Voice AI¶
Deploy AI agents that make and receive phone calls. 4Geeks Voice AI uses a native speech-to-speech engine for natural-sounding AI phone agents with real-time conversation capabilities.
Overview¶
Voice AI capabilities:
- Inbound Calls β AI answers incoming calls 24/7
- Outbound Calls β AI makes calls for campaigns, follow-ups, reminders
- Phone Provisioning β Bring your own trunk or get dedicated phone numbers
- Calling Campaigns β Bulk outbound call campaigns with CSV upload
- Custom Voice Cloning β Clone any voice for personalized agent experiences
- Cross-Channel Handoffs β Seamlessly switch between WhatsApp, Voice, and SMS
- Call Analytics β Monitor call duration, outcomes, transcripts, and sentiment
Setting Up Voice AI¶
Step 1: Choose Voice Channel¶
When creating an agent:
- Open the Agent Studio
- At the Channel step, select Phone
- The voice AI configuration options appear
Step 2: Provision a Phone Number¶
- Click Connect Phone Number
- Choose a number:
- Search for a new number β By country code and area code
- Use an existing number β From your account
- Bring Your Own Trunk β Register your own phone number via SIP
- Confirm the number assignment
- The number is linked to your agent
Step 3: Configure Voice Settings¶
| Setting | Description | Options |
|---|---|---|
| Voice | Pre-built or custom voice selection | Eve, Ara, Leo, Rex, Sal, or cloned voices |
| Language | Call language | English, Spanish, Portuguese, French |
| First Message | Opening greeting | Custom message |
| End Call Phrase | How the agent ends calls | Configurable |
| Web Search | Enable real-time web search during calls | On / Off |
Step 4: Configure AI Model¶
| Setting | Description | Default |
|---|---|---|
| Model | Voice AI model selection | 4Geeks Voice Engine (fast) |
| System Prompt | Agent behavior instructions | Auto-generated or custom |
4Geeks Voice Engine¶
4Geeks Voice AI is powered by our proprietary native speech-to-speech engine. Unlike traditional architectures that chain separate speech-to-text, LLM, and text-to-speech services, our engine processes voice conversations end-to-end for lower latency and more natural interactions.
Built-in Voices¶
Five professionally crafted voices are available out of the box:
| Voice | Character | Best For |
|---|---|---|
| Eve | Warm, professional | Receptionists, customer support |
| Ara | Expressive, versatile | Sales, marketing |
| Leo | Confident, articulate | Collections, operations |
| Rex | Deep, authoritative | Compliance, legal notifications |
| Sal | Friendly, upbeat | Appointments, reminders |
Custom Voice Cloning¶
Create custom voices tailored to your brand:
- Navigate to Voice Cloning in the AI Agents section
- Click Record Your Voice
- Read the provided script (or record your own)
- Name your voice and select a language
- Click Clone Voice
- The cloned voice is available for use across all your agents
Cloned voices support English, Spanish, Portuguese, and French.
Phone Number Registration (BYO Trunk)¶
Register your own phone numbers with the platform:
- Navigate to Channels
- Click Register Phone Number
- Provide your SIP trunk details:
- SIP URI β Your telephony provider’s SIP address
- Authentication β IP allowlist or SIP digest credentials
- The number is verified and linked to your workspace
Server-Side Tools¶
Voice agents can execute tools during calls:
| Tool | Description |
|---|---|
| Calendar Booking | Schedule appointments during the call |
| CRM Lookup | Look up customer information |
| CRM Update | Update records after the call |
| Transfer Call | Transfer to a human agent |
| End Call | Politely end the conversation |
| Web Search | Search the web for real-time information |
Tools are configured per agent and executed automatically when the AI detects the need during a conversation.
Outbound Calling Campaigns¶
Create and manage bulk outbound call campaigns.
Creating a Campaign¶
- Navigate to the Campaigns tab in AI Agents
- Click Create Campaign
- Configure:
- Campaign Name β Descriptive name
- Agent β Which voice agent to use
- Phone List β Upload a CSV of phone numbers
- Schedule β When to make calls (immediate or scheduled)
- Concurrency β How many calls simultaneously
- Click Launch Campaign
CSV Format for Phone Lists¶
phone_number,customer_name,notes
+15551234567,John Doe,Follow up on quote
+15559876543,Jane Smith,Appointment reminder
+15555555555,Bob Wilson,Payment reminder
Campaign Dashboard¶
Monitor campaign performance:
| Metric | Description |
|---|---|
| Total Calls | Number of calls in the campaign |
| Completed | Calls that reached a person |
| No Answer | Calls that went to voicemail or weren’t answered |
| Failed | Calls that failed (invalid number, etc.) |
| Average Duration | Mean call length |
| Success Rate | Percentage of completed calls |
Campaign Status¶
| Status | Description |
|---|---|
| Pending | Campaign created, not yet started |
| Running | Calls are being made |
| Paused | Campaign temporarily stopped |
| Completed | All calls have been attempted |
| Failed | Campaign encountered an error |
Campaign Controls¶
- Pause β Stop making new calls (in-progress calls continue)
- Resume β Continue from where the campaign paused
- Cancel β Stop the campaign entirely
- View Logs β See individual call results
Call Logs¶
Each call in a campaign is logged:
| Field | Description |
|---|---|
| Phone Number | Number called |
| Status | Completed, No Answer, Failed |
| Duration | Call length in seconds |
| Transcript | Full conversation transcript |
| Outcome | Result summary |
| Timestamp | When the call was made |
Phone Number Management¶
Viewing Connected Numbers¶
- Navigate to the Channels tab
- View all connected phone numbers:
| Number | Agent | Status | Country |
|---|---|---|---|
| +1 (555) 123-4567 | Sales Agent | Active | US |
| +506 2222-3333 | Receptionist | Active | CR |
Disconnecting a Number¶
- Navigate to Channels
- Find the phone number
- Click Disconnect
- Confirm
The number is released and can be reassigned later.
Call Quality¶
Voice Engine Performance¶
The 4Geeks Voice Engine is optimized for natural, low-latency conversations:
| Metric | Typical Value |
|---|---|
| Total Response Latency | ~800ms |
| Speech Recognition | Real-time, multi-language |
| Voice Synthesis | Natural-sounding, low latency |
| Concurrent Calls | Unlimited (based on plan) |
Call Analytics¶
Every call is tracked with structured analytics:
| Metric | Description |
|---|---|
| First Response Time | How quickly the AI responds |
| Total Tokens | Language model tokens consumed |
| Call Duration | Total call length |
| Cost | Per-call cost tracking |
Cross-Channel Handoffs¶
Voice agents can seamlessly hand off conversations to other channels when needed:
- Voice to WhatsApp β When a customer prefers text chat
- Voice to SMS β When WhatsApp is unavailable
- WhatsApp to Voice β When a customer requests a phone call
Handoffs are triggered automatically when the AI detects channel-switch intent, or can be configured per agent.
Billing¶
Voice AI calls consume tokens from your agent balance:
| Component | Cost |
|---|---|
| Per-minute charge | Varies by destination country |
| AI tokens | Standard token pricing |
| Voice synthesis | Included in per-minute charge |
| Speech recognition | Included in per-minute charge |
Auto-refill ensures your agents never stop mid-call.
Best Practices¶
- Write clear prompts β Voice agents need concise, conversational prompts
- Set expectations β The first message should clearly state it’s an AI
- Test thoroughly β Use the playground before deploying to production
- Monitor transcripts β Review call logs to improve agent performance
- Start small β Begin with a small campaign, then scale up
- Use custom voices β Brand-specific voices improve customer trust
What’s Next¶
- Voice Cloning β Create custom voices for your agents
- Cross-Channel β Orchestrate conversations across channels
- Agent Analytics β Monitor agent performance
- Knowledge Base β Add domain knowledge to voice agents
- Website Analysis β Auto-generate voice agent prompts
Still questions? Ask on Discord or explore tutorials