Phone & Voice AI

GoWeBa's Phone module transforms your web browser into a complete business phone system. Through Twilio integration and WebRTC technology, you can make and receive calls, send SMS, and even deploy a Voice AI agent that automatically answers your customers — all without leaving the platform.

Telecom Architecture

GoWeBa uses a hybrid telecom architecture that combines:

• Twilio as the telephony connectivity provider (PSTN)
• WebRTC for calls directly from the browser (no physical phone required)
• Voice AI powered by Twilio Gather + LLM for intelligent automatic responses
• Two-way SMS for text communications
• IVR (Interactive Voice Response) configurable for inbound call routing

GoWeBa Numbers

GoWeBa assigns Canadian phone numbers to each organization:

Numbers are provisioned through Twilio and billed at provider cost × 1.30 (telecom margin).

VoIP Calls — Click-to-Call

Making an Outbound Call

GoWeBa lets you make calls directly from the browser with no physical phone:

1. Click the call button — From a contact card, the inbox, or the dialer
2. Select the source number — Choose from the organization's numbers
3. WebRTC connection — The browser establishes a secure audio connection
4. Communication — Speak directly via your computer's mic/speakers
5. End call — Hang up via the button or close the panel 💡 Use a headset with microphone for optimal audio quality during WebRTC calls.

Receiving an Inbound Call

When a call comes in on a GoWeBa number, the system follows this flow:

1. Inbound call detected — Twilio routes the call to GoWeBa
2. Voice AI enabled? — If yes → the Voice AI agent answers automatically (section 4)
3. IVR configured? — If yes → the voice menu guides the caller
4. Browser notification — The user receives a ring notification in GoWeBa
5. Answer — The user accepts the call — WebRTC connection established
6. Recording — If enabled, the call is recorded with consent

Call Settings

Call Log

All calls are logged in the Inbox → Calls section with the following information:

Post-Call AI Features

Transcription

GoWeBa automatically transcribes recorded calls into numbered segments:

• Segmented transcription: each intervention is a distinct segment with timestamp
• Speaker identification: distinction between CALLER and AGENT (user)
• Final vs. partial text: partial text is displayed in real-time, final after processing
• Detected language: transcription adapts to the conversation's language

Automatic Call Summary

After a call, WEBA can generate an intelligent summary in one click:

1. User clicks "Summary" — From the call log or the call detail card
2. Transcription retrieval — Segments are assembled into full text
3. GPT-4.1 analysis — The AI model analyzes the conversation
4. Summary generation — Key points, decisions made, next steps
5. Saved — The summary is stored with the call for future reference The summary includes: participants, main topic, decisions made, commitments by each party, and recommended follow-up actions.

Action Items

In addition to the summary, WEBA automatically extracts concrete actions mentioned during the call:

• Tasks to create: "I'll send you the quote by Friday"
• Reminders to schedule: "Call me back next week to confirm"
• Documents to prepare: "I'll prepare the service agreement"
• Follow-ups: "Check the conference room availability" 💡 Action items can be converted into GoWeBa tasks in one click, directly from the summary.

Recording and Playback

Recorded calls are stored securely:

• Format: MP3 or WAV depending on configuration
• Storage: Secure cloud storage with encryption at rest
• Access: Play directly in GoWeBa, one-click download
• Retention: Configurable (30, 60, 90 days or unlimited)
• Compliance: Consent warning before recording ⚠️ In Canada (Bill C-36) and Quebec (Civil Code), the consent of at least one party is required for recording a phone conversation.

Voice AI — Intelligent Voice Agent

Concept

Voice AI is GoWeBa's intelligent voice agent that automatically answers incoming calls. It uses speech recognition (STT), a language model (LLM), and text-to-speech (TTS) to have a natural conversation with the caller — in French, English, or Spanish. Unlike a traditional IVR that offers a rigid menu ("press 1 for..."), Voice AI understands natural language requests and responds intelligently and contextually.

Configuration

Each phone number can have its own Voice AI configuration:

Available Voices (Amazon Polly)

Multilingual Mode

When voiceAiMultiLang is enabled, Voice AI presents a language selection menu at the start of the call: "Bonjour! Hello! ¡Hola! Pour le français, dites français ou appuyez sur 1. For English, say English or press 2. Para español, diga español o presione 3." Each language can have:

• Its own custom greeting message
• Its own system prompt (specific AI instructions)
• Its own voice synthesis

Fallback

When Voice AI cannot respond (timeout, error, transfer request), the fallback action is triggered:

Two-Way SMS

Sending SMS

GoWeBa allows you to send SMS from the organization's numbers:

• Individual send: from a contact card or the Webabox
• Mass send: via marketing campaigns (Module 15)
• Automated SMS: triggered by workflows or WEBA
• Templates: reusable SMS templates with variables

Receiving SMS

Incoming SMS appear in the Webabox like regular messages:

• Instant browser notification
• Automatic contact association (if the number is known)
• Conversation history: SMS appear in the contact's thread
• Direct reply: respond to SMS from the Webabox

SMS Billing

The 30% telecom margin (1.30×) applies to all communication costs.

IVR — Interactive Voice Response

IVR Configuration

The IVR allows you to configure an automated voice menu:

• Custom greeting message for each phone number
• Menu options: "Press 1 for sales, 2 for support..."
• Routing to specific users or groups
• Schedules: different messages based on business hours

IVR vs Voice AI

Phone Administration

Number Management

Accessible via Admin → Phone. Allows you to:

1. View active numbers — List of all organization numbers with status
2. Configure IVR — Modify each number's greeting message
3. Enable/disable Voice AI — Configure the AI voice agent per number
4. Provision a new number — Order a local or toll-free number via Twilio
5. Release a number — Deactivate a number that is no longer in use

Call Statistics

The /api/phone/call-stats API provides consolidated statistics:

Voicemail

Voice messages are accessible via the Inbox → Voicemails section:

• Direct playback: play the audio message in the browser
• Automatic transcription: the voice message is transcribed to text
• Notification: alert when a new voicemail is received
• Contact association: the message is linked to the calling contact
• Read/unread status: manage read status

Integration with WEBA

WEBA Phone Agent

WEBA's phone agent (Agent #15 — Phone Agent) can:

• Make a call via chat: "Call Jean-Pierre at 514-555-1234"
• Summarize a recent call: "Summarize my last call"
• Extract actions: "What are the action items from my call with Acme?"
• Schedule a callback: "Call Acme back Thursday at 10 AM" → creates event + task

Voice AI → WEBA Flow

When Voice AI is active, the processing flow is:

1. Inbound call — Twilio routes to the GoWeBa number
2. Voice AI activated — voiceAiEnabled = true detected
3. Voice greeting — The greeting message is spoken (Polly TTS)
4. Listen (Gather) — Twilio STT transcribes the caller's speech
5. LLM Analysis — The transcribed text is sent to GPT-4.1 with the system prompt
6. Response — The LLM response is spoken via TTS
7. Loop — Return to step 4 until end of conversation or timeout
8. Post-processing — Full transcription, summary, and action items generated

Telecom Billing

Telecom billing follows the 1.30× rule (30% margin on provider cost):

Voice AI costs combine telecom fees (1.30×) and AI fees (1.40×) which add up.

Concrete Use Cases

Scenario: 24/7 Voice AI Customer Support

Situation: An SMB wants to offer phone support outside business hours.

1. Setup — Enable Voice AI on the toll-free number with a support-oriented prompt
2. Greeting — "Hello, thank you for calling [Company]. How can I help you?"
3. Customer calls — The customer asks: "I have a problem with my order #12345"
4. Voice AI responds — The LLM checks the knowledge base and answers with relevant info
5. Fallback — If the issue is complex, Voice AI offers to leave a voicemail
6. Follow-up — The next day, the team sees the transcription, summary, and action items

Scenario: Click-to-Call from CRM

1. User opens a contact card — The phone number is displayed
2. Click the phone icon — The dialer opens with the number pre-filled
3. WebRTC call — The call is made from the browser, no phone required
4. During the call — The user takes notes in GoWeBa in parallel
5. Call ends — AI summary + action items automatically generated
6. Tasks are created — Action items become tasks in the pipeline

Scenario: Multilingual Voice AI Menu

1. Inbound call — The customer calls the main number
2. Multilingual menu — "Bonjour! For French, press 1. For English, press 2. Para español, presione 3."
3. Selection — The customer says "English" or presses 2
4. Language switch — Voice AI switches to English with the Joanna voice
5. Conversation — The AI agent responds in English with the English system prompt
6. Call ends — English transcription, English summary

Scenario: Voice Appointment Booking

1. A customer calls — Voice AI answers with the greeting
2. Customer requests — "I'd like to book an appointment for a consultation"
3. Voice AI checks — Checks the calendar for available slots
4. Proposal — "I have openings Tuesday at 10 AM or Wednesday at 2 PM. Which works for you?"
5. Confirmation — The customer chooses, the event is created automatically
6. Confirmation SMS — An SMS is sent with the appointment details

Troubleshooting & FAQ

Common Issues

FAQ

Q: Can I use my own phone number? A: No, GoWeBa provisions numbers via Twilio. However, you can port your existing number to Twilio — contact support for the procedure. Q: Can Voice AI take orders? A: Currently, Voice AI can answer questions and book appointments. Voice ordering is planned for a future update. Q: How much does a 5-minute Voice AI call cost? A: Approximately $0.15 in telecom fees (5 min × $0.029/min) + approximately $0.10 in AI fees (~3–4 conversation turns). Estimated total: ~$0.25 per 5-minute call. Q: Are calls encrypted? A: Yes. WebRTC calls are end-to-end encrypted (SRTP). Recordings are stored encrypted at rest in the cloud. Q: Can Voice AI be set to only answer outside business hours? A: Yes, via the IVR schedule configuration. Voice AI can be activated only during specified hours (e.g., 5 PM–9 AM and weekends). Q: How many languages does Voice AI support simultaneously? A: Up to 3 languages with multilingual mode (French, English, Spanish). Each language has its own voice and instructions. Q: How can I improve Voice AI response quality? A: Three levers:

1. Optimize the system prompt with precise instructions
2. Feed the Knowledge Base with common FAQs
3. Test and adjust the formality level and communication style

Best Practices

1. Use neural voices — Gabrielle (FR) and Joanna (EN) offer the best quality
2. Keep prompts concise — A 200–300 word system prompt is ideal for Voice AI
3. Configure fallback — Always have a fallback option (voicemail or transfer)
4. Test regularly — Call your own number to verify the customer experience
5. Monitor costs — Use call statistics to optimize conversation duration
6. Record with consent — Always inform the caller if the call is being recorded

— End of Module 04 —

GoWeBa — Operations Bible Generated May 06, 2026