DocsTelephony
How the Bivariant platform handles voice calls and SMS through SIP trunks, phone number management, and LiveKit-powered real-time audio.
The telephony channel connects your platform workspace to the public phone network. It supports inbound and outbound voice calls, SMS messaging, and integration with AI agents for automated voice interactions.
Architecture
Telephony on the platform is built on:
| Component | Role |
|---|---|
| SIP Trunks | Connect the platform to telecom carriers |
| Phone Numbers | Inbound numbers mapped to spaces and agents |
| LiveKit | Real-time audio/video infrastructure |
| Agents | AI-powered voice agents that handle calls |
Phone numbers
Phone numbers are the entry point for telephony. Each number is associated with a space and can be routed to an agent or flow.
Provisioning
Phone numbers can be:
- Purchased through a supported carrier integration
- Ported from an existing provider by configuring SIP trunk routing
Routing
When a call arrives at a phone number, the platform:
- Identifies the associated space and channel configuration
- Creates or resumes a conversation
- Routes the call to the configured agent or flow
- Streams audio in real time via LiveKit
SIP trunks
SIP trunks provide the connection between the platform and telecom carriers.
Configuration
To configure a SIP trunk:
- Navigate to Channels → Telephony → SIP Trunks
- Enter the trunk credentials:
- SIP server address
- Authentication credentials (username/password or IP allowlist)
- Codec preferences
- Map phone numbers to the trunk
Supported features
| Feature | Support |
|---|---|
| Inbound calls | ✓ |
| Outbound calls | ✓ |
| SMS (inbound) | ✓ |
| SMS (outbound) | ✓ |
| Call transfer | ✓ |
| DTMF tones | ✓ |
| Call recording | ✓ |
Voice agents
AI agents can handle voice calls through the telephony channel. When connected to a phone number, the agent:
- Receives the audio stream — real-time audio from the caller
- Transcribes speech — converts audio to text using the configured ASR provider
- Processes the message — runs the agent's response pipeline (context, knowledge base, LLM)
- Synthesizes speech — converts the response text to audio using TTS
- Streams the response — delivers audio back to the caller in real time
This creates a natural conversational experience over the phone, powered by the same agent configuration used in chat channels.
Call lifecycle
| Phase | What happens |
|---|---|
| Ringing | Incoming call arrives; platform identifies the channel |
| Connected | Call is answered; audio streams are established |
| Active | Conversation in progress; messages exchanged in real time |
| Hold / Transfer | Call can be placed on hold or transferred |
| Ended | Call terminates; conversation is closed or paused |
Recording and transcription
Calls can be recorded and transcribed automatically:
- Recordings are stored as assets in the space's file storage
- Transcriptions are attached to the conversation as messages
- Both are subject to the space's retention and access policies
Related concepts
- Channels — overview of all channel types
- Agents — configure AI agents for voice interactions
- Conversations — how calls become conversations
- Real-time Communication — WebSocket and streaming infrastructure
Presence Widget
How to embed the Bivariant Presence Widget on any website — a lightweight chat interface that connects visitors to your agents and workflows.
Identity and Access Management
The platform's IAM system provides authentication, role-based access control, API keys, and service accounts for secure multi-tenant operations.