Prerequisites
Before creating an agent, ensure you have:
- An active Meetzy account with dashboard access
- At least one phone number available in your account
- Basic understanding of your use case (sales, support, etc.)
The Advanced Editor
The Advanced Editor is a three-panel interface for configuring your agent:
| Panel | Description |
|---|
| Left Panel | Navigation between configuration sections |
| Central Editor | Main editing area for system prompt and settings |
| Right Panel | Playground for testing and debugging |
| Button | Description |
|---|
| ← Back | Return to agents list |
| 🏷️ Version Badge | Current version (e.g., v.1, v.2) |
| ✓ Check | Validate configuration |
| ⬇️ Pull | Sync with latest version |
| 🚀 Publish | Deploy to production |
Left Panel
The left sidebar contains all configuration sections:
| Icon | Section | Description |
|---|
| ⚙️ | Settings | Basic agent configuration |
| 👋 | Greetings | Welcome messages for calls |
| 📄 | Input Params | Dynamic variables for personalization |
| 📚 | Knowledge Base | Documents and FAQs the agent can reference |
| 🔧 | Actions & Tools | Integrations and external APIs |
| ✅ | Evaluations | Automated tests to validate agent behavior |
| 📥 | Output Settings | Data the agent should capture |
Step 1: Basic Settings
Click the Settings icon (Cog) to configure:
Agent Name
Internal identifier for your agent. Choose a descriptive name like:
- “Sales Qualification Agent”
- “Customer Support - Billing”
- “Appointment Reminder”
Company
Select which company this agent belongs to (for multi-company accounts).
Twilio Account
Choose the Twilio account for phone calls.
Phone Number
Phone numbers are exclusive resources. Each number can only be assigned to one agent at a time.
- Release a number from another agent
- Request new numbers via the “Request more numbers” link
Language Model (LLM)
Select the AI model that powers your agent’s conversations. Available models:
Anthropic (Claude)
| Model | Features |
|---|
| Claude Haiku 4.5 | Recommended. Ultra Fast. Best for tool calling. |
OpenAI (GPT)
| Model | Features |
|---|
| GPT-4o | Balanced speed and quality |
| GPT-4.1 | Latest full model |
| GPT-4.1 Mini | Faster, cost-effective |
| GPT-4.1 Nano | Fastest, most economical |
Google (Gemini)
| Model | Features |
|---|
| Gemini 3 Flash Preview | Newest. Preview release. |
| Gemini 2.5 Flash | Fast and capable |
| Gemini 2.5 Flash Lite | Lighter, faster version |
| Gemini 2.0 Flash | Stable, fast |
| Gemini 2.0 Flash Lite | Lightweight version |
| Gemini 1.5 Pro | Most capable, higher latency |
| Gemini 1.5 Flash | Ultra Fast. Good balance. |
For most use cases, start with Claude Haiku 4.5 or Gemini 2.5 Flash. They offer the best balance of speed, quality, and cost.
Call Recording
Toggle to enable/disable call recordings. When enabled:
- All calls are recorded and stored
- Recordings are available in the Calls section
- Required for quality assurance and training
Meetzy Foundation
Optional base components that enhance your agent:
| Component | Description |
|---|
| Safety Guardrails | Prevents harmful or inappropriate responses |
| Conversational Aspects | Improves natural conversation flow |
| Agent Work Experience | Pre-trained behaviors for common scenarios |
Step 2: Greetings
Click the Greetings icon (Hand Raised) to configure welcome messages.
Inbound Greeting
The first message when a customer calls your agent:
Hello! Thank you for calling {{company_name}}. How can I help you today?
Outbound Greeting
The first message when your agent calls a customer:
Hi {{customer_name}}, this is {{agent_name}} from {{company_name}}.
I'm calling regarding {{reason}}. Is this a good time to talk?
Templates
Select from pre-built greeting templates:
- Sales Call
- Customer Support
- Appointment Reminder
- Survey Introduction
Dynamic Variables
Use variables to personalize greetings:
| Variable | Example |
|---|
{{customer_name}} | ”John” |
{{company_name}} | ”Acme Corp” |
{{agent_name}} | ”Sarah” |
{{appointment_date}} | ”March 15th at 2pm” |
Step 3: System Prompt
The central editor contains the System Prompt - the core instructions that define your agent’s behavior.
Structure
A good system prompt includes:
- Role Definition: Who the agent is
- Objective: What the agent should accomplish
- Guidelines: How to handle specific situations
- Constraints: What the agent should NOT do
Example
You are a friendly sales qualification agent for Acme Corp.
## Objective
Qualify leads by understanding their needs and budget,
then schedule a demo with a sales representative.
## Guidelines
- Be conversational and friendly
- Ask about their current challenges
- Understand their timeline and budget
- If qualified, offer to schedule a demo
## Constraints
- Never discuss pricing directly
- Don't make promises about features
- Transfer to human if customer is upset
| Parameter | Type | Description |
|---|
phone | String | Customer’s phone number |
name | String | Customer’s name |
email | String | Customer’s email |
company | String | Customer’s company |
lead_source | String | Where the lead came from |
{{param_name}} syntax.
Step 5: Knowledge Base
Click the Knowledge Base icon (Book Stack) to add reference documents.
Upload documents the agent can reference during calls:
- Product FAQs
- Pricing information
- Company policies
- Common objections and responses
The agent searches the knowledge base contextually during conversations. It doesn’t memorize everything - it retrieves relevant information when needed.
| Action | Description |
|---|
| Transfer Call | Route to human agent |
| Send SMS | Send text message to customer |
| Send Email | Send email to customer |
| Calendar Booking | Schedule appointments |
| CRM Update | Update customer records |
| Custom Webhook | Call any external API |
Step 7: Evaluations
Click the Evaluations icon (Check Badge) to set up automated tests.
Create test scenarios to validate your agent:
Scenario: Customer asks about pricing
Expected: Agent should NOT discuss specific prices
Expected: Agent should offer to connect with sales team
Step 8: Output Settings
Click the Output Settings icon (Inbox) to define data extraction.
Specify what information the agent should capture:
| Field | Type | Description |
|---|
qualified | Boolean | Is the lead qualified? |
budget_range | String | Estimated budget |
timeline | String | Purchase timeline |
next_steps | String | Agreed follow-up actions |
pain_points | Array | Customer’s challenges |
Publishing Your Agent
When your configuration is complete:
Validate
Click Check to validate your configuration. Fix any errors.
Test
Use the Playground (right panel) to test conversations.
Make test calls
Use “Init Call” from the agents list to make real phone calls.
Publish
Click Publish to deploy your agent to production.
Publishing creates a new version. The previous version remains available for rollback if needed.
Next Steps
