifier: customer_name
Type: string
Description: "Customer's full name"
Appointment Type:
Identifier: appointment_type
Type: string
Description: "Type of appointment requested"
Step 10: Testing
Save and Test
- Click "Save as Draft"
- Click "Test Agent"
Testing Methods
Chat Test (Quick)
- Text-based interaction
- Fast iteration
- Good for logic testing
Voice Test (Realistic)
- Actual phone call
- Test voice quality
- Check timing
WebRTC Test (Browser)
- In-browser voice testing
- No phone needed
- Real-time feedback
Test Scenarios
Test these conversations:
-
Basic Greeting
- "Hello"
- "Hi, I need help"
-
Information Requests
- "What are your hours?"
- "Where are you located?"
- "What services do you offer?"
-
Specific Tasks
- "I want to book an appointment"
- "What's the price for [service]?"
- "Do you accept [insurance/payment]?"
-
Edge Cases
- Unclear requests
- Multiple questions
- Off-topic queries
Step 11: Refinement
Based on testing, refine:
- Greeting: Make more welcoming or informative
- System Prompt: Add missing information
- Knowledge Base: Fill gaps
- Voice Settings: Adjust speed or tone
- Turn Detection: Fine-tune timing
Step 12: Activation
When satisfied with testing:
- Review all settings
- Toggle "Active" status
- Click "Save"
- Agent is now live!
🚀 Post-Creation Steps
Widget Deployment
- Click "Get Widget Code"
- Copy the JavaScript snippet
- Add to your website's HTML
Phone Number Setup
- Go to "Numbers" tab
- Purchase or port number
- Assign to your agent
Monitor Performance
- Check "Analytics" regularly
- Review conversation transcripts
- Identify improvement areas
- Update knowledge base
📊 Success Metrics
Track from day one:
- Success Rate: Completed vs. failed calls
- Average Duration: Time per call
- Escalation Rate: Human handoffs
- Customer Satisfaction: Feedback scores
- Task Completion: Successful outcomes
🔧 Optimization Tips
Week 1
- Monitor all conversations
- Note common questions
- Update knowledge base daily
- Refine greeting based on responses
Week 2
- Analyze failure patterns
- Adjust turn detection
- Optimize system prompt
- Add missing information
Month 1
- Review analytics trends
- A/B test settings
- Gather customer feedback
- Plan feature additions
🚨 Common Pitfalls to Avoid
-
Overcomplicating Initial Prompt
- Start simple
- Add complexity gradually
-
Insufficient Testing
- Test all scenarios
- Include edge cases
-
Ignoring Analytics
- Check daily initially
- Act on insights
-
Static Knowledge Base
- Update regularly
- Add new information
🆘 Getting Help
If you need assistance:
- Check Troubleshooting Guide
- Review Best Practices
- Contact support@voisa.ai
⏭️ Next Steps
Congratulations! You've created a professional VOISA AI agent! 🎉