P2P first. Relay when needed.
Voice traffic takes the shortest path the network allows. When NAT or firewall policy blocks direct connections, a passive relay forwards ciphertext between endpoints — never holding keys, never inspecting payload.
Arquitectura de red
P2P directo o relay de servidor. En ambos casos, el servidor es ciego.
Modo A
P2P Directo
Llamante
Peer
Cuando el NAT lo permite. Mínima latencia, mínimo coste.
Modo B
Modo Relay
Llamante
Peer
Cuando las redes bloquean el P2P. El relay solo reenvía ciphertext.
El servidor nunca posee las claves
Los endpoints derivan el secreto de sesión
ML-KEM-1024 + AES-GCM
Misma criptografía en ambos modos
Selección automática
Intenta P2P primero, recae con elegancia
El relay es un reenviador pasivo de paquetes. No puede leer, modificar ni reproducir el flujo de ciphertext — ni siquiera bajo coacción.
NAT traversal mechanics
Most home and enterprise networks sit behind NAT or strict firewalls that block inbound UDP. Our client probes peer reachability with ICE-style candidate gathering, then opens a direct path when both endpoints can be reached.
- Candidate gathering on host, server-reflexive and relayed transports
- Symmetric NAT detection falls back to relay automatically
- No external STUN dependency on operator-controlled deployments
Fallback logic
The session always negotiates the same post-quantum key. Network mode is chosen by reachability, not by trust: if P2P fails, the relay takes over within the same RTT budget. The cryptographic envelope does not change.
- P2P attempt window: ≤ 800 ms before relay is engaged
- Mid-call switch supported when network conditions change
- Same ML-KEM-1024 session secret in both modes
No metadata retention
The relay forwards packets and discards them. No call detail records, no call duration logs, no participant graphs. The only state retained is short-lived NAT mapping required to route the next packet.
- No CDRs, no per-call billing logs, no participant graph stored
- Packet metadata kept only for the lifetime of the NAT mapping
- Operator-deployable: relay can run inside your own datacenter
Need a private relay deployment?
Air-gapped, sovereign, on-premise — your operators, your network, your control plane.
Request architecture brief