Consumer Messaging Platform — WhatsApp-Class System  ·  View 14 of 23

Voice and Video Call Architecture

WebRTC signalling over the existing chat socket, keys derived from the Signal session, and an SFU that forwards frames it cannot read.

Editable source SVG draw.io All views
Initiate
Initiate
Caller Device
Caller Device
Call Signalling Service
Over the chat socket
Call Signalling Service...
Callee Wake
VoIP push, high priority
Callee Wake...
Negotiate
Negotiate
SDP Offer and Answer
Sent as ciphertext
SDP Offer and Answer...
ICE Candidates
Trickled
ICE Candidates...
Key agreement
Key agreement
SRTP Master Key
From the Signal session
SRTP Master Key...
DTLS-SRTP Binding
DTLS-SRTP Binding
Path selection
Path selection
STUN Reflexive
STUN Reflexive
Direct Peer to Peer
When reachable
Direct Peer to Peer...
TURN Relay
Symmetric NAT fallback
TURN Relay...
Group media
Group media
Selective Forwarding Unit
Forwards encrypted frames
Selective Forwarding Unit...
Simulcast Layers
Bandwidth adaptive
Simulcast Layers...
Jitter & FEC
Loss concealment
Jitter & FEC...
Terminate
Terminate
Missed Call Push
Missed Call Push
Call Record
Metadata only
Call Record...
fallback on ICE failure
fallback on ICE failure
Voice and Video Call Architecture
Voice and Video Call Architecture
Application we own
Application we own
Interface / broker
Interface / broker
Security / platform
Security / platform
Data store
Data store
failure / alternate
failure / alternate
The SFU forwards encrypted frames and never holds the media key. No call content is recorded.
The SFU forwards encrypted frames and never holds the media key. No call content is recorded.
v 1.0 · owner Platform Architecture · date 2026-08
v 1.0 · owner Platform Architecture · date 2026-08Text is not SVG - cannot display

Signalling

  • SDP offers and answers travel as ordinary encrypted messages on the chat socket
  • No separate signalling protocol, so a call inherits the messaging security model
  • A high-priority VoIP push wakes a callee whose app is not running

Media path

  • Direct peer to peer whenever ICE succeeds, which is the common case
  • TURN relay for symmetric NAT; the relay sees encrypted packets only
  • Group calls use an SFU with simulcast so each participant gets a layer it can afford

What is retained

  • Call records hold participants, direction, start time and duration
  • No media is recorded, buffered to disk or transcoded in clear
  • Missed-call notifications carry no caller-supplied content
Solution Architecture v1.0 · Platform Architecture · 2026-0814-calls-architecture