Welcome to Edesy's AI Agent Builder! This guide will walk you through creating your first AI agent from scratch. By the end, you'll have a working AI agent ready to handle customer conversations.
What You'll Learn
- Creating your first AI agent
- Configuring AI settings
- Adding knowledge and training
- Deploying to your channels
- Testing and iteration
Time required: 30 minutes
Prerequisites
- An Edesy account (sign up free)
- Basic understanding of what you want your agent to do
- Some content for your knowledge base (FAQs, docs, etc.)
Step 1: Create Your Agent
After logging in, you'll see the dashboard. Let's create your first agent.
-
Click "Create Agent" button
-
Choose a template or start from scratch:
- Customer Support - For handling inquiries
- Sales/Lead Qualification - For capturing and qualifying leads
- Appointment Booking - For scheduling
- Blank - Start from scratch
-
Give your agent a name (e.g., "Support Bot" or "Sales Assistant")
-
Add a description for internal reference
Step 2: Configure Basic Settings
Agent Personality
Define how your agent should communicate:
Personality Options:
├── Professional - Formal, business-like
├── Friendly - Warm, conversational
├── Casual - Relaxed, informal
└── Custom - Define your own tone
Best Practice: Match your brand voice. If your website is casual and fun, your AI should be too.
Language Settings
- Primary Language: The main language your agent speaks
- Additional Languages: Enable multilingual support (50+ available)
- Auto-detect: Let AI detect and respond in user's language
Step 3: Add Your Knowledge Base
Your AI agent's intelligence comes from the knowledge you provide.
Ways to Add Knowledge
-
Upload Documents
- PDF, DOCX, TXT files
- FAQ documents
- Product catalogs
- Policy documents
-
Website Crawl
- Enter your website URL
- AI crawls and extracts relevant content
- Automatically stays updated
-
Manual Entry
- Add Q&A pairs directly
- Perfect for specific responses
- Override AI for critical questions
-
API Integration
- Connect to CRM, helpdesk, e-commerce
- Real-time data access
- Dynamic responses
Knowledge Base Tips
| Do | Don't |
|---|---|
| Use clear, simple language | Use internal jargon |
| Include common variations | Assume one phrasing |
| Keep content current | Let info get stale |
| Structure with headings | Upload walls of text |
Step 4: Configure AI Settings
Model Selection
Choose the AI model that powers your agent:
| Model | Best For | Speed | Cost |
|---|---|---|---|
| GPT-4o | Complex reasoning | Medium | Higher |
| GPT-4o-mini | General use | Fast | Lower |
| Claude 3.5 Sonnet | Nuanced responses | Medium | Medium |
| Gemini 2.0 Flash | Speed-critical | Fastest | Lowest |
Recommendation: Start with GPT-4o-mini for most use cases. Upgrade if needed.
Temperature
Controls response creativity:
- 0.1-0.3: Consistent, predictable (best for support)
- 0.4-0.6: Balanced
- 0.7-1.0: Creative, varied (best for content)
Response Length
- Short: Quick, concise answers
- Medium: Balanced detail
- Long: Comprehensive explanations
Step 5: Set Up Actions
Actions let your AI agent actually do things, not just talk.
Available Actions
Built-in Actions:
├── Send Email
├── Create Ticket
├── Schedule Meeting
├── Look Up Order
├── Update CRM Record
└── Transfer to Human
Custom Actions (via API):
├── Any REST API call
├── Webhook triggers
└── Database operations
Example: Order Lookup Action
- Go to Actions > Add Action
- Select "API Call"
- Configure:
Name: Look Up Order Method: GET URL: https://api.yourstore.com/orders/{order_id} Headers: Authorization: Bearer {{api_key}} - Map response fields to variables
- Create response template
Step 6: Configure Channels
Deploy your agent where your customers are.
Web Chat
- Go to Channels > Web Chat
- Customize appearance (colors, position, avatar)
- Copy embed code
- Add to your website
<!-- Add before </body> -->
<script src="https://cdn.edesy.in/agent-widget.js"></script>
<script>
EdesyAgent.init({ agentId: 'your-agent-id' });
</script>
- Connect your WhatsApp Business account
- Verify phone number
- Set up message templates
- Enable the channel
Other Channels
- Email: Connect inbox, AI responds to emails
- SMS: Via Twilio integration
- Voice: Phone call handling
- Slack/Teams: Internal use
Step 7: Test Your Agent
Before going live, test thoroughly.
Built-in Test Console
- Click "Test Agent" in the dashboard
- Simulate real conversations
- Check knowledge retrieval
- Verify actions work
Testing Checklist
- Common FAQs answered correctly
- Handles unknown questions gracefully
- Actions execute properly
- Escalation works
- Tone matches your brand
- Multi-turn conversations flow naturally
Test Scenarios
1. Happy Path
"What are your business hours?"
→ Should answer from knowledge base
2. Action Required
"Where is my order #12345?"
→ Should look up and respond with status
3. Unknown Question
"Can you help me with quantum physics?"
→ Should gracefully decline or redirect
4. Escalation
"I want to speak to a human"
→ Should transfer or provide contact
Step 8: Go Live
Ready to launch? Here's a safe rollout plan:
Soft Launch (Week 1)
- Enable for 10-20% of traffic
- Monitor all conversations
- Review AI responses daily
- Fix any issues immediately
Expanded Rollout (Week 2-3)
- Increase to 50% of traffic
- Continue monitoring
- Gather customer feedback
- Refine knowledge base
Full Launch (Week 4+)
- 100% traffic
- Establish regular review cadence
- Set up alerts for edge cases
- Continuous improvement
Monitoring and Analytics
Track your agent's performance:
Key Metrics
| Metric | Target | Description |
|---|---|---|
| Resolution Rate | >60% | Issues resolved without human |
| CSAT Score | >4.0 | Customer satisfaction |
| Escalation Rate | <30% | Transferred to humans |
| Response Time | <5s | AI response speed |
| Fallback Rate | <10% | "I don't know" responses |
Dashboard Features
- Real-time conversation monitoring
- Response quality scoring
- Trend analysis
- Export and reporting
Common Issues and Fixes
"AI gives wrong answers"
- Review and update knowledge base
- Add explicit Q&A pairs for problem areas
- Adjust temperature settings
"AI doesn't know enough"
- Add more source documents
- Enable website crawling
- Create manual entries for gaps
"Responses feel robotic"
- Adjust personality settings
- Add example responses in training
- Increase temperature slightly
"Escalation happens too often"
- Expand knowledge coverage
- Add more action capabilities
- Review escalation triggers
Next Steps
Congratulations! Your AI agent is live. Here's what to do next:
- Monitor daily - Review conversations for the first week
- Iterate - Improve based on real interactions
- Expand - Add more channels and capabilities
- Integrate - Connect to more systems
- Scale - Handle more use cases
Resources
- How to Build a Customer Support AI Agent
- AI Agent vs Chatbot: What's the Difference?
- 10 Use Cases for AI Agents in E-commerce
Need Help?
Happy building!